Artificial Intelligence for Hard-to-Treat and Unknown-Origin Cancers

Artificial Intelligence for Hard-toTreat and Unknown-Origin Cancers Authors: Mehrdad Farrokhi ERIS Research Institute Niloofar Taheri Shahroud University of Medical Sciences Omid Fakharzadeh Moghadam Mashhad University of Medical Sciences Mahdi Armoon K. N. Toosi University of Technology Susan Samimi Shiraz University of Medical Sciences Narges Torkashvand Tehran University of Medical Sciences Saeed Motevalli Yasuj University of Medical Sciences Shabahang Moghaddasi Esfivashi Northumbria University Elham Sadat Mostafavi Ferdowsi University of Mashhad Sepideh Hassanpour Khodaei Eastern Mediterranean University ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 1 Fatemeh Amini Sichuan University Zahra Taheri Shiraz University Sepide Javankiani Tehran University of Medical Sciences Amir Nasrollahizadeh Tehran University of Medical Sciences Reyhaneh Mahbubi Arani Shahid Beheshti University of Medical Sciences Oveis Ahmadzadeh Kashan University of Medical Sciences Kimia Kowsari Azad University of Sari Kourosh Shahraki Zahedan University of Medical Sciences Mahyar Noorbakhsh Kashan University of Medical Sciences Mohammad Bdaqli University of Tabriz Roozbeh Roohinezhad Iran University of Medical Sciences Arezoo Heidary Qazvin Islamic Azad University Kimia Baghebani Jeonbuk National University DR. MEHRDAD FARROKHI 2 Farzaneh Kianifar Iran University of Medical Sciences Reza Mohajer Shirazi Islamic Azad University of Tehran Medical Sciences Mohammad Eslami Shahid Beheshti University of Medical Sciences Shahram Asgari University of Georgia Yousef Fekri Dabanloo University of Georgia Saman Abdollahpour Shahid Beheshti University of Medical Sciences Sanaz Amiri Marbini University Medical Center Hamburg-Eppendorf Seyed Ali Hashemi Kiapey Mashhad University of Medical Sciences Atena Talebpoor Amirhandeh Shahid Beheshti University of Medical Sciences Mohammad Yaghoubi Azerbaijan Medical University Neda Gorjizadeh Tehran University of Medical Sciences ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 3 DR. MEHRDAD FARROKHI 4 Book Details: Publisher: PreferPub and Kindle Publication Date: December 2025 Language: English Dimensions: 5 x 0.39 x 8 inches © PreferPub and Kindle 2025 ISBN-13: 979-8277533666 This peer-reviewed book is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specially the rights of translation, reprinting, result of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 5 DR. MEHRDAD FARROKHI 6 Contents Chapter 1. AI for Hard-to-Treat Hematologic Malignancies 2. AI for Hard-to-Treat Neurological Cancers 3. AI for Hard-to-Treat Gastrointestinal Cancers 4. AI for Hard-to-Treat Breast and Lung Cancers 5. AI for Hard-to-Treat Urogenital Cancers 6. AI for Hard-to-Treat Skin and Soft Tissue Cancers 7. AI for Hard-to-Treat Head and Neck Cancers 8. AI for Hard-to-Treat Oral Cancers 9. AI for Hard-to-Treat Bone and Musculoskeletal Cancers 10. AI for the Diagnosis and Treatment of Unknown-Origin Cancers ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 7 DR. MEHRDAD FARROKHI 8 1. AI FOR HARD-TO-TREAT HEMATOLOGIC MALIGNANCIES Background Leukemia, commonly known as blood cancer, is a type of hematologic malignancy in which the bone marrow begins producing abnormal and dysfunctional blood cells, particularly white blood cells. These immature cells that lack normal immune function accumulate within the blood and bone marrow, replacing healthy cells and interfering with normal hematopoiesis. In general terms, leukemia refers to a group of cancers that originate from blood forming tissues including the bone marrow and the lymphatic system. Leukemia may develop suddenly with rapid progression as seen in acute forms, or it may emerge gradually while following a slower clinical course as seen in chronic forms. Based on the origin of the malignant cell, whether myeloid or lymphoid, this disease is classified into four principal categories which are Acute Lymphoblastic Leukemia, Acute Myeloid Leukemia, Chronic Lymphocytic Leukemia, and Chronic Myeloid Leukemia. Acute Lymphoblastic Leukemia occurs more frequently in children and is marked by the rapid proliferation of immature lymphocytes that 9 are unable to combat infections effectively. Even though this disease progresses aggressively, early diagnosis accompanied by appropriate treatment often results in a high rate of remission and long term recovery potential in many cases. Acute Myeloid Leukemia, in contrast, generally appears in adults and arises from abnormal growth of myeloid cells that are responsible for producing red blood cells, white blood cells, and platelets. The clinical outcome of Acute Myeloid Leukemia is typically poorer than that of Acute Lymphoblastic Leukemia and treatment often involves intensive chemotherapy and occasionally bone marrow transplantation as a standard therapeutic approach. Chronic Lymphocytic Leukemia occurs primarily in older adults and usually follows a slow and sometimes asymptomatic course for extended periods. It is often detected incidentally through routine blood examinations. Malignant lymphocytes in Chronic Lymphocytic Leukemia gradually accumulate in the blood and lymphatic tissues which, if left untreated, can lead to significant immune system dysfunction and susceptibility to infections. Chronic Myeloid Leukemia is associated with the presence of a specific genetic alteration known as the Philadelphia chromosome which results in abnormal activation of proliferative signals and excessive production of myeloid cells. Although treatment of this disease was historically difficult, DR. MEHRDAD FARROKHI 10 the introduction of targeted therapies such as tyrosine kinase inhibitors has greatly improved survival outcomes and quality of life for affected patients. In addition to the four major categories, rarer forms of leukemia also exist including Hairy Cell Leukemia, Promyelocytic Leukemia, and mixed phenotype leukemias. Promyelocytic Leukemia, a subtype of Acute Myeloid Leukemia, is defined by an excessive accumulation of promyelocytes in the bone marrow and severe coagulation disorders. Due to the possibility of life threatening bleeding, this subtype requires rapid diagnosis and urgent intervention to prevent mortality. From a cellular and histopathological viewpoint, leukemias characteristically present as blasts which are immature undifferentiated cells with high proliferative potential. Their morphological characteristics are visible in peripheral blood smears or bone marrow biopsy samples. Cytochemical and immunophenotypic studies utilize specific molecular markers such as CD19 and CD10 in Acute Lymphoblastic Leukemia or CD33 and myeloperoxidase in Acute Myeloid Leukemia to support accurate diagnosis and classification. Precise molecular and genetic profiling is essential for determining prognosis and selecting effective treatment regimens. For instance, certain gene mutations in Acute Myeloid Leukemia can strongly influence therapeutic responses and overall clinical outcomes. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 11 The prevalence of leukemia varies by age group. International health statistics indicate that leukemia is the most common cancer diagnosed in children under 15 years of age while chronic leukemias are more commonly observed in elderly populations. The incidence rate also tends to be slightly higher in males than in females. Reports from some regions suggest that leukemia rates are rising possibly due to environmental and industrial exposures, chemical contact, and improvements in detection systems. Several classification systems have been developed to categorize leukemias based on morphological, immunophenotypic, and cytogenetic characteristics. The French American British classification was one of the earliest systems and defined subtypes of Acute Myeloid Leukemia from M0 to M7 based on cellular morphology. Modern classifications incorporate genetic abnormalities, specific chromosomal mutations, and the degree of cellular differentiation, allowing more precise disease categorization. With advancements in genomic sequencing, medical imaging, and machine learning, more accurate leukemia classification is becoming increasingly vital not only for diagnostic accuracy but also for prognosis prediction, therapeutic planning, and implementation of personalized medicine. Artificial intelligence based methods such as bioinformatics modelling, RNA DR. MEHRDAD FARROKHI 12 sequencing analysis, and integration of laboratory data have recently enabled the development of sophisticated computational tools capable of detecting leukemia subgroups with high accuracy. These emerging technologies will be explained in later sections of this study. Etiological and Predisposing Factors of Leukemia Although leukemia originates from genetic and cellular abnormalities, its onset often results from a complex interaction between internal factors including genetic and epigenetic alterations and external influences such as environmental exposure, occupational hazards, lifestyle, and infectious agents. Understanding these contributing factors is essential for exploring disease mechanisms and designing predictive systems using artificial intelligence based methods. Among the most significant genetic contributors are chromosomal abnormalities which are frequently observed in leukemia subtypes. The Philadelphia chromosome which is caused by a translocation between chromosomes 9 and 22 is a hallmark of Chronic Myeloid Leukemia and is responsible for continual activation of proliferative signalling pathways that stimulate uncontrolled cell growth. Mutations in several genes including NPM1, FLT3, TP53, DNMT3A, and IDH1 or IDH2 ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 13 appear frequently in patients with Acute Myeloid Leukemia and have considerable effects on prognosis and treatment responses. Some of these genetic alterations result in incomplete differentiation, enhanced resistance to cell death, and promotion of malignant proliferation. In children, inherited genetic syndromes such as Down syndrome, Bloom syndrome, Fanconi anemia, and neurofibromatosis markedly increase susceptibility to leukemia because of impaired DNA repair mechanisms. Environmental and occupational exposures also play a major role in leukemia onset. Ionizing radiation, especially when exposure occurs during prenatal development or early childhood, is a well recognized risk factor. Workers frequently exposed to X rays or radioactive substances also face increased likelihood of disease development. Chemical substances including benzene, pesticides, and industrial solvents have been widely acknowledged as factors that can induce leukemia, particularly cases with myeloid origin. Lifestyle habits influence leukemia risk as well. Cigarette smoking contains carcinogenic compounds capable of damaging hematopoietic stem cells and contributing to malignant transformation. Poor nutrition, reduced antioxidant intake, and consumption of foods containing synthetic preservatives increase oxidative stress which can disrupt DNA stability in DR. MEHRDAD FARROKHI 14 bone marrow cells. Viral factors have gained increasing attention as certain viruses are linked with hematologic malignancies. Human T lymphotropic virus type 1 is definitively associated with Adult T cell Leukemia and may remain dormant for years prior to disease onset. Epstein Barr virus and several herpesvirus strains are believed to contribute to immune dysregulation and may predispose individuals to leukemia development under specific conditions. Immunosuppression contributes significantly to leukemia susceptibility, particularly among organ transplant recipients or individuals with chronic immune dysfunction such as HIV infection. Exposure to chemotherapy or radiotherapy during prior cancer treatment may also lead to secondary leukemia which is a therapy related complication. Recent research has turned its attention toward the gut microbiome and its role in hematologic malignancies. Alterations in beneficial microbial communities can impair immune regulation and promote chronic inflammation which in turn increases risk for malignant transformation in blood forming tissues. Psychosocial influences have also been studied. Although a direct causal relationship is not fully established, chronic psychological stress may disrupt normal immune responses and therefore contribute indirectly to cancer development including leukemia. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 15 The interaction between genetic vulnerability and environmental triggers is fundamental in leukemia pathogenesis. An individual carrying a predisposing mutation may only experience disease manifestation when exposed to a specific environmental factor. This concept forms the basis of many artificial intelligence based predictive models which estimate leukemia risk by integrating genetic, biological, demographic, and environmental data. Machine learning models that combine multi dimensional datasets have shown superior accuracy compared with analyses relying on isolated risk factors alone. Recognizing and addressing these predisposing factors not only supports early diagnosis but also lays the foundation for targeted interventions, optimized preventive strategies, and development of personalized treatment planning. These elements will be further expanded upon in subsequent chapters focusing on artificial intelligence applications in leukemia diagnosis and management. Clinical Manifestations and Diagnostic Methods of Leukemia The clinical manifestations of leukemia are often gradual and nonspecific, and in the early stages may be mistaken for viral infections or simple anemias. This characteristic can delay diagnosis, particularly in chronic leukemias. Symptom presentation depends on the type of DR. MEHRDAD FARROKHI 16 leukemia, whether acute or chronic and myeloid or lymphoid, the rate of disease progression, the extent of bone marrow involvement, and the spread to other organs. Common general symptoms include severe fatigue, weakness, pallor, fever, night sweats, unexplained weight loss, bone and joint pain, abnormal bleeding, spontaneous bruising, recurrent infections, and lymphadenopathy. In acute leukemias, symptoms appear suddenly within a few weeks, and the patient’s condition can deteriorate rapidly. For instance, in Acute Lymphoblastic Leukemia, patients may present with fever, severe infections, and reduced levels of consciousness. In children with Acute Lymphoblastic Leukemia, the most common signs include anemia, gum bleeding, leg pain, and splenomegaly. In Acute Myeloid Leukemia, symptoms result from impaired bone marrow function and the accumulation of blasts. Patients may experience neutropenia, thrombocytopenia, and reduced red blood cell counts, which manifest as fever, bleeding, recurrent infections, and anemia. In contrast, chronic leukemias are often discovered incidentally during routine blood tests. In Chronic Lymphocytic Leukemia, patients may remain asymptomatic for a long period, with the only abnormal finding being an elevated white blood cell count. As the disease progresses, splenomegaly, lymphadenopathy, night sweats, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 17 artificial intelligence. In this context, artificial intelligence can act as an intelligent assistant to physicians in analyzing clinical data, predicting disease progression, detecting hidden patterns in images or laboratory results, and even designing personalized treatment plans. In the scientific literature, artificial intelligence is generally divided into two broad categories, narrow artificial intelligence and general artificial intelligence. Narrow artificial intelligence refers to algorithms trained for a specific task such as image recognition, laboratory result prediction, or genetic data classification. General artificial intelligence refers to systems capable of reasoning comprehensively like humans, which remain at the theoretical and research stage. In medicine, current applications mostly fall into the narrow category and include machine learning, computer vision, natural language processing, and predictive modeling. One of the key concepts in artificial intelligence is machine learning. Instead of following predefined rules, machine learning algorithms learn patterns from experience and data and thereby acquire predictive capability. Machine learning is generally divided into three categories, supervised learning, unsupervised learning, and reinforcement learning. In supervised learning, algorithms are trained on labeled datasets such as identifying cancerous cells in microscopic images or predicting laboratory outcomes from known DR. MEHRDAD FARROKHI 30 inputs. In unsupervised learning, the goal is to discover hidden structures or clusters in data such as identifying unknown patient subgroups based on gene expression profiles. In reinforcement learning, an intelligent agent interacts with its environment through trial and error, improving its decision making based on rewards and penalties. Another fundamental concept is the artificial neural network, which is inspired by the structure of the human brain. These networks are composed of interconnected artificial neurons organized in layers. In medicine, artificial neural networks play an important role in disease diagnosis, medical image processing, genomic analysis, and modeling of complex clinical pathways. A more advanced form, deep learning, incorporates multiple hidden layers to extract increasingly complex and abstract features from raw data and has become a leading approach in many biomedical applications. A widely used branch of deep learning in medicine is the convolutional neural network, particularly applied in analyzing medical images such as magnetic resonance imaging, computed tomography scans, X rays, and blood smear images. For example, in the automated diagnosis of leukemia cells from microscopic images, convolutional neural networks have in some cases achieved accuracy levels comparable to or higher than human specialists. These models can detect complex visual features such as cell shape, nuclear ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 31 texture, and staining patterns and use them for precise disease classification. Beyond image analysis, textual data analysis is also vital in modern healthcare. Natural language processing, a branch of artificial intelligence, is designed to analyse and interpret human language. In medicine, natural language processing can automatically extract information from clinical records, physician notes, pathology reports, and scientific literature. This facilitates the identification of risk factors, medication histories, comorbidities, and laboratory findings from large amounts of unstructured data and enhances clinical research and decision making. Modern language models have demonstrated remarkable performance in natural language understanding and are expected to be increasingly integrated into health information systems. Artificial intelligence is not limited to diagnosis and classification; it also plays a major role in prediction. Risk prediction models can integrate clinical, laboratory, genetic, and environmental data to estimate the likelihood of disease occurrence, progression, or relapse with high accuracy. For instance, in leukemia patients, artificial intelligence models can predict treatment response, risk of adverse effects, or survival outcomes. Such models are designed using algorithms like random forests, support vector machines, gradient boosting methods, and deep learning architectures. DR. MEHRDAD FARROKHI 32 Clinical decision support systems represent another practical product of artificial intelligence in medicine. By analyzing patient data, these systems provide physicians with recommendations that may include likely diagnoses, drug selection, optimal dosages, alerts about potential adverse events, or suggested follow up plans. Some of these systems are already deployed in hospitals and integrated with electronic health record platforms, assisting clinicians in daily practice. However, the application of artificial intelligence in medicine raises ethical, legal, and technical concerns. Issues such as algorithmic transparency, explainability, data privacy, security, and liability for machine based decisions must be carefully considered. One major criticism of deep learning models is their black box nature, where the rationale behind final decisions is unclear and difficult to interpret. In medicine, this lack of interpretability can undermine the trust of both physicians and patients and may limit the acceptance of artificial intelligence tools in critical situations. In summary, the foundational concepts of artificial intelligence in medical sciences are rapidly expanding. With the development of digital infrastructures, improvements in data quality, and closer interdisciplinary collaboration, artificial intelligence is expected to become an integral component of future diagnostic, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 33 therapeutic, and disease management processes, particularly in complex conditions such as leukemia where large scale data integration and precise decision making are crucial. Common Machine Learning Algorithms in Disease Diagnosis Machine learning as a key subset of artificial intelligence refers to algorithms that learn from data to make decisions or predictions without requiring explicit programming of all rules. In medicine, machine learning algorithms are widely applied for analyzing large datasets, identifying hidden patterns, classifying patients, predicting disease progression, and optimizing treatment strategies. Their use in diagnosing conditions such as leukemia, solid cancers, diabetes, cardiovascular diseases, and infectious diseases is rapidly expanding across clinical and research environments. The choice of algorithm depends on the nature and structure of the data, the specific objective such as classification, regression, or clustering, and the complexity of the underlying patterns that must be captured. One of the most common supervised learning algorithms is the decision tree. This algorithm implements a set of conditional rules in the form of a tree structure and is highly applicable for problems such as disease diagnosis based on clinical symptoms or laboratory results. Decision trees are popular in medical modeling because of DR. MEHRDAD FARROKHI 34 their conceptual simplicity, high interpretability, relatively low computational requirements, and fast execution, which are valuable in clinical workflows that demand transparency. An advanced version of decision trees is the random forest algorithm, which builds an ensemble of decision trees and combines their outputs to improve accuracy and stability. Random forest has been successfully applied in diagnosing blood cancers, characterizing tumour histology, and predicting treatment responses in a variety of diseases. For example, in research focused on leukemia diagnosis using blood cell features, random forest has often outperformed several other algorithms in terms of classification accuracy and robustness, demonstrating its usefulness for hematologic applications. Support vector machine is another widely used algorithm in medicine. It classifies data by identifying the optimal decision boundary in the feature space that maximizes separation between classes. Support vector machine is particularly effective for high dimensional and complex datasets, such as genomic profiles or medical imaging features, where conventional methods may struggle. In studies devoted to leukemia, support vector machine has shown remarkable performance in classifying acute leukemia subtypes using genetic and molecular data, demonstrating its power in precision diagnostics. The k nearest neighbors algorithm is a simple yet ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 35 practical method that classifies new samples based on their similarity to training data. Although it may be less accurate than more sophisticated models, it is useful in projects with limited datasets or as a baseline in medical research. A key advantage of k nearest neighbors is that it makes no strong assumptions about data distribution, which can be beneficial when the underlying statistical properties of clinical data are not fully known. Among regression algorithms, logistic regression is commonly used in clinical studies to predict disease occurrence or relapse. This model examines the relationship between input variables such as age, laboratory results, and genetic mutations and binary outcomes such as presence or absence of disease or survival status. Due to its interpretability, well understood statistical foundations, and extensive history in medicine, logistic regression remains in use alongside more advanced machine learning models and often serves as a benchmark for evaluating newer methods. Artificial neural networks are also crucial machine learning algorithms with broad applications in medicine. Inspired by the brain’s neural structure, they can model complex nonlinear relationships among variables. With the growth of medical data and computational resources, advanced neural networks known as deep learning architectures have been developed. Among them, DR. MEHRDAD FARROKHI 36 convolutional neural networks are highly effective for medical image processing and visual pattern recognition. In leukemia diagnosis, convolutional neural networks have achieved high accuracy in classifying blood cells from microscopic images and have begun to replace manual visual assessment in some automated pipelines, thereby reducing observer variability. Gradient boosted algorithms such as XGBoost and LightGBM have shown excellent performance in medical projects involving structured data such as clinical tables and laboratory indices. These models sequentially build weak decision trees and optimize the errors of previous models to gradually form stronger predictors. Gradient boosting models have consistently ranked highly in disease detection challenges and clinical data competitions, offering strong accuracy when combined with appropriate feature selection, parameter tuning, and careful validation. In unsupervised problems such as discovering new patterns in genetic data or clustering patients into subgroups, algorithms like k means clustering and principal component analysis are commonly applied. These methods reveal hidden structures without using labeled data and help in identifying novel disease subgroups, risk categories, or response clusters, which can be especially important in heterogeneous conditions such as leukemia and other hematologic malignancies. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 37 When selecting an algorithm, several factors must be considered, including the type of data such as numeric, imaging, or text, the dataset size, class balance, interpretability needs, training time, and required accuracy. For example, in clinical environments where critical decisions are made, interpretable algorithms like decision trees or logistic regression may be preferred to ensure that clinicians can understand the reasoning behind predictions. In contrast, for research purposes, complex models like convolutional neural networks or XGBoost, although more black box in nature, may be appropriate because of their higher potential accuracy and ability to capture subtler patterns. Moreover, combining algorithms in ensemble approaches can further improve performance. By leveraging the strengths of multiple models, ensemble methods reduce overall error and enhance prediction stability. Combinations such as random forest with support vector machine or ensembles that incorporate neural networks and gradient boosting have been successfully applied in various medical diagnostic projects, including hematologic disease classification. Finally, deploying machine learning algorithms in real world clinical settings requires addressing technical and ethical issues such as data quality, elimination of bias, patient privacy, transparency, and model explainability. In leukemia projects and other hematologic domains, combining these DR. MEHRDAD FARROKHI 38 algorithms with physician expertise can lead to more accurate diagnoses, optimized treatment strategies, and improved prognostic assessments, ultimately supporting better patient outcomes. Applications of Artificial Intelligence in Diagnosing Hematologic Diseases In recent years, artificial intelligence has become a key tool in diagnosing hematologic diseases, particularly because of the massive volumes of laboratory, imaging, and genomic data whose analysis often exceeds human cognitive capacity. Hematologic conditions, including various leukemias, lymphomas, anemias, coagulopathies, and myelodysplastic syndromes, have traditionally been evaluated using tests such as complete blood count, peripheral blood smear, bone marrow biopsy, and molecular assays. However, accurate and simultaneous interpretation of these diverse data sources requires advanced analytical tools. Artificial intelligence algorithms, especially machine learning and deep learning methods, enable rapid and high accuracy detection of hidden patterns in clinical and laboratory data and thereby facilitate early diagnosis and better risk stratification. One of the earliest applications of artificial intelligence in hematology is automated processing of microscopic peripheral blood smear images. Convolutional neural networks have ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 39 been used to automatically distinguish normal from abnormal cells and have achieved very high accuracy in classifying blasts, lymphocytes, monocytes, and neutrophils. In several studies, convolutional neural network based models have distinguished leukemic from normal cells with accuracy rates exceeding ninety percent. In these models, features such as nuclear shape, nucleus to cytoplasm ratio, texture, and staining characteristics are extracted from images, enabling high throughput analysis without the need for continuous human intervention. Beyond microscopy, flow cytometry data also provide a rich substrate for machine learning applications. Algorithms such as support vector machines and random forests have been applied to phenotype classification of blood cells, identification of leukemia subgroups, and differentiation of blasts from reactive lymphocytes. Using such models to analyze flow cytometry data can produce results comparable to expert diagnoses while operating in significantly shorter time frames, which is particularly valuable in emergencies or in hospitals with limited staff. Genomic and transcriptomic analysis is another rapidly expanding domain for artificial intelligence in hematology. Models such as gradient boosting algorithms and deep neural networks trained on RNA sequencing data can identify specific mutations, generate risk scores, and predict treatment response. By combining DR. MEHRDAD FARROKHI 40 clinical characteristics, cytogenetic information, and genomic data with deep learning approaches, researchers can produce personalized risk profiles for patients with acute leukemias, thereby aiding therapy selection and predicting relapse with greater precision. Another notable application is the development of alerting and decision support systems that automatically notify clinicians when dangerous patterns are detected in laboratory results or vital signs. Often deployed in intensive care units or hematology wards, these systems continuously analyze data streams to prevent clinical crises. For example, deep learning networks have been used to analyze many laboratory parameters and vital indices in order to predict organ failure, severe infection, or the need to change therapy in hematologic patients, providing early warnings that can guide timely intervention. In lymphoma diagnosis, deep learning has been employed to interpret digital histopathology images. Models such as ResNet and DenseNet analyze cellular architecture and tissue morphology on whole slide images. These convolutional neural network based systems have achieved high accuracy for classifying lymphoma subtypes, sometimes matching or surpassing the performance of human pathologists and providing decision support for complex cases. Artificial intelligence usage is also expanding in anemias, thalassemia, hemophilia, and platelet ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 41 disorders. In thalassemia screening, decision trees and random forests using hematologic indices such as mean corpuscular volume, mean corpuscular hemoglobin, and red cell distribution width can provide noninvasive identification and reduce the need for expensive molecular testing. In hemophilia, machine learning methods are being explored to predict bleeding risk and individual response to factor replacement therapy, allowing more personalized dosing strategies. A recent advance is the development of multimodal artificial intelligence models that jointly process heterogeneous inputs such as images, numeric data, and free text. By analyzing blood tests, clinical reports, and bone marrow images simultaneously, these models can make more precise decisions. For instance, systems that combine natural language processing for physician notes with convolutional neural networks for smear images have achieved high diagnostic accuracy for leukemia and related disorders, demonstrating the benefit of integrated approaches. Overall, artificial intelligence in hematologic diseases now extends beyond diagnosis to prognosis estimation, therapy selection, disease subgrouping, and post treatment monitoring. By reducing human error, accelerating analysis, and uncovering hidden patterns, artificial intelligence opens new frontiers in hematology. Nevertheless, ongoing challenges including data quality, DR. MEHRDAD FARROKHI 42 model interpretability, fairness, and ethical considerations require standardized frameworks, large scale clinical validation, and careful oversight to ensure safe and effective clinical deployment. Neural Network Based Models for Leukemia Prediction Artificial neural networks, particularly deep learning architectures, have become powerful tools for medical data analysis. In hematology, and specifically in leukemia prediction and diagnosis, these models excel because of their ability to capture complex patterns, nonlinear relationships, and large scale multimodal inputs. Unlike classical statistical methods that require strict distributional assumptions, neural networks learn patterns directly from data and often achieve superior predictive performance when adequately trained and validated. A key application is the classification of microscopic blood cell images. Convolutional neural networks have shown excellent performance in this domain by automatically extracting features such as edges, textures, shapes, and color distributions. In research on Acute Lymphoblastic Leukemia, convolutional neural network based systems have reported accuracy levels above ninety five percent in classifying leukemic cells without manual feature engineering. Early layers in the network learn ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 43 low level visual features, whereas deeper layers capture more abstract representations such as nuclear structure patterns and chromatin organization. Neural networks are also applied to numeric laboratory data including white blood cell counts, platelet counts, blast percentages, mean corpuscular volume, red cell distribution width, and biochemical markers. These features can be fed into multilayer perceptron architectures to predict leukemia presence, subtype, or risk categories. In multiple studies, multilayer perceptrons trained on a limited set of laboratory variables have achieved high accuracy in distinguishing leukemia from other hematologic conditions, illustrating the usefulness of neural networks even with relatively small feature sets. For longitudinal prediction tasks such as disease progression or relapse risk, recurrent neural networks, especially long short term memory networks, are used to model temporal dependencies in clinical trajectories, treatment histories, sequential responses, and biomarker trends. By analyzing time series data from leukemia cohorts, these models can achieve credible accuracy in predicting time to relapse or progression, thereby informing follow up schedules and treatment adjustments. A major advantage of neural networks is their capacity for data fusion. Multi input or multimodal architectures can accept clinical DR. MEHRDAD FARROKHI 44 variables, laboratory measurements, images, and genomic data simultaneously, with intermediate layers integrating these signals to produce more accurate decisions. For example, architectures in which smear images are processed by a convolutional neural network while laboratory or genomic data are fed into a multilayer perceptron have achieved improved accuracy for early leukemia diagnosis compared with single modality models. Neural networks also perform strongly in survival prediction and treatment response modeling. By integrating demographic factors, genetic markers such as FLT3, NPM1, and BCR ABL, treatment regimens, and early response indicators, models can estimate therapy success probabilities or overall survival. Analyses that combine registry data with gene expression profiles have shown that neural network models can predict outcomes in chronic leukemias and acute leukemias with high accuracy, supporting more informed clinical decision making. However, challenges remain in deploying neural network models. High performance typically requires large, diverse, and well curated datasets; explainability is limited because of black box behavior; and training demands substantial computational resources and careful hyperparameter tuning. Overfitting is a major risk with small datasets, and mitigation techniques such as dropout, regularization, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 45 data augmentation, and cross validation are essential. For interpretability, methods like feature attribution and visualization of salient image regions help explain which patterns contribute most to a given prediction. In summary, neural network models, especially convolutional neural networks, multilayer perceptrons, and long short term memory networks, offer advanced, accurate, and flexible tools for leukemia prediction and diagnosis. They are increasingly moving from research laboratories into clinical decision support systems. With broader access to structured and unstructured data, interoperable health information technology, and scalable computing resources, these technologies are poised to become integral components of leukemia care and personalized hematologic oncology. Deep Learning for Genetic and Imaging Data Analysis Deep learning, which refers to multilayer neural methods with high representational capacity, has transformed the analysis of complex medical data over the last decade. In hematology, and particularly in leukemia, applying deep learning to genetic and imaging datasets has enabled more precise identification of disease, faster diagnosis, and improved prediction of disease course. Two primary implementation arenas are analysis of gene expression, sequencing, and mutation data DR. MEHRDAD FARROKHI 46 and analysis of cellular microscopy and digital histopathology images. In genetics, analyzing RNA sequencing, DNA methylation, single nucleotide polymorphisms, and structural variants entails processing very large sets of context dependent measurements. Deep architectures, including deep neural networks, autoencoders, and sequence oriented models such as long short term memory networks, can learn nonlinear gene gene relationships and expression patterns that help classify patients, identify leukemia subgroups, and predict treatment response. For example, training deep networks on RNA sequencing data from Acute Lymphoblastic Leukemia patients has enabled the identification of high risk subgroups with accuracy that surpasses classical statistical models, demonstrating the added value of deep learning for risk stratification. On the imaging side, convolutional neural networks are widely deployed. For leukemia, peripheral blood smear images and digital bone marrow slides are primary visual sources. Convolutional neural networks automatically extract nuclear morphology, texture, cell size, and staining cues to separate normal from blast or abnormal cells. Several studies have reported accuracy close to or above ninety six percent for leukemia cell classification, exceeding many traditional image processing methods and improving consistency among assessments. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 47 A key advantage over traditional techniques is automated feature learning, which eliminates the need for manual feature engineering. This is particularly crucial in genomics, where thousands of variables interact in complex ways. Autoencoders and related architectures can effectively reduce dimensionality, generating compact gene embeddings that subsequently power survival, relapse, or drug response prediction models. This approach enables better modeling of relationships among genes and pathways involved in leukemogenesis. Combining imaging and genetic data is an emerging and powerful approach. In such multimodal pipelines, smear images are analyzed with convolutional neural networks while genomic or laboratory data are modeled with deep neural networks or multilayer perceptrons; their outputs are then fused for final decisions. Studies that apply this strategy for classification and survival prediction in acute leukemia have reported significantly higher performance compared with single modality models, highlighting the value of integrating complementary information sources. For advanced imaging, such as digital bone marrow and lymph node slides, deep architectures including ResNet, Inception, and EfficientNet have shown outstanding performance in detecting abnormal or tumoral tissue. Automated systems based on these architectures can identify and DR. MEHRDAD FARROKHI 48 quantify blasts in bone marrow images of acute leukemia, supporting more objective assessment and reducing interobserver variability. Deep learning has also been leveraged to predict drug resistance. By integrating genotype, gene expression levels, treatment type, and clinical response, models estimate resistance to agents such as tyrosine kinase inhibitors or targeted small molecules in chronic and acute leukemias. Multiple studies report high accuracy in predicting which patients are likely to develop resistance, facilitating early modification of treatment strategies. Challenges include the need for large, well labeled datasets; risks of overfitting on limited data; and the clinical imperative for interpretability and robustness. Mitigation approaches include data augmentation, dropout, early stopping, and regularization, alongside explainability tools such as gradient based visualization to highlight salient image regions and feature attribution tools for genomic variables. These tools help clinicians better understand model behavior and increase trust in deep learning predictions. Overall, deep learning for genetic and imaging analysis in leukemia has opened new horizons in diagnosis, prognosis, and personalized therapy. Integrating these models with health information systems, electronic health records, and clinical algorithms can yield decision support platforms that are likely to become integral components of ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 49 in designing effective repositories that support reproducible and generalizable research. Performance Metrics for Evaluating AI Models Evaluating the performance of artificial intelligence models in medicine, especially in critical applications such as leukemia diagnosis, requires precise, reliable, and context appropriate metrics. These metrics are not only used to compare algorithms but are also essential for ensuring clinical safety, regulatory compliance, and trustworthiness. Models that perform well in controlled laboratory settings may not succeed in real world clinical environments unless they are assessed against rigorous and meaningful indicators. One of the most common metrics is accuracy, defined as the ratio of correct predictions to the total number of predictions. While accuracy is useful when datasets are balanced, meaning similar numbers of positive and negative cases, medical data are often imbalanced. For leukemia diagnosis, only a small fraction of samples may be positive. In such cases, a model that predicts no disease for all samples might achieve high accuracy yet be clinically useless because it fails to detect any true patients. Therefore, sensitivity and specificity become more critical in medical evaluation. Sensitivity, also called recall or true positive rate, refers to the DR. MEHRDAD FARROKHI 62 ability to correctly identify positive cases, which is vital in medicine because missing a true patient can have serious or even fatal consequences. Specificity, the true negative rate, measures the ability to correctly identify negative cases such as healthy individuals. A clinically useful model in leukemia must maintain high sensitivity to avoid missed diagnoses while preserving sufficient specificity to limit unnecessary anxiety and interventions in healthy or low risk individuals. Another important metric is precision, which indicates how many of the predicted positive cases are truly positive. Precision, combined with recall, is crucial in conditions where false positives carry high costs, such as initiating expensive, toxic, or invasive treatments. The F1 score, which is the harmonic mean of precision and recall, is often used to balance these metrics and to provide a single summary value that reflects performance on positive cases. The receiver operating characteristic curve and its area under the curve are widely applied for classification performance evaluation. The receiver operating characteristic curve illustrates the trade off between false positive rates and true positive rates across different decision thresholds. The area under the curve ranges from zero to one, with values closer to one indicating better overall performance across thresholds. Values above 0.90 generally represent a strong model in many medical contexts and provide an intuitive ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 63 summary of discriminative ability. For multi class problems, such as distinguishing between Acute Lymphoblastic Leukemia, Acute Myeloid Leukemia, Chronic Lymphocytic Leukemia, and Chronic Myeloid Leukemia, a confusion matrix becomes crucial. It provides class specific counts of correct and incorrect predictions and highlights strengths and weaknesses in identifying particular subtypes. This information can guide model refinement and inform clinicians about which categories may require additional human review. When models are designed for regression tasks such as survival prediction, gene expression level estimation, or treatment duration, metrics such as mean squared error, root mean squared error, and the coefficient of determination are used. The coefficient of determination measures the proportion of variance in the dependent variable that is explained by the model; values closer to one indicate stronger explanatory and predictive power. In clinical settings, prediction time and computational efficiency also matter. Even a highly accurate model is impractical if its predictions take many minutes or hours to compute, especially in emergencies or high throughput laboratories. Thus, algorithms that balance speed and accuracy, such as optimized gradient boosting or compact neural networks, are often preferred unless the absolute highest DR. MEHRDAD FARROKHI 64 accuracy is essential and time constraints are less critical. Interpretability is another informal yet critical metric. Models such as decision trees and logistic regression are easily interpretable by physicians, who can review their rules and coefficients, whereas deep networks may deliver higher accuracy but lack transparency. In medicine, decision explainability is vital for gaining clinician and patient trust, and in some jurisdictions regulatory systems restrict or scrutinize the use of black box models for direct clinical decision making. Given the multidimensional nature of model performance, relying on a single metric is insufficient. Most successful studies report multiple measures, including accuracy, precision, recall, F1 score, area under the curve, and processing time, so that decision makers can select the most balanced option for their operational needs and risk tolerance. Domestic Studies on Leukemia Modeling In recent years, with increased access to medical data and computational capacity, researchers in Iran have also applied artificial intelligence methods for leukemia diagnosis, prediction, and classification. Although challenges such as limited datasets, labeling accuracy, and hardware and software infrastructure remain, domestic studies ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 65 have taken significant steps toward developing intelligent medical systems and present a promising outlook for national advancement in this field. Some early work used support vector machine algorithms to classify blood test data from leukemia patients. Features such as white blood cell count, platelet level, blast percentage, and other hematologic indices served as inputs and achieved high accuracy levels in distinguishing affected individuals from controls. These studies emphasized the role of machine learning algorithms in early stage leukemia screening and highlighted their potential as decision support tools for clinicians. Other research has applied decision trees and random forests to laboratory and clinical data from Acute Myeloid Leukemia patients. In many of these studies, random forest outperformed individual decision trees, achieving high accuracy in detecting positive cases and demonstrating the importance of ensemble methods. Feature selection techniques, including dimensionality reduction with principal component analysis, were also used to identify the most informative variables and to improve model performance. Artificial neural networks have been explored for identifying abnormal cells in blood smear images. After preprocessing and extracting geometric, textural, and color features, neural networks classified leukemic cells with high accuracy and DR. MEHRDAD FARROKHI 66 demonstrated the feasibility of computer assisted image analysis for microscopy in local settings. These efforts represent some of the earliest attempts to apply deep learning related methods to hematologic image analysis using domestic data. Ensemble approaches have also been proposed, combining algorithms such as gradient boosting, support vector machines, and neural networks for survival prediction in leukemia patients. Data from specialty hospitals, including age, leukemia type, gene expression levels, and treatment histories, have been used to build models with high area under the curve values, outperforming individual algorithms and illustrating the benefit of model combination in complex prognostic tasks. Some groups have analyzed RNA sequencing data from Iranian Acute Lymphoblastic Leukemia patients using deep neural networks to differentiate high and low risk subgroups. Autoencoder layers have been used for dimensionality reduction and extraction of key genetic patterns, demonstrating the feasibility of genomic analysis with relatively limited patient numbers when leveraging deep learning to capture essential features. Recurrent neural networks, including long short term memory architectures, have been employed to predict treatment response to targeted therapies in Chronic Myeloid Leukemia patients. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 67 Using sequential laboratory data, these models have accurately forecasted response trajectories and provided some of the first examples of applying deep learning to medical time series data in the local context. Clinical decision support systems have also been tested. Some designs analyze daily hospital data to predict disease exacerbation or therapy adjustment needs, using machine learning backends and clinician facing dashboards. These systems illustrate how artificial intelligence can be embedded into hospital information environments to provide practical guidance. Overall, domestic leukemia modeling studies show growing progress and diversification. Despite limitations such as small datasets, limited computational infrastructure, and lack of standardization across centers, there is a clear trend toward modern algorithms, genomic analysis, and real world system development. Future research that emphasizes multimodal data integration, advanced deep learning methods, and collaboration between universities, hospitals, and research institutes may drive major advances in artificial intelligence powered leukemia diagnosis and prognosis within the country. Summary of Theoretical Foundations and Conceptual Framework Reviewing the theoretical foundations of DR. MEHRDAD FARROKHI 68 leukemia and the applications of artificial intelligence in medical diagnosis, particularly in genomic and imaging contexts, indicates that advanced machine learning and deep learning algorithms can significantly improve the quality, accuracy, and speed of leukemia detection and classification. Earlier sections discussed leukemia types, risk factors, diagnostic and therapeutic methods, together with information technology based tools, especially machine learning, and their potential for clinical data modeling. Findings from prior studies suggest that deep neural networks, especially convolutional neural networks, are highly effective for blood smear image analysis and for distinguishing normal from blast cells. Meanwhile, models such as random forest, gradient boosting, and support vector machine perform strongly with structured laboratory data, particularly when datasets are of modest size. Long short term memory based models and other recurrent architectures have been applied for time series predictions, such as disease progression and treatment response, by modeling temporal patterns in clinical data. Given the diagnostic sensitivity and complexity of leukemia, relying on a single algorithm is often insufficient. Hence, recent trends emphasize hybrid and multimodal models that integrate laboratory features, imaging data, genomic information, and treatment histories. By combining these inputs in joint learning ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 69 layers, final decisions become more accurate and robust. Hybrid approaches such as convolutional neural network plus multilayer perceptron, or gradient boosting plus recurrent neural networks, have consistently shown superior performance in complex medical contexts compared with single model strategies. A review of domestic studies shows that, while Iran is in relatively early stages of developing intelligent leukemia diagnostic systems, notable efforts have already applied classification algorithms and neural networks to local patient data. These studies indicate that even with small datasets, careful model tuning, feature selection, and dimensionality reduction can achieve acceptable predictive accuracy and provide valuable decision support. Data infrastructure has also been highlighted as critical. Resources such as publicly available image and genomic repositories are foundational for model training and validation. Because centralized datasets are limited domestically, collaboration between medical institutions and universities to establish local repositories of images, genomic profiles, and treatment records is strongly recommended. Evaluation metrics were also discussed, and it was emphasized that relying solely on accuracy is insufficient. Metrics such as sensitivity, specificity, F1 score, and area under the curve must be considered together. In leukemia, where false DR. MEHRDAD FARROKHI 70 negatives carry especially high costs because of missed diagnoses and delayed treatment, sensitivity is of paramount importance, although specificity and precision also remain relevant to avoid unnecessary interventions. This conceptual framework underpins the subsequent chapters, which will cover model design, data collection, algorithm implementation, and results analysis. The main hypothesis is that combining deep learning with multimodal datasets can significantly enhance leukemia diagnosis performance compared with conventional methods and single source models, thereby supporting more precise and timely clinical decision making. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 71 DR. MEHRDAD FARROKHI 72 2. AI FOR HARD-TO-TREAT NEUROLOGICAL CANCERS Background Neurological cancers form a dangerous category of cancers that cause severe harm to patients because they lead to poor survival rates and high incidence rates. The three brain tumours glioblastoma and diffuse intrinsic pontine glioma and specific types of medulloblastoma are classified as hard to treat because of their aggressive nature and unfavourable treatment outcomes and restricted available treatment choices. Glioblastoma functions as the most common aggressive brain tumour that affects adults because it makes up 50% of gliomas and patients survive only 15 to 18 months with current medical interventions. The treatment of diffuse intrinsic pontine glioma and other pediatric tumours remains challenging because these tumours exist in locations that prevent surgical removal and conventional radiotherapy offers only short term benefits. The worldwide healthcare system faces a significant problem because these cancers present ongoing treatment obstacles to both patients and medical staff. The distinct biological characteristics of these tumours make them resistant to treatment. 73 The protective blood brain barrier and intra tumoral heterogeneity and adaptive signalling networks function as multiple obstacles that prevent drugs from effectively targeting their intended sites. The process of traditional clinical decision making becomes more challenging because it requires the combination of multiple data sources which include imaging results and histopathology findings and genomic sequencing data and ongoing clinical patient records for proper diagnosis and treatment planning. The data contains complex nonlinear patterns that humans cannot effectively understand. People now understand that artificial intelligence and machine learning methods can effectively address these issues. The application of artificial intelligence technologies in radiology and pathology and molecular profiling has revolutionized oncology practice. Artificial intelligence systems achieve better performance through large scale datasets that enable them to identify complex patterns that traditional statistical models cannot match for improved image segmentation and tumour classification and biomarker discovery. Artificial intelligence based models in neuro oncology show successful results for MRI based tumour boundary definition and glioma grading and molecular subtype prediction. Radiomic and radiogenomic methods allow doctors to create individualized DR. MEHRDAD FARROKHI 74 treatment plans through their ability to link imaging characteristics with genetic changes. Artificial intelligence technology shows promise for neurological cancer treatment but its application exists in the initial stages of development. The lack of sufficient high quality brain tumour datasets hinders the development of dependable algorithms because it makes it difficult to train and validate them. The use of different imaging protocols across institutions leads to inconsistent results when trying to obtain accurate outcomes. Artificial intelligence models need to be implemented in clinical environments as soon as possible because they need to solve essential privacy protection problems and create transparent prediction systems and unbiased data platforms. The drive toward artificial intelligence based healthcare continues to accelerate and neuro oncology will gain significant advantages because it requires immediate development of innovative diagnostic and therapeutic methods. This chapter presents an in depth evaluation of artificial intelligence based treatments for neurological cancers that show resistance to conventional therapies. The first section provides information about tumour biology and clinical aspects before discussing artificial intelligence applications for tumour diagnosis and treatment planning and drug discovery and prognosis. The research paper identifies existing barriers to ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 75 practical deployment of proposed methods while proposing three promising research directions that include federated learning and multi modal data integration and large scale foundation models. This chapter shows how artificial intelligence technology provides benefits and drawbacks to neuro oncology that will enhance precision medicine for treating dangerous brain tumours. Artificial intelligence has significantly transformed the diagnosis, treatment, and overall management of hard to treat neurological cancers, particularly glioblastoma and diffuse midline gliomas. Recent progress has demonstrated clinical grade performance in multiple areas, ranging from diagnostic accuracy comparable to or surpassing human specialists to real time guidance systems used during neurosurgical procedures. The integration of machine learning algorithms with advanced neuroimaging has created new opportunities for highly personalized cancer care, and several AI systems are now nearing regulatory approval for implementation in clinical environments. The most notable breakthrough is found in diagnostic applications, where deep learning models reach exceptional accuracy in brain tumor classification and subtype identification. Current findings show that AI systems can match or exceed the diagnostic capabilities of experienced DR. MEHRDAD FARROKHI 76 neuroradiologists, signaling a major shift in clinical workflows and diagnostic standards. Modern deep learning architectures have reshaped diagnostic practices for highly aggressive brain tumors. An extensive clinical study developed an automated diagnostic system trained on tens of thousands of patient cases that achieved strong multiclass accuracy across numerous brain tumor categories. When neuroradiologists incorporated AI support into their assessments, their diagnostic accuracy showed significant improvement compared to unaided evaluation. This result highlights the clinical advantage of a collaborative model where human reasoning and artificial intelligence operate together to achieve superior performance. The use of transfer learning techniques has gained attention due to impressive performance in brain tumor detection tasks. Models based on optimized convolutional networks have demonstrated extremely high accuracy in identifying tumor presence, setting new standards for automated diagnostic capability. These systems have shown particular strength in distinguishing glioblastoma from other brain tumor types, and this is essential for rapid intervention and treatment planning in urgent clinical situations. Research efforts are increasingly focused on rare and highly malignant tumor types. Innovative hybrid methods combining whole slide digital ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 77 pathology with multiparametric MRI have been developed to classify glioma molecular subtypes, achieving competitive results in global evaluations. For diffuse midline gliomas, survival prediction models based on deep learning have outperformed traditional techniques, demonstrating strong predictive power even in external patient cohorts. Radiomics combined with machine learning has significantly improved prognostic strategies in neuro oncology. By extracting and analyzing detailed quantitative imaging features, these methods have produced robust survival prediction models for glioblastoma. Such approaches combine radiomic features with characteristics derived from deep learning applied to preoperative imaging, helping clinicians estimate outcomes more accurately and adjust treatment plans accordingly. Multi parametric imaging work has shown that combined diffusion weighted and perfusion weighted MRI features enhance prognostic performance when compared to the use of clinical information alone. Through careful feature selection procedures, strong predictive models have been developed that surpass traditional risk classification strategies, offering more precise patient stratification. Recent studies highlight the essential integration of clinical data, molecular biomarkers, and DR. MEHRDAD FARROKHI 78 imaging features for predicting therapeutic response. Machine learning approaches that include MGMT methylation status together with radiomic markers successfully differentiate true tumor progression from pseudoprogression, which remains one of the most challenging diagnostic dilemmas in glioblastoma management. These multimodal strategies indicate the direction of future precision neuro oncology approaches. Deep learning contributions to brain tumor imaging have also reached advanced performance in automated segmentation. Volumetric convolutional neural networks combined with 3D modeling techniques provide high quality segmentation of tumor regions and assist in monitoring treatment response through objective and repeatable measurements. These networks have demonstrated competitive performance on benchmark datasets widely adopted in neuroimaging research. Comprehensive evaluations show that ongoing advancements in deep learning have significantly impacted the field of medical image analysis. The progression from early convolutional architectures to optimized systems capable of self adapting network configurations has supported state of the art segmentation across diverse imaging modalities. Convolutional auto encoder models have also shown high accuracy in ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 79 brain tumor classification tasks, successfully identifying common tumor categories and exhibiting excellent performance across internal validation metrics. AI supported systems for neurosurgical procedures represent another rapidly advancing development. Systematic reviews of extensive research demonstrate a wide variety of applications, including recognition of surgical workflows, real time tool tracking during microsurgery, and intelligent navigation systems designed to compensate for brain shift and increase tumor resection precision. These innovations further contribute to enhanced patient safety and surgical training. Clinical validation has confirmed the practical effectiveness of artificial intelligence in neurosurgical environments. A large multicenter trial demonstrated that an AI based intraoperative imaging system using stimulated Raman histology produced diagnostic accuracy comparable to traditional pathology, while delivering results within seconds instead of the longer time required for standard processing. The speed and reliability of this approach support its potential for widespread adoption in operating rooms. Major advancements have also been reported in intraoperative detection of tumor infiltration. Deep learning systems have been tested across DR. MEHRDAD FARROKHI 80 international medical centers and demonstrated strong performance in identifying infiltrative boundaries that are often missed using standard inspection techniques. These systems have shown the ability to reduce the likelihood of leaving behind high risk residual tumor tissue, which contributes to earlier recurrence. Emerging temporal deep learning methods have improved monitoring of tumor recurrence, especially in pediatric neuro oncology. By analyzing sequential brain imaging, these models can forecast relapse risk with high accuracy and help guide personalized surveillance strategies. The development of standardized clinical guidelines for artificial intelligence evaluation ensures that new applications meet strict safety, quality, and reproducibility requirements. The integration of artificial intelligence with neuro oncology has fundamentally changed clinical capabilities across all stages of cancer care. Diagnostic systems that outperform human expertise in specific tasks and real time surgical technologies validated in clinical practice illustrate the maturity of current AI innovations. With regulatory pathways becoming more defined and validation standards established, these systems are poised to broaden access to expert level brain tumor care and support highly individualized treatment strategies for patients facing the most severe neurological cancers. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 81 of artificial intelligence in neuro oncology requires access to sensitive personal health information. Protecting privacy, obtaining informed consent, and meeting regulatory compliance remain ongoing challenges. Training data frequently contains embedded biases based on demographic information and tumour subtypes and institutional practices which can lead to inequitable outcomes in care. Safe implementation of artificial intelligence requires strong regulatory frameworks and equal access to artificial intelligence technologies so that benefits serve all patients. Biological and clinical complexity further limits artificial intelligence progress because brain tumours display significant biological diversity within individual tumours and between different patients. The complex biological mechanisms of cancer challenge even advanced algorithms because these models may oversimplify disease behaviour or fail to detect rare subtypes. Neuro oncology treatment outcomes depend on tumour biology and individual patient factors including age and comorbidities and functional status which artificial intelligence must consider. Artificial intelligence models require continued development to process multiple types of clinical and biological information to produce accurate and reliable predictions. Artificial intelligence demonstrates great potential to revolutionize neuro oncology yet DR. MEHRDAD FARROKHI 94 multiple significant obstacles continue to exist. The solution to these problems requires collaboration across institutions to create high quality datasets and develop transparent models and ensure seamless clinical workflow integration along with strong ethical governance systems. Delivering the benefits of artificial intelligence to patients with hard to treat neurological cancers requires addressing the current limitations that restrict progress. Future Directions The field of neuro oncology is expected to undergo significant transformation through artificial intelligence implementation during the upcoming decade even though its development remains in an early stage. The creation of new technological strategies shows great potential to solve present challenges while advancing precision medicine for hard to treat neurological cancers. Multi modal data integration will become increasingly important because artificial intelligence models of the future will unite imaging information with histopathology results and genomic profiles and proteomic measurements and clinical records to develop fully integrated predictive systems. The combination of diverse diagnostic methods enables researchers to analyse brain tumours within their complete biological and clinical context which improves accuracy in diagnosis ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 95 and treatment planning. Transformer based deep learning architectures support the integration of multiple large datasets, while the combination of advanced imaging techniques will establish new connections between imaging outcomes and molecular alterations which will accelerate biomarker development and treatment strategy innovation. Federated and collaborative learning represents a strong solution to the problem of restricted neuro oncology data availability and fragmented distribution. Federated frameworks allow institutions to train models independently without sharing raw data which protects privacy while benefiting from diverse and extensive patient cohorts. Large scale deployment of these techniques would improve current data limitations and reduce institutional biases which would allow artificial intelligence models to function more effectively across different clinical environments. Achieving success requires support from international consortia and multi centre collaborations. Explainability and clinical trust must improve for artificial intelligence tools to gain widespread clinical adoption. Future research will focus on creating explainable artificial intelligence methods that provide clear results through visual heatmaps and feature importance scoring and natural language descriptions. These developments will increase physician confidence DR. MEHRDAD FARROKHI 96 in artificial intelligence systems which will act as supportive decision tools instead of replacing healthcare expertise. The combination of artificial intelligence predictions with current clinical guidelines will further enhance their practical value. Foundation models and large language models show rising potential for medical applications because success in natural language processing and computer vision continues to expand into healthcare. Neuro oncology will benefit from foundation models which serve as general purpose systems that combine diverse data sources with existing medical knowledge to produce predictions and generate personalized treatment strategies. The ability of these models to scale and adapt will drive progress faster than traditional task specific approaches can support. The overall goal of artificial intelligence in neuro oncology focuses on developing precision medicine systems that provide specific treatments to individual patients at optimal stages of care. Artificial intelligence systems may eventually operate as digital twins that create virtual patient models capable of predicting disease progression and treatment responses. These virtual simulations allow clinicians to test therapies while reducing risks and generating valuable clinical insights. The new technologies show strong potential to improve outcomes for aggressive brain tumours because they can guide ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 97 timely intervention when treatment delays can lead to severe consequences. The future of artificial intelligence in neuro oncology depends on integrating improved data access with privacy preserving collaboration methods and explainable systems and strong foundation model architectures. The partnership between technological progress and clinical requirements and ethical principles enables artificial intelligence to transform the treatment of hard to treat neurological cancers into a more effective data driven system of personalized care. Conclusion The medical field recognizes glioblastoma and diffuse intrinsic pontine glioma and aggressive medulloblastoma as among the most dangerous cancer types. Many decades of research have not produced meaningful improvements in patient outcomes which demonstrates that healthcare must adopt new approaches to solve its pressing problems. Artificial intelligence technology has emerged as a powerful tool that supports diagnosis and treatment planning and drug development and patient outcome prediction. This chapter has outlined how artificial intelligence is reshaping neuro oncology by improving MRI tumour segmentation with convolutional neural networks and enhancing workflow speed and clinical accuracy with predictive models that combine radiomics DR. MEHRDAD FARROKHI 98 data with genomics information and clinical records. Machine learning technology has become essential for drug discovery and drug repurposing because it accelerates the creation of new treatment possibilities for patients who currently have limited options. The application of these models in clinical settings continues to face multiple barriers that arise from limited data access and unclear algorithmic behaviour and difficulties in system implementation and ongoing medical ethics concerns. Healthcare systems can overcome these obstacles through extensive collaboration and standardized data collection practices and transparent trustworthy modelling approaches that satisfy clinical requirements. Meaningful progress in the field will occur through the combination of multiple data sources with federated learning and explainable artificial intelligence and foundation models. The integration of artificial intelligence with clinical expertise will support the development of customized treatment strategies for challenging brain cancers which outperform traditional medical approaches and offer renewed hope for improved patient outcomes. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 99 DR. MEHRDAD FARROKHI 100 3. AI FOR HARD-TO-TREAT GASTROINTESTINAL CANCERS Background Hard to treat gastrointestinal cancers include several malignancies that remain among the deadliest in oncology. These diseases often present late, progress rapidly, and demonstrate resilience against conventional therapies. Pancreatic cancer, cholangiocarcinoma, metastatic colorectal cancer that is microsatellite stable, select forms of hepatocellular carcinoma, refractory gastric cancer, and some gastrointestinal stromal tumors are key examples where the standard of care improves survival only marginally. Their biological heterogeneity, frequent genomic instability, and involvement of dense stromal or immune suppressive tumor environments contribute to the difficulty of effective detection and intervention. For many patients, the first diagnosis coincides with an advanced stage when curative treatments are no longer feasible. This trend reflects limitations in early screening, slow symptom development, and insufficient biomarkers that can reliably detect disease onset or therapeutic response. The global burden of gastrointestinal cancers continues to rise, driven by population aging, 101 lifestyle changes such as increasing obesity and sedentary behavior, and dietary patterns rich in processed foods. In low resource settings, limited access to medical imaging, specialty care, and timely surgical intervention further deepens disparity. Mortality rates remain high not only because these tumors are aggressive but also because healthcare systems struggle to keep pace with emerging treatments, molecular diagnostics, and personalized care pathways. There is an urgent need for technologies that can accelerate diagnosis, improve decision making, optimize therapy selection, and provide ongoing monitoring using minimally invasive tools. Artificial intelligence has emerged as a promising catalyst for transformation in this field. The ability of algorithms to integrate high dimensional clinical, genomic, imaging, and pathology data aligns well with the complexity inherent to hard to treat gastrointestinal malignancies. AI can uncover subtle patterns invisible to human interpretation, predict risk and treatment responses, and guide therapy personalization. In doing so, it may reduce the historical disadvantage these cancers have faced, particularly where conventional screening and treatment strategies fall short. AI in Early Detection and Screening Delayed diagnosis is a major determinant of DR. MEHRDAD FARROKHI 102 poor outcomes in cancers such as pancreatic or cholangiocarcinoma. Symptoms are often vague, imaging may overlook small lesions, and current blood biomarkers like CA19 9 lack sufficient sensitivity and specificity. AI fueled screening tools are being developed to bridge these gaps. Machine learning models can analyze electronic health records to detect early risk signatures long before clinical suspicion arises. For example, patterns in laboratory values, weight change trajectories, diabetes onset, and abdominal imaging reports can signal elevated risk for pancreatic cancer. Rather than waiting for a tumor to reach a detectable size, risk stratification enables targeted monitoring and timely referral for imaging or molecular testing. Deep learning techniques applied to radiography, CT scans, and MRI can significantly enhance sensitivity in identifying small tumors or precancerous lesions. Algorithms trained on large image datasets can highlight subtle textural differences that precede visible changes. In bile duct cancers, AI analysis of imaging may reduce misclassification of benign strictures and shorten the diagnostic interval. Endoscopic technologies benefit in similar ways. Computer vision can assist gastroenterologists by identifying early lesions in the stomach or colon with greater precision, reducing miss rates associated with human fatigue or lack of visual contrast. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 103 Integration of liquid biopsy data is another promising avenue. Complex biomarker panels, including circulating tumor DNA and extracellular vesicle signatures, require interpretation of multidimensional molecular landscapes. AI models can improve diagnostic accuracy and potentially distinguish between malignant and benign conditions. When systems combine multiple data sources, they achieve performance exceeding any single modality. This multimodal approach is particularly appropriate for cancers that evade detection through a single standard test. AI for Personalized Treatment Planning Treatment of hard to treat gastrointestinal cancers often involves combinations of surgery, radiation, chemotherapy, immunotherapy, and targeted agents. However, many patients fail to respond or experience severe toxicity due to biological variability. Precision oncology calls for better prediction of who will benefit from which treatment strategy. Machine learning models can analyze tumor genomic profiles, transcriptomic signatures, radiomic traits, and patient clinical characteristics to estimate likely drug responses. These tools may advise clinicians when a patient is unlikely to benefit from a conventional regimen and should instead pursue experimental therapies or clinical trials. DR. MEHRDAD FARROKHI 104 For example, microsatellite stable colorectal cancer tends to show limited response to immune checkpoint inhibition. AI can help stratify subsets within this group that have immune active tumor microenvironments and might still respond to immunotherapy combinations. In pancreatic cancer, where surgical resection offers the only realistic chance of cure, predicting which patients are likely to tolerate major surgery and derive long term benefit is essential. AI models incorporating performance status, nutritional state, tumor anatomy, and biological markers can inform multidisciplinary discussions and avoid futile interventions. Radiomics, the extraction of quantitative features from imaging, has grown in importance. Subtle spatial patterns in tumor tissue captured in CT or MRI can correlate with molecular phenotypes and therapy sensitivity. AI systems translate thousands of radiomic variables into clinically actionable insights. Such approaches offer a noninvasive option for patients whose tumors are difficult to biopsy or exhibit strong intratumoral heterogeneity. Personalization also includes adaptive treatment adjustments. Throughout therapy, AI driven monitoring can detect early resistance, prompting rapid transitions to alternative strategies rather than allowing disease progression. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 105 AI in Surgical and Interventional Decision Support For many gastrointestinal cancers, surgery or locoregional interventions such as ablation or embolization remain crucial in improving survival. The complexity of these procedures demands meticulous preoperative planning. AI can enhance operative precision by mapping tumor boundaries, vascular anatomy, and potential lymph node involvement with high accuracy. Augmented intelligence platforms allow surgeons to visualize decision critical structures intraoperatively using real time image analysis, reducing the likelihood of incomplete resections. Robotic surgery and advanced endoscopic procedures benefit from AI based guidance that stabilizes instrument motion, warns of proximity to vital tissues, and automates repetitive actions. In liver cancer care, patient selection for resection, transarterial therapies, or transplantation is a nuanced process. Predictive modeling supports hepatologists in determining which patients stand to gain meaningful survival benefit while minimizing risk of postoperative liver failure. By integrating sequential clinical data, AI can also update risk models continuously rather than relying on static scoring systems. AI for Drug Discovery and Repurposing DR. MEHRDAD FARROKHI 106 The pipeline for developing new therapeutics for aggressive gastrointestinal tumors is expensive and slow. AI offers pathways to shorten discovery timelines and reduce failure rates. Computational models can simulate drug target interactions, identify vulnerabilities in tumor signaling networks, and propose novel compound structures with optimized pharmacodynamic properties. In cancers with limited actionable mutations, such as many pancreatic tumors, AI aided screens may uncover alternative dependencies or synthetic lethal relationships that could be drugged effectively. Drug repurposing strategies, powered by AI scanning of pharmacologic and clinical trial databases, reveal unexpected applications of existing medications. Given that many gastrointestinal cancer patients cannot wait years for new therapies, repurposed agents that have already passed toxicity evaluation represent an attractive opportunity. AI can match biological signatures of resistant cancers with suitable compounds originally developed for other diseases, opening new therapeutic routes. Another compelling area is the design of combination regimens. Synergistic effects among targeted agents, immunomodulators, and cytotoxic drugs are difficult to predict due to complex tumor microenvironment interactions. Machine learning models can forecast synergistic ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 107 or antagonistic drug interactions before clinical testing, reducing the trial and error burden. These methods help rationally construct regimens tailored to tumor biology while minimizing unnecessary toxicity. AI Based Monitoring and Survivorship Support After initial treatment, patients with hard to treat gastrointestinal cancers face a high risk of recurrence. Traditional follow up models often rely on periodic imaging and symptom reporting, which may delay recognition of relapse. AI enriched monitoring platforms employ continuous data streams from wearable sensors, laboratory tests, and digital health records to identify subtle shifts indicating disease progression or treatment toxicity. Earlier detection of relapse enables intervention during windows when disease remains manageable. Patient reported outcomes are increasingly recognized as essential in guiding supportive care. Natural language processing can extract clinically relevant insights from patient communication, such as messages sent through electronic portals or conversational inputs captured by virtual assistants. These analyses alert care teams to concerning patterns like worsening pain, nutrition decline, or psychological distress. Personalized care recommendations based on DR. MEHRDAD FARROKHI 108 predictive analytics can improve quality of life even for patients with limited survival prospects. Survivorship extends to caregivers and families who face significant emotional and logistical challenges. AI powered tools can provide education, coordinate appointments, and streamline access to resources. They reduce the cognitive load on patients during a vulnerable period marked by anxiety and complex treatment pathways. Challenges and Ethical Considerations Despite progress, there are persistent barriers to integrating AI into routine care for gastrointestinal oncology. One major challenge is data availability and quality. Rare cancer subtypes often lack large curated datasets required to train robust models. Biological diversity within tumors further complicates the situation. Without representative data, models may perform poorly across populations or exhibit bias. Ensuring equitable model performance across demographic groups is critical, especially considering known disparities in cancer outcomes. Interpretability is another pressing issue. Clinicians are understandably hesitant to rely on systems whose reasoning is opaque. There is a growing movement toward explainable AI that can articulate which factors are driving a ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 109 given prediction. Interpretability is even more important in high risk decisions like treatment escalation or withdrawal. AI should supplement not replace clinical judgment, providing confidence rather than confusion in complex decision making. Regulatory frameworks must evolve to match the pace of innovation. Many AI systems continuously update as new data arrives, challenging traditional approval processes that assume static medical devices. Accountability in the event of incorrect predictions must be clearly defined. To build trust, transparency in model development, validation, and monitoring is required. Additionally, patient privacy and informed consent remain paramount. AI systems thrive on large scale data collection, yet sensitive health information must be protected. Ethical usage demands that patients understand how their data will be used and can opt out without compromising their care. Collaboration between industry, government regulation, and healthcare systems is essential to ensure responsible deployment. Future Directions and Opportunities The future of AI in managing hard to treat gastrointestinal cancers is promising. As multimodal datasets grow, models will become increasingly comprehensive and accurate. DR. MEHRDAD FARROKHI 110 Integration of real world clinical data from across diverse populations will help refine risk stratification tools and expand personalized therapy approaches. Cross collaboration between oncologists, surgeons, radiologists, data scientists, and patient advocates will shape development priorities toward meaningful impact. Emerging technologies like federated learning allow AI models to train across distributed databases without transferring patient data, addressing privacy concerns while expanding access to large cohorts. Advanced generative models can simulate virtual cancer patient populations to test therapeutic strategies before implementation. Digital twins of tumors may help oncologists experiment with therapies in silico, predicting response and resistance developments dynamically. In endoscopy and pathology, real time AI will become a standard feature, increasing diagnostic accuracy and reducing workload. Predictive care pathways will automate parts of the clinical workflow, directing high risk patients to expedited evaluation while reducing unnecessary procedures for low risk cases. Seamless connection between AI insights and clinical action is key, supported by interoperable digital infrastructure. Patient empowerment will continue to grow as AI enabled tools bring clarity to complex treatment decisions. Personalized education, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 111 emotional support, and treatment navigation will be integrated into holistic care models. As outcomes improve even modestly in the short term, increased survival will lead to further advancements by expanding opportunities for clinical research and treatment refinement. Importantly, continuous evaluation of ethical, legal, and psychological impacts must accompany technological progress. The ultimate goal is not to replace human expertise but to augment the capabilities of clinicians and expand hope for patients who have historically faced limited options. Early commitments to equity, trust, and patient centered design will determine whether AI fulfills its transformative promise. Conclusion Hard to treat gastrointestinal cancers represent one of the most formidable challenges in current oncology due to biological aggressiveness, late detection, and limited treatment responses. Artificial intelligence has become a crucial tool in the mission to change this landscape. With capabilities spanning early detection, personalized therapy guidance, advanced surgical planning, accelerated drug discovery, and improved patient monitoring, AI stands to enhance every stage of the cancer care continuum. The technology is not without its hurdles, particularly in data quality, interpretability, regulatory oversight, and ethical DR. MEHRDAD FARROKHI 112 responsibility. Yet through careful collaboration and continued innovation, AI can shift outcomes in diseases where progress has long been constrained. The journey is still in its early stages. Success will depend on responsible deployment and a patient centered approach that respects clinician expertise and societal values. By harnessing the full potential of artificial intelligence, the field may finally gain momentum against cancers that have remained unrelenting for decades. The future holds real possibility that early diagnosis becomes more common, treatments more effective, and quality of life more protected, bringing long awaited advances to those confronting some of the most challenging gastrointestinal malignancies. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 113 still experience limitations such as high false positive results, reader dependent variability, and reduced sensitivity for early stage lesions. Clinical screening trials have shown that low dose computed tomography reduces lung cancer mortality in high risk populations, yet widespread implementation is hindered by limited accessibility, eligibility restrictions, insurance reimbursement challenges, and operational costs. Research has increasingly focused on blood based biomarkers including circulating tumour DNA as complementary strategies for early lung cancer detection. Liquid biopsy technology enables the identification of molecular and genetic alterations that may appear before radiological detection, increasing opportunities for timely therapeutic intervention. Combining artificial intelligence assisted imaging with biomarker based approaches allows a more precise and efficient screening strategy that improves diagnostic accuracy, supports individualized patient pathways, and reduces unnecessary follow up procedures. These advances hold strong potential to transform detection and treatment, ultimately improving survival and patient quality of life. Artificial intelligence strategies are also expanding into prognosis prediction and treatment response evaluation. Machine learning models can analyse extensive clinical, molecular, DR. MEHRDAD FARROKHI 126 and imaging datasets to forecast patient outcomes, enable therapeutic regimen selection, and identify individuals most likely to benefit from immunotherapy or targeted therapy. These capabilities help optimize resource allocation, reduce ineffective treatment exposure, and enhance personalized management in both early and advanced stage lung cancer. Large language models continue to emerge as promising tools within lung cancer care due to their ability to respond to clinical queries in free text format without requiring specific task training. They enable rapid interpretation of large amounts of medical knowledge and offer support in clinical decision assistance, patient counselling, and clinical trial selection. Medical chatbots based on large language models can generate responses comparable to clinician replies in both factual accuracy and communication quality. Artificial intelligence systems also facilitate efficient extraction of patient data for matching individuals with appropriate clinical trial eligibility criteria. However, challenges remain because large language models may generate incorrect or fabricated information as a result of relying on patterns in text rather than true comprehension. For these reasons, human supervision is essential to ensure safe and reliable integration into clinical workflows. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 127 Approved AI Devices in Lung Cancer The implementation of artificial intelligence algorithms into clinical care for hard to treat lung cancer requires formal regulatory authorization to ensure safety and effectiveness. Rapid development in artificial intelligence technologies challenges existing regulatory frameworks and requires sufficient review mechanisms. Regulatory bodies classify artificial intelligence based medical tools according to patient risk levels. Many oncology artificial intelligence applications are categorized as moderate risk devices where randomized controlled trial evidence may not always be mandatory. Although approved artificial intelligence technologies demonstrate strong performance in use cases such as lung nodule detection, diagnostic classification, and radiotherapy planning, validation across diverse populations remains essential to ensure reliable real world performance. Regulatory approvals primarily focus on imaging based applications, and successful adoption requires ongoing collaboration among technology developers, manufacturers, healthcare institutions, and policymakers. Refining regulatory standards, creating standardized development protocols, and maintaining post market surveillance are necessary to support safe translation into clinical practice. Such coordinated efforts will strengthen DR. MEHRDAD FARROKHI 128 precision oncology delivery for patients with hard to treat lung cancer. Challenges and Opportunities in AI for Lung Cancer Artificial intelligence offers significant promise for lung cancer detection and management, yet clinical implementation continues to encounter major barriers. Access to large and high quality datasets from different clinical institutions is necessary to train and validate robust artificial intelligence models. However, limitations in data sharing because of privacy concerns, regulatory requirements, and ownership issues remain a major obstacle. Methods such as centralized learning initiatives, anonymized public databases, and federated learning approaches offer partial solutions with different benefits and constraints. Bias and fairness considerations are crucial because artificial intelligence models may inadvertently favor specific demographic or socioeconomic groups, creating or worsening disparities in healthcare delivery. Interpretability presents another key challenge because many deep learning models operate as complex black box systems that provide limited insight into how decisions are produced. Lack of transparency can hinder clinician acceptance and reduce practical usability in critical decision processes. Reproducibility and generalizability represent ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 129 vital requirements because differences in imaging protocols, motion artifacts, scanner variations, noise levels, and inconsistencies in radiologist annotation can diminish model reliability. The adoption of standardized preprocessing workflows, structured reporting guidelines, and standardized feature extraction initiatives contribute to improved consistency in artificial intelligence outcomes. Continued attention to these concerns supports safe and effective deployment in clinical environments and improves precision medicine in lung cancer. Pathology Once a suspicious lesion is detected, tissue biopsy remains essential for definitive diagnosis. Traditionally, pathologists review slides manually to identify cancer cells and evaluate features such as tumor grade and biomarker expression. With recent advances, pathology slides can now be digitized into high resolution images, enabling artificial intelligence to support this process. Deep learning algorithms are able to scan entire slides and highlight regions of interest, improving speed and reducing the risk of oversight. These systems serve as valuable diagnostic partners to pathologists by identifying very small tumor deposits, such as early nodal metastases in breast cancer. Artificial intelligence is also increasingly used for quantitative biomarker assessment, including proliferation and receptor expression DR. MEHRDAD FARROKHI 130 markers, providing consistent measurements that contribute to treatment planning. Similar progress is observed in lung cancer pathology, where algorithms have demonstrated the ability to classify subtypes on standard stained slides and predict genetic mutations based on microscopic tissue architecture. These computational pathology technologies hold promise for more automated and quantitative interpretation of tissue specimens. In summary, artificial intelligence has become integrated into multiple aspects of cancer diagnosis, including interpretation of computed tomography scans and mammograms, digital slide analysis, and detection of subtle features that may otherwise be missed. These systems have already entered routine practice in some settings, with several receiving regulatory clearance for identifying lung nodules or breast abnormalities. Early adopters report that artificial intelligence acts as a dependable assistant, improving diagnostic speed and enhancing accuracy. As clinical experts note, the purpose of artificial intelligence is to support rather than replace clinical judgment, helping healthcare providers make better decisions, reduce diagnostic errors, and improve survival outcomes. Personalizing Treatment with AI Every cancer develops uniquely, which requires ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 131 treatment strategies tailored to each individual. Artificial intelligence supports precision oncology by integrating genomic markers, clinical characteristics, and imaging features to recommend therapies that are more likely to be effective. In breast cancer, machine learning can predict responses to targeted therapy and immunotherapy by analyzing tumor microenvironment properties and biological traits, making treatment choices more accurate. In lung cancer, artificial intelligence can identify actionable mutations and recognize early indicators of therapeutic resistance, such as resistance to epidermal growth factor receptor inhibitors in non small cell lung cancer. These predictive capabilities allow clinicians to modify treatments at an optimal time, improving patient outcomes and reducing exposure to ineffective medications. Predicting Outcomes with Precision Artificial intelligence is reshaping prognostic evaluation in breast and lung cancers. By combining radiomics, pathology, and clinical information, predictive models estimate recurrence risk and survival with notable accuracy. For patients with lung cancer, deep learning applied to computed tomography features contributes to forecasting long term outcomes, supporting personalized follow up planning and risk counseling. In breast cancer, DR. MEHRDAD FARROKHI 132 predictive analytics help identify patients who are unlikely to benefit from aggressive therapy, which reduces unnecessary chemotherapy while maintaining favorable prognoses. These advancements enable patients and clinicians to make more informed decisions, preparing for the future with clearer expectations. Conclusion and Future Direction Advancements in artificial intelligence have significantly influenced the field of lung cancer research and clinical management. Artificial intelligence supports early detection, screening accuracy, prognosis prediction, and treatment optimization while enabling personalized care through integrated analysis of complex data streams. Large language models and deep learning architectures show potential to enhance decision support and improve communication between clinicians and patients facing difficult therapeutic choices. However, regulatory approval processes, challenges in data access and privacy, bias risk, interpretability limitations, reproducibility issues, and generalizability concerns remain barriers to widespread clinical adoption. Continuous development of explainable artificial intelligence approaches, standardized data methodologies, and multi institution collaborations will be essential to facilitate safe integration of these systems ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 133 within healthcare. Looking ahead, artificial intelligence is poised to offer transformative contributions to precision oncology in lung cancer. Integration of artificial intelligence within multidisciplinary care pathways combined with rigorous validation and ethical oversight will support improved early detection, optimized treatment decisions, and enhanced survival and quality of life for patients experiencing this highly lethal disease. Continued collaboration between clinicians, researchers, regulatory agencies, and technology innovators remains crucial to achieving these advancements. DR. MEHRDAD FARROKHI 134 ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 135 5. AI FOR HARD-TO-TREAT UROGENITAL CANCERS Background Hard to treat urogenital cancers represent a major challenge in modern oncology because they often show aggressive biological characteristics, poor clinical responses to available therapies, and substantial diversity in tumor genetics and microenvironment features. These cancers arise in the urinary and genital systems, including prostate cancer in its metastatic or castration resistant forms, muscle invasive bladder cancer, advanced renal cell carcinoma, testicular cancers that relapse after first line therapy, and rare malignancies such as penile carcinoma or upper tract urothelial carcinoma. Patients diagnosed with these diseases may experience late stage detection, limited access to precision therapies, rapid development of resistance, and physical or psychological burdens that negatively affect survival and quality of life. Epidemiologically, prostate cancer remains one of the most commonly diagnosed malignancies in men around the world, but while early stage disease is often curable, metastatic castration resistant prostate cancer continues to carry a poor prognosis. Bladder cancer frequently recurs 136 and can progress to muscle invasive disease that requires complex multimodal treatment. Renal cell carcinoma shows unpredictable behavior and immunologic complexity, while late relapses in testicular cancer can require salvage regimens that lack optimal patient selection strategies. Health disparities and resource limitations further exacerbate outcomes, particularly where delayed diagnosis and limited access to expert oncology teams persist. Biomedical research over the last decade has made meaningful progress toward precision oncology, yet the clinical management of these complex cancers remains constrained by insufficient predictive tools and fragmented data availability. Imaging modalities, genomic sequencing, pathology interpretation, and clinical data all play a role in characterization and treatment selection, but integrating these diverse information sources remains an ongoing challenge. The urgent need for earlier diagnosis, improved risk assessment, optimized treatment allocation, and better monitoring of recurrence and toxicity motivates breakthroughs in computational and analytic techniques. Artificial intelligence is rapidly emerging as a powerful set of tools capable of changing the way clinicians detect, classify, and treat urogenital cancers. Through machine learning, deep learning, and decision support applications, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 137 AI offers abilities to analyze high dimensional data at a scale no human can match. The goal is not simply faster computation but deeper understanding of tumor biology, more accurate interpretation of medical information, and actionable predictions that improve outcomes. The integration of AI into urogenital oncology promises new opportunities in screening, image analysis, molecular discovery, treatment optimization, and personalized survivorship. This chapter examines how AI technologies are being applied across the care continuum for hard to treat urogenital cancers, the current limitations that must be addressed, and the future directions likely to define the next era of oncology innovation. AI in Early Detection and Risk Stratification Early detection remains one of the most powerful predictors of survival in urogenital cancers. Prostate cancer screening through prostate specific antigen testing has reduced mortality but also resulted in overdiagnosis and overtreatment. AI models have the potential to refine screening decisions by combining longitudinal PSA values with clinical history, imaging reports, and genetic risk markers. Risk prediction tools powered by machine learning can identify individuals who are more likely to harbor clinically significant tumors DR. MEHRDAD FARROKHI 138 and reduce unnecessary biopsies. Multiparametric MRI has become central to detecting prostate cancer, yet radiologic interpretation varies across clinicians. AI driven radiology solutions can detect suspicious regions, characterize tumor aggressiveness, and assist biopsy planning with improved consistency. These systems can improve cancer detection rates while lowering false positives. Algorithms that evaluate temporal changes in imaging can also support better surveillance strategies for patients on active monitoring programs. Bladder cancer screening remains difficult because common tests such as urine cytology have limited sensitivity. AI enhanced interpretation of urinary biomarkers, cellular morphology, and genomic signatures can increase diagnostic accuracy and may identify disease earlier than current methods. In renal cell carcinoma, incidental detection is common but prognosis depends heavily on stage at diagnosis. Machine learning applied to electronic health records can help recognize individuals at high risk due to genetic disorders, occupational exposures, or metabolic conditions, potentially enabling earlier imaging referrals. Testicular cancer often manifests with palpable symptoms, yet delayed medical attention among young men remains an issue. AI enabled educational outreach programs and digital symptom checkers can encourage earlier ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 139 evaluation, especially in underserved populations. Once a tumor is identified, algorithms that analyze serum markers and ultrasound images may support more accurate staging and help determine who can safely avoid overtreatment. Overall, intelligent risk stratification and early detection tools enhance precision at the front end of care, allowing clinicians to intervene during the window when cure is achievable. AI for Imaging Interpretation and Tumor Characterization Medical imaging plays a critical role in diagnosing and staging urogenital cancers, but conventional reading relies heavily on expert interpretation. AI methodologies transform imaging into quantitative data rich in biological information. In prostate cancer, deep learning networks trained on thousands of MRI scans can differentiate low grade from clinically significant disease more accurately than standard assessment alone. Algorithms can measure tumor volume, shape irregularities, and tissue texture that correlate with histopathology. For bladder cancer, cystoscopic visualization is the primary diagnostic technique. Automated image recognition can assist urologists by highlighting suspicious lesions in real time, reducing the number of tumors that escape detection. Fluorescent imaging techniques paired with AI can identify cancerous regions invisible DR. MEHRDAD FARROKHI 140 under white light endoscopy, improving resection completeness. In renal cell carcinoma, radiomics provides insights into tumor vascularity, necrosis, immune cell infiltration, and potential drug sensitivity. AI models that integrate CT imaging data with clinical and molecular indicators offer a platform for noninvasive tumor subtyping, helping clinicians distinguish indolent tumors from those that may require early systemic intervention. These imaging supported predictions can reduce unnecessary surgeries and guide the use of active surveillance. Metastatic disease evaluation also benefits from computational interpretation. In castration resistant prostate cancer, bone metastasis detection with whole body imaging can be challenging. AI systems outperform humans in identifying skeletal lesions and quantifying disease burden, improving treatment planning and prognosis assessment. Automated tracking of metastasis response to therapy supports adaptive treatment modifications that protect quality of life. The evolution of imaging into a data rich diagnostic platform depends significantly on AI, facilitating more comprehensive tumor characterization and more precise treatment strategies. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 141 AI for Molecular Profiling and Precision Therapy Advances in genomic sequencing have expanded knowledge of actionable mutations in urogenital cancers, yet connecting molecular data to effective therapies remains complex due to heterogeneity and limited biomarkers. Machine learning models can interpret genomic alterations, epigenetic changes, transcriptomic activity, and proteomic patterns to predict treatment response or resistance. In prostate cancer, alterations in DNA repair genes such as BRCA1 and BRCA2 indicate benefit from PARP inhibitors, but responses are variable across patient populations. AI can help identify additional molecular dependencies and reveal gene interaction patterns that modify treatment sensitivity. Predicting androgen receptor activity levels using transcriptomic signatures may optimize the use of hormonal therapies. Muscle invasive bladder cancer demonstrates significant variation in immunotherapy outcomes. AI enabled clustering of tumor molecular profiles supports improved patient selection for checkpoint blockade. Tumor microenvironment analysis through computational pathology can identify immune excluded phenotypes that may respond to combination therapy rather than monotherapy. DR. MEHRDAD FARROKHI 142 Renal cell carcinoma displays intricate immune and metabolic biology. AI tools can discover molecular subsets more likely to respond to VEGF targeted agents or immune checkpoint inhibitors, reducing unnecessary toxicity and improving survival. Multiomic integration helps reveal composite predictors far beyond what single biomarker strategies can achieve. For rare cancers such as penile carcinoma, small sample sizes hinder traditional analysis approaches. Transfer learning techniques allow models trained on more common cancer datasets to support clinical insights in low incidence diseases, potentially accelerating access to personalized therapies. By linking molecular complexity to therapeutic opportunity, AI fosters more precise and effective treatment selection across urogenital cancers. AI Enhanced Surgical and Interventional Planning Surgery and interventional oncology play primary roles in the management of many urogenital cancers. The quality of local control often determines long term outcomes, especially in diseases such as muscle invasive bladder cancer or testicular cancer. AI has a valuable impact in operative planning, intraoperative decision support, and postoperative evaluation. Robotic assisted prostatectomy is widely used for ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 143 localized disease. AI can support surgeons by predicting optimal nerve sparing strategy based on tumor location and potential extracapsular extension. Preoperative models that incorporate MRI data improve surgical margin outcomes and preserve functional results such as continence and sexual function. Bladder cancer surgery requires thorough lymph node dissection and precise decision making around urinary reconstruction. Predictive analytics can estimate the likelihood of node involvement and determine when extended dissection is justified. AI can also model postoperative recovery trajectories to personalize counseling and resource planning. Partial nephrectomy for renal cell carcinoma is technically demanding when tumors are in complex anatomical positions. Algorithms that reconstruct three dimensional kidney vasculature and tumor boundaries can guide approach selection and reduce complications such as ischemia time or renal function loss. Automated evaluation of operative video content provides performance metrics that support surgeon training and continuous improvement. Interventional techniques such as biopsy targeting and ablative therapies also benefit from computational precision. AI directed targeting minimizes sampling error in heterogeneous prostate tumors and ensures that the DR. MEHRDAD FARROKHI 144 most clinically relevant region is evaluated histologically. Thermal ablation planning guided by tumor behavior models can improve local control in kidney tumors deemed inappropriate for surgery. These enhancements demonstrate how AI amplifies surgical expertise, reduces procedural variability, and supports optimal outcomes. AI for Drug Discovery and Adaptive Therapy Developing new therapies remains critical for patients who progress after standard treatments. AI accelerates drug discovery and development by modeling complex biological pathways, identifying drug targets, and prioritizing candidate molecules. Computational simulations can analyze chemical libraries far faster than traditional laboratory screening, dramatically reducing development timelines. In metastatic prostate cancer, resistance evolves through diverse pathways involving androgen receptor signaling, neuroendocrine differentiation, and lineage plasticity. AI modeling can reveal pathways that maintain survival despite therapy, highlighting vulnerabilities for next generation treatment strategies. Drug repurposing algorithms may match existing compounds to emerging resistance phenotypes. Bladder cancer drug discovery benefits from ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 145 malignancies through automated image analysis and risk modeling. Deep learning models trained on vast datasets of dermoscopic and clinical skin lesion images can distinguish melanoma from benign nevi with accuracy comparable to or exceeding experienced dermatologists. These algorithms evaluate subtle color variations, asymmetries, border irregularities, and structural patterns that may signal malignancy but are difficult for the human eye to discern consistently. AI enabled smartphone applications extend this capability into community settings, enabling individuals in remote areas to receive prompt screening support and reduce disparities in access to care. In primary care and telemedicine environments, AI decision support systems help clinicians decide when to refer for biopsy or specialist evaluation. By analyzing medical history, lesion photographs, and symptom descriptions, risk stratification models can identify patients most likely to benefit from expedited diagnostic procedures. Screening challenges extend beyond melanoma. Merkel cell carcinoma often presents as a firm, painless lesion mistaken for a benign cyst. AI models trained with clinical context may help distinguish these lesions earlier. The same applies to dermatofibrosarcoma protuberans, where nodules develop slowly but can become locally aggressive if overlooked. Likewise, Kaposi sarcoma DR. MEHRDAD FARROKHI 158 in immunocompromised individuals may be more easily flagged through image based risk classifiers that contextualize skin changes alongside patient immune status and viral co factors. AI based monitoring tools also show potential for surveillance of individuals with genetic predispositions or histories of UV overexposure. Personalized risk scores updated in real time may identify skin transformations long before symptoms develop. These innovations expand the reach of dermatologic care and offer a pathway toward earlier and more equitable detection. AI in Imaging and Radiomics for Soft Tissue Sarcomas Soft tissue sarcomas present complex imaging and diagnostic challenges due to their diverse origins, variable patterns of growth, and tendency to infiltrate surrounding tissues. MRI and CT scans are primary modalities for preoperative evaluation, but interpretation requires significant expertise and subjective judgment. Radiomics, supported by AI, provides a powerful method to quantify tumor phenotype through high dimensional feature extraction from medical images. Machine learning algorithms applied to MRI can assess tumor margins, vascularity, necrosis, and stromal composition. These characteristics correlate with histologic grade and metastatic ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 159 potential, informing treatment decisions such as whether to pursue radiation, neoadjuvant chemotherapy, or wide surgical margins. AI helps identify specific imaging signatures associated with dedifferentiated liposarcoma, synovial sarcoma, or leiomyosarcoma, which may otherwise be challenging to distinguish without invasive biopsy. Inoperable sarcomas or those undergoing systemic therapy can be monitored more precisely using AI models that detect subtle changes in tumor response before visible shrinkage occurs. Predicting non response early can prevent months of ineffective treatment and allow clinicians to pivot toward more promising strategies. Image guided interventions such as biopsy and localized ablation also benefit from AI enhanced targeting. Algorithms can identify the most biologically active regions of heterogeneous tumors, maximizing diagnostic yield and allowing therapeutic energy delivery where it is needed most. Integration of radiomics with genomics, known as radiogenomics, holds promise in connecting molecular alterations with imaging phenotypes and ultimately supporting a comprehensive understanding of sarcoma biology. AI Enhanced Dermatopathology and Histologic Interpretation Pathology plays a central role in diagnosing DR. MEHRDAD FARROKHI 160 skin and soft tissue cancers, yet manual interpretation is time consuming and prone to interobserver variability. The introduction of whole slide digital imaging and convolutional neural networks enables automated analysis with high reproducibility. AI systems can detect architectural disorganization, mitotic rates, ulceration, tumor infiltrating lymphocyte patterns, and invasion depth in melanoma biopsy specimens. Predictive models link these features to staging and survival outcomes, offering greater precision in selecting appropriate therapy intensity. Algorithms can also flag ambiguous or borderline lesions for more thorough review, reducing the risk of missed melanoma diagnoses. Soft tissue tumor pathology benefits from similar computational advances. Many sarcomas share overlapping microscopic features, making differentiation difficult even for experts. Deep learning models can distinguish complex subtypes by analyzing cell morphology, extracellular matrix relationships, and vascular characteristics. By integrating immunohistochemistry data, AI aids in identifying relevant biomarkers such as PD L1 levels or lineage specific markers that influence treatment selection. Automated quantification of tumor infiltrating lymphocytes and other microenvironmental ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 161 attributes contributes to immunotherapy response prediction. Digital pathology tools can analyze thousands of features simultaneously, revealing patterns that may be invisible through traditional visual review. As access to specialized sarcoma pathology remains uneven worldwide, AI supported platforms offer a scalable way to democratize expert interpretation and improve diagnostic accuracy on a global scale. AI for Genomic Profiling and Precision Oncology Genomic analysis has become increasingly important in guiding therapy for skin and soft tissue cancers, particularly melanoma and sarcomas where mutation driven pathways play significant roles. However, interpreting genomic data is difficult due to the complex interactions among genes, immune influences, and tumor microenvironment factors. AI facilitates the integration of these multidimensional datasets. In melanoma, mutations in BRAF, NRAS, and other signaling pathways inform suitability for targeted therapies such as kinase inhibitors. Machine learning models help identify additional actionable alterations hidden within complex mutation profiles. Predictive systems can estimate resistance likelihood and suggest combination strategies to overcome pathway reactivation. DR. MEHRDAD FARROKHI 162 For metastatic disease, immunotherapy holds significant promise, yet only a portion of melanoma patients respond. AI can analyze transcriptomic and proteomic data to reveal immune signatures associated with durable response. Features such as T cell infiltration, antigen presentation capacity, and interferon signaling activity can be quantified and combined into predictive scores that inform clinical decisions. Soft tissue sarcomas exhibit greater molecular diversity, often requiring broad panel sequencing. Many lack well defined therapeutic targets, but AI methods can identify molecular clusters correlated with therapeutic vulnerabilities. For instance, specific fusion proteins in synovial sarcoma or Ewing sarcoma may predict sensitivity to targeted or epigenetic therapies. Deep learning tools can support discovery of novel biomarkers that recognize shared dependencies across sarcoma subtypes. AI enabled drug matching systems evaluate the compatibility of available and experimental therapies with individual molecular profiles. This personalized strategy increases the likelihood of meaningful benefit in cancers known for heterogeneous and unpredictable responses. AI Guided Surgical and Reconstructive Planning ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 163 Surgery is a cornerstone of treatment for many skin and soft tissue cancers, especially melanoma with localized disease or sarcomas that are amenable to wide resection. Achieving negative margins is essential yet challenging when tumors infiltrate complex anatomic regions. AI based planning systems use imaging and predictive models to support surgeons in achieving optimal oncologic and functional outcomes. Three dimensional reconstructions generated from MRI and CT images provide detailed maps of tumor boundaries and their relationships to muscles, nerves, and bones. Machine learning can analyze risk of local recurrence based on planned margin width and help select appropriate resection strategies. For sarcomas adjacent to major neurovascular structures, AI simulations predict outcomes associated with limb preserving surgery versus amputation, enabling informed shared decision making. Melanoma patients with sentinel lymph node involvement often undergo lymphadenectomy. AI tools can analyze lymphatic drainage imaging to identify specific nodal basins most at risk of harboring disease. This precision can reduce unnecessary surgical morbidity. Reconstructive planning benefits from predictive modeling of wound healing and tissue viability. Deep learning algorithms can determine which flap strategies will provide best function and DR. MEHRDAD FARROKHI 164 cosmetic outcome while minimizing complication risks. Intraoperative video analysis supported by AI can guide surgeons in maintaining clear margins and assessing soft tissue planes in real time. Robotic systems enhanced with intelligent navigation have potential to improve mechanical precision and expand the complexity of surgeries that can be performed minimally invasively. AI in Radiation Therapy and Response Prediction Radiotherapy is commonly used to treat high risk melanoma after surgery and for many soft tissue sarcomas either before or after resection. AI optimizes treatment planning by improving target delineation and predicting normal tissue sensitivity. Automated segmentation tools decrease planning time and reduce variability introduced by manual contouring. Machine learning algorithms model dose response relationships using extensive clinical data. These systems inform decisions about dose intensification when attempting to achieve local control in locally advanced disease or reduced dosing when patient specific toxicity risks are high. Predictive models identify patients likely to benefit from hypofractionated regimens, stereotactic radiotherapy, or proton therapy. For sarcoma, where radiation fields often involve ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 165 multiple critical structures, AI enhanced planning reduces exposure to surrounding organs. Tumor motion prediction supports highly conformal treatment while accommodating respiratory or musculoskeletal shifts. Monitoring response during radiotherapy is crucial for adaptive planning. AI can detect radiographic changes indicating tumor shrinkage or resistance early in treatment. This approach allows radiation oncologists to adjust plans dynamically to maximize therapeutic efficiency. Emerging research explores integrating radiomics, tumor genomics, and microenvironment markers into comprehensive models that forecast long term local control and patient outcomes. AI for Drug Discovery and Novel Therapeutic Development Developing effective systemic therapies for advanced skin and soft tissue tumors remains a significant unmet need. Many sarcomas lack established targets, and aggressive melanoma can become resistant to available agents. Artificial intelligence accelerates drug discovery by analyzing genetic networks, predicting structure activity relationships, and identifying synergistic drug combinations. Deep learning models screen compound libraries for molecules likely to bind key oncogenic proteins or restore function to mutated pathways. DR. MEHRDAD FARROKHI 166 These approaches can rapidly evaluate millions of chemical structures in silico, narrowing the number needing laboratory validation. AI identifies drug resistance pathways in melanoma by modeling how cells adapt to BRAF and MEK inhibitors. This helps researchers design combination regimens that preemptively block escape routes. Repurposing strategies also benefit from pattern recognition across clinical trial databases, revealing unexpected anticancer properties of approved drugs. In sarcoma drug discovery, where patient numbers are small and resources limited, AI helps prioritize targets most likely to translate effectively into therapy. Simulations of tumor evolution support the concept of adaptive therapy, where drug dosing changes dynamically to prevent resistant population dominance. These methods bring hope for expanding the therapeutic arsenal against cancers that have historically offered few options at advanced stages. AI Based Real Time Monitoring and Survivorship Care For many patients, the journey through skin or soft tissue cancer does not end after initial treatment. Surveillance for recurrence requires repeated imaging, physical exams, and sometimes invasive procedures. AI enhances monitoring ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 167 efforts by identifying subtle indications of relapse sooner than traditional approaches. Melanoma survivors depend heavily on self examination and dermatologic follow up. AI assisted mobile applications can track lesion evolution over time and alert patients or clinicians to concerning changes. Patterns of recurrence risk can adjust visit frequency to match individual risk. Soft tissue sarcoma recurrence is often detected through shear volume growth on imaging. Radiomic analysis powered by AI enables earlier detection of aggressive changes and supports timely salvage therapy. Treatment related complications such as lymphedema, neuropathy, or radiation induced fibrosis can be monitored using wearable sensors that continuously assess functional mobility and extremity changes. AI translates these signals into actionable insights that prompt early intervention, improving patient outcomes. Quality of life fluctuates during survivorship. Natural language processing applied to patient journals or digital communications provides clinicians with insights into distress, fatigue, anxiety, and social difficulties. Personalized coaching and supportive care suggestions can be automatically delivered through AI platforms to meet dynamic needs. DR. MEHRDAD FARROKHI 168 By empowering proactive health management, AI supports not only survival but also long term wellness and recovery. Challenges and Ethical Considerations Even with remarkable potential, the integration of artificial intelligence into care for skin and soft tissue cancers brings significant challenges. High performing AI requires large, well curated datasets representative of diverse populations. Many rare skin cancers and most sarcoma subtypes suffer from limited data availability. Small datasets can lead to overfitting and poor real world performance. Bias in training data may impair model accuracy in populations underrepresented in clinical trials or dermatologic image libraries, such as individuals with darker skin tones. This disparity poses a risk of perpetuating or worsening existing healthcare inequalities. Ongoing efforts must ensure that datasets reflect global diversity. Interpretability remains essential. Clinicians need to understand how an algorithm reaches a conclusion, especially when guiding high stakes decisions like immunotherapy escalation or wide margin surgery. Black box models can erode trust and obscure errors that require human oversight. Regulatory pathways for continuously learning AI systems are still evolving. Safety monitoring must ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 169 be robust, with mechanisms for accountability when AI recommendations contribute to negative outcomes. Privacy concerns arise when sensitive patient information including images of identifiable skin areas or genomic data is shared for model development. Ethical use requires secure data handling and clear patient consent processes. Successful implementation demands collaboration between clinicians, data scientists, technologists, regulators, and patient advocates to ensure that innovation aligns with patient centered values and clinical realities. Future Directions The future of AI in managing hard to treat skin and soft tissue cancers promises expanded capabilities and deeper integration into personalized care. Multimodal AI systems will combine dermoscopic images, radiologic features, pathology data, and molecular signatures into cohesive models that capture tumor biology comprehensively. Digital twin models representing unique patient characteristics may simulate treatment responses before clinical implementation. Robotic systems with intelligent vision will refine surgical precision, while augmented reality overlays assist in visualizing hidden tumor margins. AI guided immunotherapy optimization DR. MEHRDAD FARROKHI 170 is expected to grow as models gain deeper understanding of the tumor microenvironment and immune evasion mechanisms. Teledermatology and remote oncology care will expand, making expert level guidance accessible globally. These tools may form the backbone of preventive care, helping detect malignancies before they pose life threatening risks. Research collaboration networks powered by AI will speed discovery by enabling real time analysis of clinical outcomes across institutions. More effective therapies will likely emerge as AI accelerates molecular research and adaptive clinical trials. Despite these advances, maintaining patient trust will be essential. Ensuring transparency, preserving clinician judgment, and anchoring innovation in ethical frameworks will determine whether AI truly transforms outcomes for patients facing aggressive skin and soft tissue cancers. Conclusion Hard to treat skin and soft tissue cancers remain among the most challenging diseases in oncology due to biological aggression, limited treatment response, and risk of recurrence. Artificial intelligence offers new tools that reimagine each phase of cancer care. From early screening and improved diagnostic accuracy to precision ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 171 radiotherapy, personalized drug selection, enhanced surgery, and ongoing surveillance, AI brings detailed analysis and predictive power to support clinicians and empower patients. While barriers exist related to data access, fairness, interpretability, regulation, and privacy, the trajectory of innovation is clear. Responsible adoption of AI has the potential to elevate clinical care and provide hope to individuals who historically have faced limited options. By fostering collaborative development and ensuring equitable design, AI may play a defining role in shifting outcomes and quality of life for patients with hard to treat skin and soft tissue malignancies. Continual progress will bring us closer to a future where early detection is routine, treatments are precise, and survivorship is more secure for all affected individuals. DR. MEHRDAD FARROKHI 172 ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 173 7. AI FOR HARD-TO-TREAT HEAD AND NECK CANCERS Background Hard to treat head and neck cancers represent a group of malignancies originating in the oral cavity, pharynx, larynx, sinonasal region, salivary glands, and other supporting structures. Many are diagnosed late, present with aggressive biological behavior, and require complex treatment strategies involving combinations of surgery, radiation, and systemic therapy. Even when treated with curative intent, patients frequently experience functional impairments involving speech, swallowing, appearance, and respiration. The survival outcomes have improved modestly over time, but advanced or recurrent disease still leads to a high mortality rate. A major challenge in this field lies in disease heterogeneity. Head and neck squamous cell carcinoma alone encompasses numerous anatomic locations, risk factors, and molecular subtypes that respond differently to therapy. Human papillomavirus associated oropharyngeal cancer demonstrates a more favorable prognosis and different therapeutic sensitivities compared to cancers caused by tobacco and alcohol exposure. Rare tumors of the salivary glands or sinonasal 174 tract often have unique molecular drivers but suffer from limited research investment and few options for targeted treatments. Precision oncology in this space is advancing but still remains inconsistently applied. Traditional diagnostic and treatment pathways depend heavily on expert interpretation of imaging, pathology, and clinical examination. These tasks are labor intensive and subject to significant variability, particularly in resource limited settings. There is an urgent need for tools that enhance early detection, improve risk stratification, guide therapy personalization, and support functional recovery following treatment. Artificial intelligence has emerged as a transformative solution capable of filling these gaps. AI technologies leverage machine learning and deep learning to analyze complex clinical data, recognize patterns beyond human perception, and generate real time decision support. Applications span a wide spectrum: early cancer screening using voice or image analysis, automated evaluation of radiology or digital pathology, multimodal prognosis prediction, radiotherapy optimization, drug discovery, and long term survivorship support. By integrating diverse data sources, AI offers a comprehensive approach to understanding tumor behavior and improving patient outcomes. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 175 This chapter explores the current and emerging roles of artificial intelligence in addressing the many clinical challenges associated with advanced and hard to treat head and neck cancers, as well as the limitations, ethical considerations, and future directions shaping the integration of these technologies into routine care. AI in Early Detection and Screening Many head and neck cancers are diagnosed only once they have caused visible or functional symptoms, such as persistent voice changes, dysphagia, or neck masses. Earlier detection dramatically improves survival, yet reliable screening methods remain limited. Artificial intelligence offers innovative solutions in identifying early stage disease where conventional diagnostics may fail. Computer vision algorithms trained on clinical imagery have improved recognition of suspicious oral lesions. These programs analyze subtle textural or vascular changes on mucosal surfaces that may indicate malignant transformation. Mobile AI tools can extend screening capabilities to primary care or community health environments, enabling early risk identification without requiring immediate access to specialty providers. Voice analysis is another area of rapid growth. Machine learning models can detect acoustic DR. MEHRDAD FARROKHI 176 changes associated with early laryngeal cancer before visible lesions appear during laryngoscopy. By incorporating patient voice recordings into automated risk scoring systems, these tools serve as non invasive and low cost methods for ongoing monitoring of high risk individuals such as professional voice users or individuals exposed to tobacco smoke. Artificial intelligence is enhancing radiology based screening as well. CT and MRI scans performed for unrelated reasons sometimes reveal incidental abnormalities in the upper airway or neck. AI models that scan radiology reports and images can flag concerning changes for secondary review, reducing missed opportunities for intervention. In populations with human papillomavirus risk, models combining demographic data, sexual behavior patterns, and serologic markers can help determine who may benefit most from targeted screening initiatives. As immune profiling technologies advance, AI will further support risk stratification based on the biological predisposition toward viral driven carcinogenesis. AI systems that integrate multiple signals, including symptom reports, social determinants of health, and clinical history, likely represent the most powerful future approach. By identifying early stage disease and minimizing diagnostic delay, AI driven screening may significantly ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 177 optimizing interventions, predicting outcomes, and supporting survivorship throughout recovery. As AI continues to evolve, the most successful implementations will be those that maintain clinician involvement, protect patient privacy, ensure equity, and prioritize meaningful improvements in quality of life. Through careful and collaborative innovation, AI has the potential to shift the paradigm of head and neck cancer care toward earlier cures, longer survival, and greater functional preservation. By embracing these capabilities responsibly, the field moves closer to a future in which patients facing these difficult cancers can receive timely, tailored, and life enhancing care. DR. MEHRDAD FARROKHI 190 ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 191 8. AI FOR HARD-TO-TREAT ORAL CANCERS Background Oral cancers remain a significant global health challenge due to their aggressive biological behavior, the difficulties associated with early detection, and the functional burden imposed by curative therapies. Classified primarily as squamous cell carcinomas arising from the mucosal surfaces of the mouth, these malignancies can also involve the lip vermilion, tongue musculature, gingiva, buccal mucosa, palate, and floor of mouth. Although they are grouped within the wider spectrum of head and neck cancers, oral cancers have unique features that set them apart, including distinct etiologic patterns, differences in genetic drivers, and specific implications for speech, chewing, swallowing, and appearance. The disease often progresses rapidly and may infiltrate deep tissue planes early in its development, spreading to regional lymph nodes or metastasizing systemically before a diagnosis is confirmed. These tumors are frequently diagnosed at advanced stages when treatment becomes more complex, resource intensive, and less likely to achieve long term control. 192 A variety of modifiable and non modifiable risk factors contribute to the development of oral cancers. Tobacco and alcohol use remain the most prominent environmental drivers in many regions, particularly when these behaviors coexist and amplify malignant transformation risk. Betel nut chewing contributes substantially to incidence in South Asia and the Western Pacific. Chronic irritants, poor oral hygiene, microbial dysbiosis, and genetic predispositions also play roles in carcinogenesis. Unlike in oropharyngeal cancers, human papillomavirus involvement is generally much lower, which means immunotherapy strategies that have worked effectively elsewhere in the upper aerodigestive tract may be less successful in the oral cavity without identification of more tailored biomarkers. The global distribution of this disease reveals considerable disparities. Low and middle income countries bear a disproportionate share of incidence and mortality. Socioeconomic factors influence the ability to access dental services, receive routine screenings, and secure specialized cancer care. These barriers lead to late diagnosis in many patients, lowering survival outcomes and increasing the risk of extensive surgical interventions followed by debilitating functional impairment. Even in high income settings where state of the art treatment is accessible, outcomes ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 193 remain far from ideal for individuals with locally advanced or recurrent disease. Traditional oral cancer care relies heavily on human expertise in visual examination, clinical judgment, imaging interpretation, pathology evaluation, and coordination of complex treatment decisions. Yet each of these steps carries inherent limitations in speed, consistency, and sensitivity. Subtle early lesions escape attention because they appear similar to benign ulcers or inflammatory conditions. Imaging may confirm deeper infiltration only after substantial structural damage has occurred. Histopathology can be subject to interpretive variability, particularly in cases involving borderline dysplasia or small biopsy samples that do not fully represent tumor heterogeneity. The time required for multidisciplinary evaluation may delay initiation of therapy, allowing disease to progress. Artificial intelligence represents a powerful approach to overcoming these limitations. With its ability to recognize patterns within large and complex datasets at speeds impossible for manual evaluation, AI can enhance every stage of the care continuum. It expands access to screening by supporting frontline clinicians and even patients directly through portable devices. It improves diagnostic accuracy by analyzing multimodal data from photographs, imaging DR. MEHRDAD FARROKHI 194 scans, and histologic slides. It contributes to personalized treatment decisions by integrating genomic, immunologic, and clinical variables into predictive models. It augments surgical planning, radiotherapy delivery, rehabilitation, and survivorship monitoring through real time guidance and predictive analytics. These capabilities align closely with the urgent needs present in oral oncology. This chapter provides an in depth exploration of how AI technologies are reshaping the management of hard to treat oral cancers. It considers opportunities in early detection, diagnosis, staging, surgical and radiologic therapy, systemic treatment personalization, and rehabilitation. It also acknowledges the ethical, regulatory, and practical barriers that must be addressed to ensure equitable implementation. As AI continues to evolve, its role within oral cancer care is likely to expand dramatically, redefining both outcomes and patient experience. AI in Early Detection and Screening Early detection represents the single most impactful way to improve survival in oral cancer. The difference in outcome between early stage and late stage disease is profound. However, subtle mucosal lesions are often invisible to untrained observers and can be misdiagnosed even by experienced clinicians. Artificial intelligence ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 195 has introduced new approaches to overcoming these diagnostic obstacles by supporting lesion identification, risk stratification, and triaging of patients into appropriate care pathways. The rapid expansion of computer vision techniques has enabled deep neural networks to evaluate intraoral images with impressive accuracy. By analyzing characteristics such as color progression, keratinization patterns, architectural distortion, and vascular changes, AI can identify abnormalities that warrant further assessment. Unlike conventional screening methods requiring specialized equipment, AI enabled mobile phone applications can be used at the point of care in dental clinics or community health settings, making screening more accessible. Telemedicine has also found a strengthened role through AI. Dental practitioners or general physicians can photograph a suspicious lesion and receive reliable automated insights that inform whether a referral to oncology is appropriate. Patients in rural or remote communities can be evaluated rapidly without delaying care, helping to close geographic disparities in diagnosis. Some systems incorporate longitudinal lesion monitoring into their evaluation. When a lesion persists or evolves in concerning ways over time, AI models detect subtle changes that human observers may overlook. In patients with established oral potentially malignant disorders DR. MEHRDAD FARROKHI 196 such as leukoplakia or lichen planus, continuous tracking reduces the risk of late recognition of malignant transformation. Beyond image analysis, AI can support risk prediction by integrating data on patient demographics, lifestyle habits, past dental records, and medical comorbidities. For example, a model might weigh a history of tobacco exposure and heavy alcohol intake alongside lesion morphology to produce a personalized malignancy likelihood estimate. These innovations strengthen the first line of defense against oral cancer progression. By improving early recognition and more timely specialist involvement, AI powered screening tools have the potential to shift diagnosis toward earlier stages when treatment is less destructive and survival prospects are significantly higher. AI in Diagnostic Imaging and Tumor Characterization Accurate assessment of tumor extent is essential for determining the best course of treatment. Oral cancers can invade anatomical structures that are critical for essential functions. Imaging evaluation is therefore key to planning interventions that balance oncologic control with preservation of quality of life. Yet interpretation of CT, MRI, and PET scans can be challenging because the oral cavity contains a dense network of ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 197 muscles, nerves, and bones packed into a small space, creating imaging artifacts and overlapping densities. Artificial intelligence enhances imaging interpretation through automated segmentation, radiomic analysis, and prediction modeling. Rather than relying solely on human visual expertise, AI extracts quantitative information from each pixel, identifying features associated with depth of invasion, nodal spread, and early mandibular involvement. Radiomics helps uncover correlations between imaging features and aggressiveness, enabling risk stratification beyond conventional tumor size and shape criteria. The integration of multimodal imaging data further improves precision. For example, MRI derived soft tissue contrast can complement PET metabolic activity maps, while AI models synthesize information from both techniques into a unified interpretation that provides a more holistic understanding of tumor biology. These approaches reduce diagnostic uncertainty and assist clinicians in determining whether surgical, radiotherapeutic, or combined approaches will yield the best outcome. Another significant benefit lies in monitoring treatment response. After radiation or systemic therapy, inflammation and fibrosis may mimic persistent cancer on imaging. AI pattern DR. MEHRDAD FARROKHI 198 recognition allows earlier and more accurate distinction between active disease and healing changes. This reduces unnecessary biopsies or surgical exploration and helps identify cases requiring prompt treatment modification. In high complexity regions such as the base of tongue or deep floor of mouth, three dimensional reconstructions created by AI systems support surgeons in evaluating how a tumor encroaches upon critical structures. This assists in planning surgical margins that achieve both complete tumor removal and functional preservation. These imaging innovations transform radiology into a predictive and dynamic component of oral cancer care. AI Enhanced Digital Pathology Pathologic confirmation remains indispensable in the diagnosis of oral cancers. However, tumor heterogeneity and subjective interpretation contribute to variability in grading, staging, and margin assessment. Transitioning from manual microscopy to digital whole slide imaging creates a platform where artificial intelligence can significantly enhance pathology workflows. Deep learning systems analyze cellular morphology and tissue architecture at a granular level. They evaluate criteria such as nuclear enlargement, pleomorphism, mitotic activity, keratin pearl formation, and immune infiltrate ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 199 distribution, detecting microscopic cues that predict behavior more reliably than qualitative evaluation alone. Such models standardize grading decisions, reducing discrepancies between pathologists and ensuring greater confidence in diagnosis. Margin assessment in surgical specimens benefits tremendously from computational pathology. Manual review may miss isolated tumor cells, particularly when evaluating large resections that require many slide sections. AI rapidly screens entire specimens, highlighting regions requiring targeted human review. This assists surgeons in achieving complete tumor removal and lowers the likelihood of needing re operation. Predictive pathology extends beyond diagnosis into prognosis. Machine learning models trained on outcome data can estimate expected survival or recurrence probability by integrating histologic features with genetic and clinical context. This informs decisions about adjuvant therapy and follow up intensity. Digital pathology also supports resource limited settings. AI enabled interpretation can supplement areas where experienced pathologists are scarce, widening access to expert level evaluation through telepathology networks. Such technology democratizes diagnostic accuracy globally and plays a vital role in reducing international disparities in oral cancer care. DR. MEHRDAD FARROKHI 200 AI Guided Personalized Treatment Decision Making A defining challenge in oral oncology involves tailoring treatment approaches to individual patient needs. Surgery may remove visible tumor, yet if disease is biologically aggressive, systemic therapy or radiation may be warranted to prevent recurrence. Conversely, overtreatment must be avoided to preserve form and function when disease is less threatening. Striking this balance manually can be difficult because clinical presentations are nuanced and response variability is substantial. Artificial intelligence guides treatment planning by integrating numerous prognostic indicators. Predictive models assess clinical staging, radiomic signatures, histopathology findings, genomic alterations, immune environment features, and patient health status to estimate which therapies will offer the greatest benefit. For example, AI may identify a subset of patients in whom chemoradiation alone will achieve cure without the need for extensive surgery, preserving critical functions such as speech and swallowing. Systemic therapy selection benefits from AI analysis of molecular characteristics. Although immunotherapy response has been moderate in oral cancer compared to other head and neck sites, AI based biomarker evaluation can define ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 201 which specific patients have a tumor immune landscape compatible with durable responses. Decision support tools allow clinicians to simulate how treatments may perform based on real world outcomes drawn from multiple institutions and patient populations. AI interpretation of toxicity risk enhances shared decision making. For instance, a model might indicate that a particular radiation plan poses a substantial chance of long term xerostomia or osteoradionecrosis in a patient with preexisting dental vulnerabilities. Understanding risks in advance allows for adaptation of modality or intensity before initiating therapy. Personalized oncology ideally combines outcome prediction with preservation of quality of life, and AI is instrumental in enabling that evolution. AI in Surgery and Reconstructive Planning Surgery remains central to the management of many oral cancers, especially when diagnosis occurs at a stage where tumor removal offers the only realistic chance of cure. Yet surgical intervention must minimize post operative dysfunction. The tongue, mandible, and other structures are essential for articulation, mastication, and swallowing, meaning even small resections can have significant impacts. AI contributes meaningfully to preoperative DR. MEHRDAD FARROKHI 202 planning by analyzing the anticipated results of various surgical strategies. Three dimensional models show how tumor excision will alter anatomy and allow surgical teams to anticipate consequences for mobility and airway protection. Predictive tools estimate postoperative speech intelligibility and swallowing ability, guiding discussions about reconstruction options that preserve these functions. For cases involving mandibular involvement, AI evaluation of cortical and medullary bone invasion helps determine whether a segmental resection is required. The decision to remove bone has enormous implications for prosthodontic rehabilitation and facial symmetry. Using AI insights, surgeons can optimize surgical margins while preserving structural integrity when safe. Intraoperatively, real time AI assistance improves precision. Image guided navigation systems enhanced by machine learning differentiate tumor from healthy tissue based on optical or fluorescence cues. This reduces both under treatment and overtreatment. Robotic assistance guided by AI helps maintain stable instrument positioning and reduces fatigue associated with long and technically demanding procedures. Reconstructive planning benefits from predictive modeling that assesses tissue viability, healing probability, and functional restoration. Whether bone flaps or soft tissue flaps are used, AI ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 203 helps match reconstruction type to the specific functional priorities of the patient. Through these contributions, AI fosters outcomes that preserve dignity, identity, and the essential elements of communication. AI Enhanced Radiotherapy Radiation plays a major role in oral cancer care as either a definitive approach for unresectable tumors or as a postoperative treatment to reduce recurrence. However, the close proximity of numerous sensitive anatomical structures challenges the safe delivery of therapeutic doses. Side effects can severely impact quality of life for survivors, sometimes resulting in chronic disability. AI improves radiotherapy planning and execution by analyzing anatomical and functional relationships between tumor and organs at risk. Automated segmentation tools reduce planning time and minimize variation between planners. Machine learning algorithms evaluate dose distributions to find optimal strategies that protect salivary glands, taste buds, dental arches, and swallowing musculature. Predictive toxicity modeling allows high risk individuals to receive tailored support even before symptoms emerge. When integrated with imaging during treatment, adaptive radiotherapy techniques utilize AI to modify delivery plans if DR. MEHRDAD FARROKHI 204 the tumor shrinks or patient anatomy changes. This maintains precision and reduces collateral tissue damage. Together, these advancements strengthen the therapeutic ratio, enhancing control while limiting harm. AI in Systemic Therapy and Immunotherapy Response Prediction Oral cancers with advanced local spread or distant metastasis often require systemic therapy to complement local control measures. Yet treatment success remains variable. The complex biology of oral cancer contributes to resistance, and trial and error approaches can expose patients to unnecessary toxicity without benefit. Artificial intelligence models support therapy selection by predicting tumor susceptibility to chemotherapy, targeted agents, or immunotherapy based on genomic signatures and microenvironmental patterns. When a tumor displays molecular features linked to aggressive progression, AI may recommend more intensive multimodal therapy for a greater chance of long term survival. Immunotherapy represents a major area of exploration. Response prediction remains challenging due to mixed involvement of immune pathways in oral carcinoma compared to other ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 205 sites. AI aids in identifying which patients express markers associated with durable benefit, ensuring that checkpoint inhibition is reserved for those most likely to respond effectively. AI tools also track treatment response through liquid biopsies, radiomics, or symptom monitoring to detect signs of non response early. Adjusting therapy based on evolving clinical signals represents a shift toward proactive rather than reactive management. AI Supported Rehabilitation and Survivorship Cancer survival means little if survivors are left with severe functional deficits, chronic pain, and diminished social participation. The oral cavity plays central roles in nourishment, communication, identity expression, and social interaction. Post treatment rehabilitation must consider each patient’s goals and challenges. AI supports rehabilitation by analyzing speech and swallowing performance data, interpreting subtle limitations in lingual motion or pharyngeal coordination. Personalized exercise programs can be generated automatically and adjusted in real time based on measurable improvements or setbacks. Integrating this technology into home based care ensures continuous support between clinic visits. Wearable sensors help track nutritional intake, DR. MEHRDAD FARROKHI 206 weight change, and activity levels. AI identifies trends associated with decline and alerts providers to intervene early. Mental health support is another important dimension. Natural language analysis of patient communication can identify emotional distress that may not surface in brief clinical encounters. Targeted counseling can be offered promptly to those at risk of depression or social isolation. Advanced prosthodontic planning uses AI to anticipate the functional advantages of various inserts or implants following surgical alteration of oral anatomy. These innovations restore confidence and practical ability for eating and speaking. Survivorship improved by AI emphasizes autonomy, dignity, and long term health, supporting individuals well beyond completion of primary treatment. Ethical and Operational Considerations The promise of artificial intelligence brings with it important responsibilities. Data used to train models must reflect the populations most affected by oral cancer, which often include underserved communities where access to high quality care is limited. Without careful dataset curation, AI may inadvertently widen disparities by delivering lower quality recommendations to those whose ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 207 characteristics are underrepresented in training material. Protecting patient privacy is also paramount because oral images and voice data are uniquely identifiable. Robust safeguards are necessary to maintain trust and comply with regional regulations governing health data. Clinicians must understand how AI arrives at a given recommendation so that they can evaluate its appropriateness and explain it to their patients. Transparent design and external validation processes build confidence that AI operates reliably and safely. Successful integration into clinical practice requires thoughtful workflow design so that the technology reduces burden rather than increasing it. Stakeholder collaboration across dentistry, oncology, radiology, pathology, engineering, and patient advocacy will be necessary to develop solutions that meet practical needs rather than simply fulfilling a technological vision. Future Directions As AI continues to mature, multimodal intelligence integrating clinical, pathological, radiological, genomic, and behavioral data will become foundational to oral cancer care. Digital twins that model individual patients could allow clinicians to simulate the outcomes of varied treatment strategies before committing to a DR. MEHRDAD FARROKHI 208 chosen plan. Real time monitoring throughout therapy may lead to fully adaptive care that optimizes outcomes as situations evolve. The extension of AI enhanced mobile health tools into widespread public use may reduce delays that currently drive poor outcomes by empowering individuals to recognize harmful changes early. Globally connected research networks will allow shared learning and continuous improvement of models, bringing advancements to every corner of the world where oral cancer persists. Conclusion Oral cancers continue to impose a devastating burden on individuals and healthcare systems, particularly when detected late or when the disease displays aggressive biological behavior. The complexity of the oral cavity, the challenges of early recognition, and the balance required between oncologic control and preservation of essential functions make traditional approaches difficult to optimize consistently. Artificial intelligence presents transformative opportunities to address these difficulties. It enhances early detection, improves diagnostic accuracy, supports precision therapy selection, strengthens surgical and radiotherapy planning, accelerates the discovery of systemic treatments, and provides personalized rehabilitation support for survivors. Responsible implementation grounded in equity, transparency, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 209 and function. When pediatric patients are treated, models assess growth plate sensitivity to reduce the risk of stunting or deformity. Adaptation during treatment is achieved through AI systems that examine changes in tumor volume or body alignment across sessions. If the target shrinks significantly, plans can be updated to maintain accuracy and limit collateral damage. Predictive models also estimate risk of long term complications such as fracture or fibrosis, enabling early preventive strategies. Through improved precision and adaptability, AI enhanced radiotherapy contributes to superior clinical outcomes and reduced disability. AI Accelerated Drug Discovery and Personalized Systemic Therapy Systemic therapy for bone and musculoskeletal cancers remains an area of unmet need because many tumors develop resistance to current agents and progress despite intensive treatment. Artificial intelligence accelerates discovery of new drug candidates by modeling interactions at the molecular level and predicting compound performance before human trials. AI driven virtual drug screening reduces research time substantially, allowing millions of molecules to be evaluated computationally in the time that traditional methods evaluate only a small fraction. These models identify agents that are more likely DR. MEHRDAD FARROKHI 222 to overcome resistance mechanisms by targeting pathways implicated in metastatic progression and immune evasion. Drug repurposing also benefits greatly. AI finds similarities between musculoskeletal tumor genomic landscapes and diseases treated by existing drugs, potentially offering rapid therapeutic alternatives. For example, if a kinase inhibitor used for another cancer targets pathways active in certain sarcomas, AI might flag this agent for further investigation. Personalized therapy selection is supported through predictive modeling that integrates genomics, immune composition, and response history. AI analysis of circulating tumor DNA during therapy helps identify early resistance and guides changes that preserve treatment momentum. Altogether, AI contributes to a more dynamic and personalized approach to systemic therapy. AI Supported Rehabilitation and Survivorship Survivorship for patients with bone and soft tissue cancers involves significant rehabilitation due to functional losses associated with tumor location and treatment. Limb salvage procedures often require lengthy recovery periods and physical therapy to regain stability and mobility. Amputation survivors rely on prosthetics that ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 223 must be tailored carefully to optimize comfort and performance. AI helps personalize rehabilitation by assessing gait patterns, range of motion, and force distribution during movement. Wearable sensors transmit data to learning algorithms that identify subtle inefficiencies or increasing strain, prompting targeted exercises. Real time monitoring allows continuous progress evaluation and enables early response when recovery slows. Pain and emotional trauma can persist long after treatment completion. Natural language processing applied to patient communication highlights distress and helps clinical teams intervene with timely psychosocial support. Muscle and bone integrity predictions inform activity recommendations and protect against stress fractures or joint deterioration. Advances in prosthetic robotics paired with AI enable more natural limb control and adaptation to varied terrains or tasks. The integration of neural and mechanical signals through machine learning expands possibilities for independent living. Long term survivorship supported by AI emphasizes ability rather than disability and gives survivors greater freedom to pursue fulfilling life goals. DR. MEHRDAD FARROKHI 224 Ethical and Implementation Considerations Artificial intelligence promises tremendous benefit, but these innovations raise important responsibilities. Data quality and diversity must be prioritized because musculoskeletal cancers disproportionately affect younger populations whose data may be scattered across institutions or incomplete. Underrepresentation of certain demographics risks biased models that fail to serve every patient equally. Clinician familiarity and trust are essential. AI systems must be interpretable enough that healthcare providers understand and confidently integrate recommendations. Patients should be able to ask questions and receive clear explanations that affirm their autonomy and support shared decision making. Privacy and data protection are particularly important because imaging and rehabilitation data can reveal identifiable information. Regulatory oversight must ensure that AI development and deployment are carried out under strict ethical frameworks and validated rigorously before clinical implementation. Most importantly, technological advances must benefit all patients, not only those treated at elite academic centers. Accessibility and broad integration into diverse healthcare environments ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 225 remain crucial goals. Future Directions The future of AI in bone and musculoskeletal cancer care involves tighter integration across specialties. Digital twins that create virtual patient models may allow clinicians to simulate treatment outcomes before choosing a plan. Multimodal AI systems will link radiologic imaging with pathology and molecular modeling to produce up to the minute adaptive care strategies. Robotics will expand surgical possibilities with increased safety and precision. Virtual rehabilitation programs will become more sophisticated, giving patients continuous support in their homes. Global collaboration will expand knowledge sharing and improve rare cancer data availability, strengthening predictive accuracy and broad access to innovation. AI will increasingly support public health initiatives by recognizing environmental exposures and emerging risk trends. The ethical imperative will be to harness these advancements responsibly and inclusively. Conclusion Bone and musculoskeletal cancers impose severe burdens on patients through pain, disability, and uncertainty. Traditional diagnostics and treatments, while essential, do not always provide DR. MEHRDAD FARROKHI 226 timely insight or preserve long term function. Artificial intelligence offers new pathways to reduce diagnostic delays, improve surgical and radiologic precision, accelerate therapy development, and support survivors throughout recovery. When deployed thoughtfully within multidisciplinary care, AI can transform outcomes and quality of life for individuals facing these difficult cancers. Commitment to ethical implementation, equitable distribution, interpretability, and clinician partnership will determine how fully this technology fulfills its potential in redefining care for bone and musculoskeletal cancers. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 227 DR. MEHRDAD FARROKHI 228 10. AI FOR THE DIAGNOSIS AND TREATMENT OF UNKNOWNORIGIN CANCERS Background In oncology, identifying the tissue of origin of a malignancy is fundamental to selecting effective therapy. Yet a subset of patients presents with metastatic cancer lacking any detectable primary site, classified as cancers of unknown primary, or CUP. CUP accounts for approximately 2 to 5 percent of all malignancies and is consistently associated with poor outcomes. Without an identified primary, patients are deprived of site specific treatment options and face significantly worse prognoses. Conventional empirical therapies demonstrate limited efficacy in this setting, which underscores the urgent need for new diagnostic approaches to guide management. Recently, artificial intelligence has emerged as a promising tool, with machine learning models capable of predicting tumor origins with high accuracy and providing additional support to clinicians. CUP represents a relatively rare but clinically significant entity that contributes disproportionately to cancer related mortality despite its relatively low incidence. Although 229 incidence has gradually declined with advances in imaging and pathology, CUP remains a persistent diagnostic and therapeutic challenge. Historically, patients underwent exhaustive investigations and, when no primary was found, were treated with empiric broad spectrum chemotherapy. Over time, classification systems stratified CUP cases into prognostic subsets. Around 15 to 20 percent of patients present with clinicopathological patterns suggestive of a specific primary, often referred to as a favorable subset, and may benefit from site directed therapy, whereas the remaining 80 to 85 percent form an unfavorable subset with poor responses to empiric regimens and markedly worse outcomes. Biologically, CUP is highly heterogeneous and encompasses diverse tumor lineages and genomic profiles that reflect multiple potential tissues of origin and complex evolutionary pathways. Persistent diagnostic challenges stem from the limitations of conventional modalities. Even with advanced imaging techniques and comprehensive histopathological evaluation, CUP often remains a diagnosis of exclusion. Immunohistochemistry can identify a presumptive origin in only a proportion of cases and is particularly unreliable for poorly differentiated tumors, where staining patterns are ambiguous or non specific. Molecular classifiers based on gene expression, methylation profiling, or sequencing DR. MEHRDAD FARROKHI 230 show promise in predicting tissue of origin but lack fully standardized validation and have yet to consistently demonstrate survival benefits in large randomized trials. These limitations often lead to reliance on empiric chemotherapy, leaving most high risk patients with a median survival of only 6 to 10 months and restricting their access to targeted therapies, immunotherapies, or appropriately matched clinical trials. Artificial intelligence is now being deployed to address these diagnostic gaps. Machine learning classifiers trained on large multi omic datasets can detect tumor specific molecular signatures and have achieved accuracies of approximately 83 to 90 percent in predicting tissue of origin. For example, targeted gene mutation profiling combined with intelligent algorithms has enabled high confidence predictions in a substantial proportion of CUP patients, effectively doubling the number eligible for genomically guided therapy and leading to significantly improved survival for those treated according to artificial intelligence based predictions. Deep learning applied to histopathology has identified primary sites with approximately 80 percent accuracy, while cytology based systems such as newer tumor origin classifiers have shown about 83 percent top one accuracy and have been associated with prolonged survival when treatment was guided by model predictions. Importantly, a ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 231 randomized trial demonstrated that use of a 90 gene expression classifier to direct site specific therapy significantly extended progression free survival compared with empiric chemotherapy, providing strong clinical proof of concept that such approaches can change outcomes. Artificial intelligence enabled clinical decision support systems extend these advances into day to day clinical workflows. By embedding predictive models into oncology practice, clinical decision support tools can classify tumor origin and recommend therapy options, thereby reducing diagnostic uncertainty and improving clinician confidence. Deep learning systems trained on histopathology and cytology already provide structured differential diagnoses that support personalized therapy decisions in challenging cases. Integration with electronic health records enables real time synthesis of multimodal data, including pathology, imaging, laboratory results, and genomics, which streamlines oncologists’ ability to make informed treatment choices. Early implementations have shown high clinician satisfaction, and growing evidence suggests that multimodal artificial intelligence integration can improve both decision making and patient outcomes by standardizing assessment and revealing options that might otherwise be missed. Therapeutically, artificial intelligence is advancing precision strategies in CUP. Machine learning DR. MEHRDAD FARROKHI 232 can integrate molecular and clinical data to recommend targeted treatments or repurpose existing drugs based on tumor specific biomarkers and pathway alterations. Adaptive trial designs now leverage artificial intelligence for response adaptive randomization, dynamically steering patients toward more effective therapies as data accumulate over time. Critically, clinical evidence supports tangible outcome gains. Trials using a 90 gene classifier have demonstrated prolonged progression free survival with site specific therapy compared to empiric chemotherapy, while genomic profiling linked to targeted therapy has produced longer progression free survival than standard regimens in CUP cohorts. These findings highlight the growing potential of artificial intelligence driven therapeutic strategies to improve prognosis in this historically intractable disease and to convert biologic insights into practical clinical benefit. Despite the significant progress to date, further innovations are on the horizon to fully realize the potential of artificial intelligence in CUP care. Emerging trends in research include the development of explainable artificial intelligence techniques that increase transparency in how models predict tumor origin, thereby improving clinician and patient trust in artificial intelligence guided recommendations. Alongside explainability, multimodal learning is gaining ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 233 traction, in which algorithms concurrently analyze histopathology images, radiology scans, genomic sequences, and clinical data. Such integrated models are expected to yield more robust and accurate predictions than single modality systems. Another frontier is federated learning, which enables artificial intelligence models to be trained on data from multiple institutions without pooling sensitive patient data in a central location. This approach could greatly benefit CUP, a relatively rare cancer, by leveraging large, distributed datasets while preserving patient privacy and respecting data security regulations. In addition to technical advances, there is a growing focus on ethical and policy considerations for artificial intelligence in oncology. Model developers and clinicians are increasingly aware of the need to mitigate biases in training data that could otherwise lead to unequal performance of artificial intelligence tools across different patient populations. Ensuring that algorithms are audited for bias and fairness is crucial for maintaining patient trust and avoiding the perpetuation of healthcare disparities. Likewise, improving the interpretability of artificial intelligence outputs, for example through user friendly explanations or visualizations of what the model found significant, will be key for clinician acceptance DR. MEHRDAD FARROKHI 234 of artificial intelligence driven recommendations. On the regulatory side, professional societies and agencies have begun outlining frameworks for validation and deployment of artificial intelligence in clinical settings. Consensus recommendations stress rigorous prospective validation of any artificial intelligence diagnostic tool against the current standard of care and the development of guidelines to govern how artificial intelligence advice should be integrated into oncologists’ decision making workflows. As these policy frameworks solidify, they will provide much needed guidance on issues of accountability, data governance, and quality control for artificial intelligence systems in medicine. The long term vision is that artificial intelligence will enable truly personalized treatment for CUP to become standard of care. In the coming years, it is anticipated that every CUP patient’s tumor will undergo comprehensive molecular profiling and artificial intelligence based analysis at the outset, yielding an origin profile and a map of targetable vulnerabilities. Clinicians would then use this information to select effective, tailored therapies rather than defaulting to nonspecific chemotherapy. Ideally, CUP would cease to be a diagnosis of therapeutic uncertainty and instead would be managed with a precise, individualized plan comparable to the approach used for cancers of known primary. Achieving this vision ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 235 will likely require continued multidisciplinary collaboration, with oncologists, data scientists, pathologists, bioinformaticians, ethicists, and regulators working together, but recent studies have already laid the groundwork. Notably, the success of trials that identify a tissue of origin and treat accordingly suggests that tissue specific and biology driven strategies can improve outcomes, supporting the notion that an artificial intelligence driven, personalized approach may become the new standard of care for CUP in the near future. In summary, the convergence of explainable and multimodal artificial intelligence, ethical best practices, and supportive policy development is poised to transform CUP management and to bring renewed hope for better patient outcomes in this challenging area of oncology. AI in Early Diagnostic Evaluation When a patient presents with metastatic disease and no clear primary site, clinicians follow a diagnostic algorithm that includes targeted imaging, biopsies, laboratory tests, and sometimes invasive procedures. This evaluation is resource intensive and may still fail to uncover definitive answers. Artificial intelligence enhances this early diagnostic phase by identifying clues that human interpretation could overlook and by prioritizing tests that are most likely to yield informative results based on the unique characteristics of each DR. MEHRDAD FARROKHI 236 case. In the first stage of evaluation, radiologic imaging plays a central role. AI assisted interpretation of computed tomography, magnetic resonance imaging, and positron emission tomography improves visualization of subtle lesions that might represent the missing primary tumor. Deep learning models have shown capability in detecting very small primary tumors within the gastrointestinal tract, breast tissue, pancreas, and lungs based on texture irregularities, perfusion signals, or metabolic patterns that fall below typical human threshold recognition. These computational systems learn from thousands of confirmed examples to identify regions deserving additional scrutiny, increasing the chance of identifying the tumor origin early in the diagnostic process. Artificial intelligence also supports triage decisions by modeling which investigations are most likely to provide diagnostic answers for a specific clinical presentation. For example, if a patient arrives with liver predominant metastasis and limited symptom history, AI may estimate the probability of a primary in the gastrointestinal tract or pancreas and suggest the most informative endoscopic or sampling strategies. This avoids excessive and unnecessary procedures and reduces the time until treatment is initiated. Suspicion of cancer of unknown origin often ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 237 creates emotional distress for patients because of the uncertainty and fear that accompany a diagnosis of advanced malignancy. AI enabled systems can guide clinicians in communicating prognostic and diagnostic expectations more clearly by offering predictive confidence intervals and scenario modeling. This creates a more transparent and supportive experience, even when uncertainty remains. AI in Imaging Based Primary Site Identification Medical imaging remains an indispensable tool in the diagnosis and staging of metastatic cancers. When the primary site is unknown, radiology must be interpreted with heightened attention to patterns of spread. Metastatic pathways differ based on tumor lineage and influence lymphatic, hematologic, or transcoelomic dissemination. Artificial intelligence excels in analyzing complex patterns and correlating them with known disease trajectories. Machine learning models evaluate spatial relationships among metastases, comparing them to large datasets of known tumor origins. For instance, a metastatic lesion in the liver combined with bone involvement could originate from gastrointestinal or prostate primaries, whereas regional lymph node involvement in the cervical region paired with thoracic imaging anomalies DR. MEHRDAD FARROKHI 238 may suggest head and neck lineage. AI identifies these patterns quicker than manual review and generates probable primary sites ranked by likelihood. Even when accuracy is not absolute, narrowing diagnostic range significantly helps clinicians refine further testing. Radiomics extends this precision by quantifying imaging features such as lesion shape, surface sharpness, heterogeneity, and metabolic profile. Tumors from different organs maintain distinct radiomic signatures even after metastasis. Deep learning can cluster tumors based on these signatures and link them to tissue origin. Artificial intelligence also addresses complex imaging presentations such as peritoneal carcinomatosis or diffuse bone metastases where human interpretation alone may struggle to identify a clear source. Monitoring disease progression with AI creates additional advantages. When new lesions appear during early therapy response evaluation, models can reassess likelihood of origin based on emerging patterns. Tumors that evolve in specific trajectories over time reveal lineage characteristics through disease kinetics. By integrating radiomics, clinical features, and staging patterns, AI provides a continually improving system for identifying cancer origin from imaging data alone. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 239 AI Enhanced Pathology and Tissue Classification Biopsy analysis is fundamental to diagnosing cancer, yet metastatic lesions are often less differentiated than primary tumors and may lose key histologic features. Immunohistochemistry helps refine diagnosis but can leave significant uncertainty. The increasing availability of digital pathology has opened new avenues for artificial intelligence to contribute detailed microscopic evaluation capable of detecting lineage specific traits that remain even when tumors appear morphologically ambiguous. Computational pathology uses convolutional neural networks to study entire slide images and extract subtle image based biomarkers. Nuclear atypia, stromal composition, and architectural patterns can be linked through machine learning to tissue origin with predictive accuracy that exceeds manual estimation. Sarcomatoid or poorly differentiated carcinomas that once defied clear classification may now reveal hidden lineage associations through AI supported pattern recognition. One promising area involves prediction of tumor primary based solely on metastatic biopsy images. Studies suggest that when a model is trained on large databases of known primary tissues, it can differentiate metastatic colorectal cancer from DR. MEHRDAD FARROKHI 240 metastatic breast or lung cancer by analyzing features that cannot be articulated easily even by expert pathologists. These discoveries enrich multidisciplinary decision making and reduce reliance on broader empiric treatments. Molecular interpretation builds on pathology by integrating AI at the genomic level. Gene expression signatures, mutational landscapes, and epigenetic markers vary by tissue type. Machine learning methods evaluate countless data points simultaneously to identify combinations that best predict tumor origin. For example, a panel of microRNA signatures can guide classification of carcinoma origin more accurately when enriched by AI clustering algorithms that reflect tumor behavior rather than isolated markers. AI enhanced tissue diagnostics remove guesswork by enabling a more comprehensive and biologically grounded understanding of metastatic tumor identity. AI for Precision Systemic Therapy Once a likely primary is identified or a molecular phenotype becomes clearer through AI analysis, treatment selection becomes more precise. Historically, therapy for unknown origin cancers has consisted of empiric chemotherapy with modest efficacy. Artificial intelligence redefines systemic therapy strategy by enabling personalized variability in biologic targeting. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 241 R, Joven J, Arenas M. Artificial intelligencedriven integration of multi-omics and radiomics: A new hope for precision cancer diagnosis and prognosis. Biochim Biophys Acta Mol Basis Dis. 2025;1871(6):167841. 30. Cannarozzi AL, Biscaglia G, Parente P, Latiano TP, Gentile A, Ciardiello D, et al. Artificial intelligence and whole slide imaging, a new tool for the microsatellite instability prediction in colorectal cancer: Friend or foe? Crit Rev Oncol Hematol. 2025;210:104694. 31. Caraka RE, Supardi K, Gondhowiardjo SA, Isnaniawardhani V, Gio PU, Chen RC, et al. Artificial Intelligence in Cancer Oncology Through Comprehensive Bibliometric Mapping of Global Trends Impact and Conceptual Structures. J Healthc Leadersh. 2025;17:591-621. 32. Cece A, Agresti M, De Falco N, Sperlongano P, Moccia G, Luongo P, et al. Role of Artificial Intelligence in Thyroid Cancer Diagnosis. J Clin Med. 2025;14 )7(. 33. Chadha S, Mukherjee S, Sanyal S. Advancements and implications of artificial intelligence for early detection, diagnosis and tailored treatment of cancer. Semin Oncol. 2025;52(3):152349. 34. Chamouni G, Lococo F, Sassorossi C, Atuhaire N, Ádány R, Varga O. Ethical and legal concerns in artificial intelligence applications DR. MEHRDAD FARROKHI 254 for the diagnosis and treatment of lung cancer: a scoping review. Front Public Health. 2025;13:1663298. 35. Chen L, Zhao J, Liu Z, Kong L, Zhou D, Wang F. Global research hotspots and trends in the application of artificial intelligence in gastric cancer: a bibliometric analysis from 2005 to 2024. Front Oncol. 2025;15:1591655. 36. Chen S, Ding P, Guo H, Meng L, Zhao Q, Li C. Applications of artificial intelligence in digital pathology for gastric cancer. Front Oncol. 2024;14:1437252. 37. Chen Y, Sun Y, Chen Y, Zhang J, Zhang H, Liu K, et al. Artificial intelligenceassisted PET imaging for predicting neoadjuvant chemotherapy response in breast cancer: a systematic review and meta-analysis. Eur J Nucl Med Mol Imaging. 2025. 38. Cheng CH, Shi SS. Artificial intelligence in cancer: applications, challenges, and future perspectives. Mol Cancer. 2025;24(1):274. 39. Cheng CH, Wong CC. The role of artificial intelligence-based foundation models and "copilots" in cancer pathology: potential and challenges. J Exp Clin Cancer Res. 2025. 40. Cherukuri SP, Kaur A, Goyal B, Kukunoor HR, Sahito AF, Sachdeva P, et al. Artificial Intelligence-Enhanced Liquid Biopsy and Radiomics in Early-Stage Lung Cancer Detection: ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 255 A Precision Oncology Paradigm. Cancers (Basel). 2025;17 )19(. 41. Chia JLL, He GS, Ngiam KY, Hartman M, Ng QX, Goh SSN. Harnessing Artificial Intelligence to Enhance Global Breast Cancer Care: A Scoping Review of Applications, Outcomes, and Challenges. Cancers (Basel). 2025;17 )2(. 42. Chierici A, Lareyre F, Iannelli A, Salucki B, Goffart S, Guzzi L, et al. Applications of artificial intelligence in liver cancer: A scoping review. Artif Intell Med. 2025;169:103244. 43. Ciancia M, Moro F, Bertoni M, Baldassari G, Schips P, Fanfani F, et al. Role of artificial intelligence applied to ultrasound in endometrial cancer: a systematic review. Int J Gynecol Cancer. 2025;35(11):102653. 44. Ciccone V, Garofano M, Del Sorbo R, Mongelli G, Izzo M, Negri F, et al. Improving Early Prostate Cancer Detection Through Artificial Intelligence: Evidence from a Systematic Review. Cancers (Basel). 2025;17 )21(. 45. Dalboni da Rocha JL, Lai J, Pandey P, Myat PSM, Loschinskey Z, Bag AK, et al. Artificial Intelligence for Neuroimaging in Pediatric Cancer. Cancers (Basel). 2025;17 )4(. 46. Dankwa-Mullan I, Ndoh K, Akogo D, Rocha HAL, Juaçaba SF. Artificial Intelligence and Cancer Health Equity: Bridging the Divide or Widening the Gap. Curr Oncol Rep. 2025;27(2):95-111. DR. MEHRDAD FARROKHI 256 47. De Luca GR, Diciotti S, Mascalchi M. The Pivotal Role of Baseline LDCT for Lung Cancer Screening in the Era of Artificial Intelligence. Arch Bronconeumol. 2025;61(6):359-67. 48. Debellotte O, Dookie RL, Rinkoo F, Kar A, Salazar González JF, Saraf P, et al. Artificial Intelligence and Early Detection of Breast, Lung, and Colon Cancer: A Narrative Review. Cureus. 2025;17(2):e79199. 49. Derbal Y. Adaptive Treatment of Metastatic Prostate Cancer Using Generative Artificial Intelligence. Clin Med Insights Oncol. 2025;19:11795549241311408. 50. Desai M, Desai B. Evolution and integration of artificial intelligence across the cancer continuum in women: advances in risk assessment, prevention, and early detection. Cancer Causes Control. 2025;36(11):1343-60. 51. Di Mauro A, Berretta M, Santorsola M, Ferrara G, Picone C, Savarese G, et al. Towards Post-Genomic Oncology: Embracing Cancer Complexity via Artificial Intelligence, Multi-Targeted Therapeutics, Drug Repurposing, and Innovative Study Designs. Int J Mol Sci. 2025;26 )16(. 52. Dimopoulos P, Mulita A, Antzoulas A, Bodard S, Leivaditis V, Akrida I, et al. The role of artificial intelligence and image processing in the diagnosis, treatment, and prognosis of ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 257 liver cancer: a narrative-review. Prz Gastroenterol. 2024;19(3):221-30. 53. Dommalapati L, Guenter R, Golivi Y, Ganji SP, Sandhya T, Nagaraju GP, et al. Advances in artificial intelligence for spatial transcriptomics in cancer: Special focus on Yin Yang 1 (YY1) and Raf kinase inhibitor protein (RKIP). Biochim Biophys Acta Rev Cancer. 2025;1880(6):189456. 54. Donatello D. An overview of the use of cutting-edge artificial intelligence (AI) modeling to produce synthetic medical data (SMD) in decentralized clinical machine learning (ML) for ovarian cancer(OC) and ovarian lymphoma(OL). J Ultrasound. 2025;28(2):483-92. 55. Duan C, Sheng J, Ma X. Innovative approaches in colorectal cancer screening: advances in detection methods and the role of artificial intelligence. Therap Adv Gastroenterol. 2025;18:17562848251314829. 56. Duranti L, Tavecchio L, Rolli L, Solli P. New Perspectives on Lung Cancer Screening and Artificial Intelligence. Life (Basel). 2025;15 )3(. 57. Ejaz ZH, Shaikh RH, Fatimi AS, Khan SR. Unlocking artificial intelligence, machine learning and deep learning to combat therapeutic resistance in metastatic castrationresistant prostate cancer: a comprehensive review. Ecancermedicalscience. 2025;19:1953. 58. Elahi R, Karami P, Amjadzadeh M, Nazari M. DR. MEHRDAD FARROKHI 258 Radiomics-based artificial intelligence (AI) models in colorectal cancer (CRC) diagnosis, metastasis detection, prognosis, and treatment response prediction. Abdom Radiol (NY). 2025. 59. Elsaigh M, Baqar S, Awan B, Saleh OS, Hesham Gamal M, Abara N, et al. Comparative Effectiveness of Artificial Intelligence Versus Conventional Methods for Detecting Peritoneal Metastasis in Colorectal Cancer: A Systematic Review. Cureus. 2025;17(10):e95484. 60. Farhoudian A, Heidari A, Shahhosseini R. A new era in colorectal cancer: Artificial Intelligence at the forefront. Comput Biol Med. 2025;196(Pt C):110926. 61. Fatima SH, Faisal SJ, Batool SV, Faisal M, Aamir T, Shahid ZUA, et al. Evaluating the combined efficacy of Telisotuzumab Vedotin and artificial intelligence in the treatment of nonsquamous non-small cell lung cancer: a narrative review focusing on pharmaceutical and technical insights. Front Oncol. 2025;15:1673586. 62. Felici A, Peduzzi G, Pellungrini R, Campa D. Artificial intelligence to predict cancer risk, are we there yet? A comprehensive review across cancer types. Eur J Cancer. 2025;222:115440. 63. Feng K, Yi Z, Xu B. Artificial Intelligence and Breast Cancer Management: From Data to the Clinic. Cancer Innov. 2025;4(2):e159. 64. Firuzpour F, Heydari M, Aram C, Alishvandi ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 259 A. The role of artificial intelligence in enhancing breast cancer screening and diagnosis: A review of current advances. Bioimpacts. 2025;15:30984. 65. Fogli S, Barberis A, Re MD, Crucitta S, Ruglioni M, Luculli G, et al. Artificial intelligencebased pharmacological approach in non-small cell lung cancer in the precision medicine era. Cancer Treat Res Commun. 2025;45:101023. 66. Freitas V, Ghai S, Au F, Muradali D, Kulkarni S. The Transformative Power of Digital Breast Tomosynthesis and Artificial Intelligence in Breast Cancer Diagnosis. Can Assoc Radiol J. 2025;76(2):302-12. 67. Fu M, Xu J, Lv Y, Jin B. Artificial intelligence in advanced gastric cancer: a comprehensive review of applications in precision oncology. Front Oncol. 2025;15:1630628. 68. Fuster-Matanzo A, Picó-Peris A, BellvísBataller F, Jimenez-Pastor A, Weiss GJ, MartíBonmatí L, et al. Prediction of oncogene mutation status in non-small cell lung cancer: a systematic review and meta-analysis with a special focus on artificial intelligence-based methods. Eur Radiol. 2025. 69. Galani A, Zikopoulos A, Moustakli E, Potiris A, Paraskevaidi M, Arkoulis I, et al. Cervical Cancer Screening in the HPV-Vaccinated and Digital Era: Reassessing Strategies in Light of Artificial Intelligence and Evolving Risk. Cancers (Basel). DR. MEHRDAD FARROKHI 260 2025;17 )19(. 70. Gao Y, Chen J, Fu T, Gu Y, Du W. Quantitative analysis of studies that use artificial intelligence on thyroid cancer: a 20-year bibliometric analysis. Front Oncol. 2025;15:1525650. 71. Garcia-Atutxa I, Villanueva-Flores F, Barrio ED, Sanchez-Villamil JI, Martínez-Más J, BuenoCrespo A. Artificial intelligence for ovarian cancer diagnosis via ultrasound: a systematic review and quantitative assessment of model performance. Front Artif Intell. 2025;8:1649746. 72. Ge KK, Jin X, Zhong JG, Sun XY, Xie HS, Peng SB, et al. [Applications of artificial intelligence in the diagnosis and treatment of lung cancer]. Zhonghua Zhong Liu Za Zhi. 2025;47(11):1057-65. 73. Geppert J, Asgharzadeh A, Brown A, Stinton C, Helm EJ, Jayakody S, et al. Software using artificial intelligence for nodule and cancer detection in CT lung cancer screening: systematic review of test accuracy studies. Thorax. 2024;79(11):1040-9. 74. Ghane G, Karimi R, Chekeni AM, Darvishi M, Imani R, Vafaeinezhad FZ. Pain Management in Cancer Patients With Artificial Intelligence: Narrative Review. Scientifica (Cairo). 2025;2025:6888213. 75. Ghasemi A, Hashtarkhani S, Schwartz DL, Shaban-Nejad A. Explainable artificial intelligence ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 261 in breast cancer detection and risk prediction: A systematic scoping review. Cancer Innov. 2024;3(5):e136. 76. Goh S, Goh RSJ, Chong B, Ng QX, Koh GCH, Ngiam KY, et al. Challenges in Implementing Artificial Intelligence in Breast Cancer Screening Programs: Systematic Review and Framework for Safe Adoption. J Med Internet Res. 2025;27:e62941. 77. Gonçalves N, Chaves J, Marques-Sá I, DinisRibeiro M, Libânio D. Early diagnosis of gastric cancer: Endoscopy and artificial intelligence. Best Pract Res Clin Gastroenterol. 2025;75:101979. 78. Grapă C, Mocan T, Mocan LP, Motofelea A, Stănciulescu R, Crăciun R, et al. Turning the TideArtificial Intelligence in the Evolving Landscape of Liver Cancer. Cancers (Basel). 2025;17 )18(. 79. Grunhut J, Newland JJ, Brown RF. Implications of Artificial Intelligence for Colorectal Cancer in Young Populations. J Surg Oncol. 2025;131(7):1368-72. 80. Guo Y, Li T, Gong B, Hu Y, Wang S, Yang L, et al. From Images to Genes: Radiogenomics Based on Artificial Intelligence to Achieve Non-Invasive Precision Medicine in Cancer Patients. Adv Sci (Weinh). 2025;12(2):e2408069. 81. Hachache R, Yahyaouy A, Riffi J, Tairi H, Abibou S, Adoui ME, et al. Advancing personalized oncology: a systematic review on the DR. MEHRDAD FARROKHI 262 integration of artificial intelligence in monitoring neoadjuvant treatment for breast cancer patients. BMC Cancer. 2024;24(1):1300. 82. Hagen J, Hornung L, Barham W, Mukhopadhyay S, Bess A, Contrera K, et al. Artificial Intelligence in Head and Neck Cancer: Towards Precision Medicine. Cancers (Basel). 2025;17 )18(. 83. HaghighiKian SM, Shirinzadeh-Dastgiri A, Vakili-Ojarood M, Naseri A, Barahman M, Saberi A, et al. A Holistic Approach to Implementing Artificial Intelligence in Lung Cancer. Indian J Surg Oncol. 2025;16(1):257-78. 84. Hamamoto R, Komatsu M, Yamada M, Kobayashi K, Takahashi M, Miyake M, et al. Current status and future direction of cancer research using artificial intelligence for clinical application. Cancer Sci. 2025;116(2):297-307. 85. Haque F, Simon BD, Özyörük KB, Harmon SA, Türkbey B. Generative Artificial Intelligence in Prostate Cancer Imaging. Balkan Med J. 2025;42(4):286-300. 86. Hardacre C, Hibbs T, Fok M, Wiles R, Bashar N, Ahmed S, et al. Predicting Surgical Difficulty in Rectal Cancer Surgery: A Systematic Review of Artificial Intelligence Models Applied to PreOperative MRI. Cancers (Basel). 2025;17 )5(. 87. Hasan MR, Ibraheem N, Rahman ME, Tamanna R. Artificial Intelligence Across ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 263 the Prostate Cancer Pathway: Screening, Imaging, Pathology, and Biomarkers. Cureus. 2025;17(11):e96226. 88. Hashem H, Sultan I. Revolutionizing precision oncology: the role of artificial intelligence in personalized pediatric cancer care. Front Med (Lausanne). 2025;12:1555893. 89. Hashim HT, Alhatemi AQM, Daraghma M, Ali HT, Khan MA, Sulaiman FA, et al. Artificial intelligence versus radiologists in detecting earlystage breast cancer from mammograms: a metaanalysis of paradigm shifts. Pol J Radiol. 2025;90:e1-e8. 90. Hassan C, Bisschops R, Sharma P, Mori Y. Colon Cancer Screening, Surveillance, and Treatment: Novel Artificial Intelligence Driving Strategies in the Management of Colon Lesions. Gastroenterology. 2025;169(3):444-55. 91. Haue AD, Hjaltelin JX, Holm PC, Placido D, Brunak SR. Artificial intelligence-aided data mining of medical records for cancer detection and screening. Lancet Oncol. 2024;25(12):e694e703. 92. Hays P. Artificial intelligence in cytopathological applications for cancer: a review of accuracy and analytic validity. Eur J Med Res. 2024;29(1):553. 93. He S, Li X, Xie Y, Hou J, Han B, Nie S. [Artificial intelligence in predicting pathological DR. MEHRDAD FARROKHI 264 complete response to neoadjuvant chemotherapy for breast cancer: current advances and challenges]. Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2025;42(5):1076-84. 94. Hengky A, Lionardi SK, Kusumajaya C. Can artificial intelligence aid the urologists in detecting bladder cancer? Indian J Urol. 2024;40(4):221-8. 95. Hilbers D, Nekain N, Bates A, Nunez JJ. Patient Attitudes Toward Artificial Intelligence in Cancer Care: Scoping Review. JMIR Cancer. 2025;11:e74010. 96. Hofman P, Ourailidis I, Romanovsky E, Ilié M, Budczies J, Stenzinger A. Artificial intelligence for diagnosis and predictive biomarkers in NonSmall cell lung cancer Patients: New promises but also new hurdles for the pathologist. Lung Cancer. 2025;200:108110. 97. Howard HR, Hasanova M, Tiwari A, Ghose A, Winayak R, Nahar T, et al. The landscape of conventional and artificial intelligence-based clinical prediction models in non-small-cell lung cancer: from development to real-world validation. ESMO Open. 2025;10(9):105557. 98. Huang C, Shen Y, Galgano SJ, Goenka AH, Hecht EM, Kambadakone A, et al. Advancements in early detection of pancreatic cancer: the role of artificial intelligence and novel imaging techniques. Abdom Radiol (NY). ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 265 2025;50(4):1731-43. 99. Huang D, Li Z, Jiang T, Yang C, Li N. Artificial intelligence in lung cancer: current applications, future perspectives, and challenges. Front Oncol. 2024;14:1486310. 100. Huang L, Wu X, You J, Jin Z, He W, Sun J, et al. Artificial Intelligence Can Predict Personalized Immunotherapy Outcomes in Cancer. Cancer Immunol Res. 2025;13(7):964-77. 101. Huang R, Jin X, Liu Q, Bai X, Karako K, Tang W, et al. Artificial intelligence in colorectal cancer liver metastases: From classification to precision medicine. Biosci Trends. 2025;19(2):150-64. 102. Hussain MS, Ramalingam PS, Chellasamy G, Yun K, Bisht AS, Gupta G. Harnessing Artificial Intelligence for Precision Diagnosis and Treatment of Triple Negative Breast Cancer. Clin Breast Cancer. 2025;25(5):406-21. 103. Ichimasa K, Kudo SE, Kouyama Y, Takashina Y, Chung H, Maeda Y, et al. Artificial Intelligence-Assisted Whole Slide Image Analysis for Lymph Node Status Prediction in Early Colorectal and Gastric Cancer. Dig Endosc. 2025. 104. Imani S, Li X, Chen K, Maghsoudloo M, Jabbarzadeh Kaboli P, Hashemi M, et al. Computational biology and artificial intelligence in mRNA vaccine design for cancer immunotherapy. Front Cell Infect Microbiol. 2024;14:1501010. DR. MEHRDAD FARROKHI 266 105. Ingman WV, Britt KL, Stone J, Nguyen TL, Hopper JL, Thompson EW. Artificial intelligence improves mammography-based breast cancer risk prediction. Trends Cancer. 2025;11(3):188-91. 106. Javanmard Z, Zarean Shahraki S, Safari K, Omidi A, Raoufi S, Rajabi M, et al. Artificial intelligence in breast cancer survival prediction: a comprehensive systematic review and metaanalysis. Front Oncol. 2024;14:1420328. 107. Jeong S, Choi HI, Yang KI, Kim JS, Ryu JW, Park HJ. Artificial Intelligence in the Diagnosis of Tongue Cancer: A Systematic Review with MetaAnalysis. Biomedicines. 2025;13 )8(. 108. Jerjes W, Stevenson H, Ramsay D, Hamdoon Z. Enhancing Oral Cancer Detection: A Systematic Review of the Diagnostic Accuracy and Future Integration of Optical Coherence Tomography with Artificial Intelligence. J Clin Med. 2024;13 )19(. 109. Ji J, Duan F, Liao Q, Wang H, Liu S, Liu Y, et al. Artificial Intelligence-Based Pathology to Assist Prediction of Neoadjuvant Therapy Responses for Breast Cancer. Cancer Med. 2025;14(15):e71132. 110. Jiang CQ, Li XJ, Zhou ZY, Xin Q, Yu L. Imaging based artificial intelligence for predicting lymph node metastasis in cervical cancer patients: a systematic review and meta-analysis. Front Oncol. 2025;15:1532698. 111. K RU, Elango AP, Subramanian R, Ks ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 267 S. The Evolving Landscape of Ovarian Cancer: Innovations in Biotechnology and Artificial IntelligenceBased Screening and Treatment. Curr Treat Options Oncol. 2025;26(7):622-37. 112. Kadir A, Asaduzzaman M, Kundu J, Rahman MA, Rabbany MG, Shemanto MU, et al. Integrating Genetic Insights and Artificial Intelligence for Enhanced Oral and Maxillofacial Cancer Care. Methods Mol Biol. 2025;2952:107-24. 113. Kagan S, Huynh L, Strickland C, Wang A, Kwan JYY, Ng TL, et al. Survival prediction in metastatic breast cancer using artificial intelligence: a scoping review. Curr Opin Support Palliat Care. 2025;19(4):279-89. 114. Kaidar-Person O, Pfob A, Valentini V, Aznar M, Dekker A, Meattini I, et al. Artificial intelligence in breast cancer radiotherapy: Insights from the Toolbox Consortium Delphi study. Breast. 2025;83:104537. 115. Karimzadhagh S, Ghodous S, Robati RM, Abbaspour E, Goldust M, Zaresharifi N, et al. Performance of Artificial Intelligence in Skin Cancer Detection: An Umbrella Review of Systematic Reviews and Meta-Analyses. Int J Dermatol. 2025. 116. Karuppan Perumal MK, Rajan Renuka R, Kumar Subbiah S, Manickam Natarajan P. Artificial intelligence-driven clinical decision support systems for early detection and precision therapy DR. MEHRDAD FARROKHI 268 in oral cancer: a mini review. Front Oral Health. 2025;6:1592428. 117. Kedar P, Bhattacharya S, Kanugo A, Prajapati BG. Novel Strategies for the Treatment of Lung Cancer: An In-depth Analysis of the Use of Immunotherapy, Precision Medicine, and Artificial Intelligence to Improve Prognoses. Curr Med Chem. 2025;32(35):7752-76. 118. Kemna R, Zeeuw JM, Ziesemer KA, Ali M, Bereska JI, Marquering H, et al. From Development to Implementation: A Systematic Review on the Current Maturity Status of Artificial Intelligence Models for Patients with Colorectal Cancer Liver Metastases. Oncology. 2025:1-10. 119. Khalid U, Gurung J, Doykov M, Kostov G, Hristov B, Uchikov P, et al. Artificial Intelligence Algorithms and Their Current Role in the Identification and Comparison of Gleason Patterns in Prostate Cancer Histopathology: A Comprehensive Review. Diagnostics (Basel). 2024;14 )19(. 120. Khosravi M, Jasemi SK, Hayati P, Javar HA, Izadi S, Izadi Z. Transformative artificial intelligence in gastric cancer: Advancements in diagnostic techniques. Comput Biol Med. 2024;183:109261. 121. Khosravi P, Fuchs TJ, Ho DJ. Artificial Intelligence-Driven Cancer Diagnostics: Enhancing Radiology and Pathology ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 269 through Reproducibility, Explainability, and Multimodality. Cancer Res. 2025;85(13):2356-67. 122. Kim SJ, Clark V, Hancock JT, Rawassizadeh R, Liu H, Taylor EA, et al. Leveraging artificial intelligence-mediated communication for cancer prevention and control and drug addiction: A systematic review. Transl Behav Med. 2025;15 )1(. 123. Kotoulas SC, Spyratos D, Porpodis K, Domvri K, Boutou A, Kaimakamis E, et al. A Thorough Review of the Clinical Applications of Artificial Intelligence in Lung Cancer. Cancers (Basel). 2025;17 )5(. 124. Krengli M, Kruszyna-Mochalska MM, Pasqualetti F, Malicki J. Can Advances in Artificial Intelligence Strengthen the Role of Intraoperative Radiotherapy in the Treatment of Cancer? Cancers (Basel). 2025;17 )19(. 125. Kumar A, Goel S, Chaudhary A, Dutt S, Mishra VK, Kumar R. Artificial IntelligenceBased Wearable Sensing Technologies for the Management of Cancer, Diabetes, and COVID-19. Biosensors (Basel). 2025;15 )11(. 126. Kumar Verma K. Artificial intelligence models for lung cancer diagnosis: A comprehensive review. Curr Probl Diagn Radiol. 2025. 127. Kuppanda PM, Janda M, Soyer HP, Caffery LJ. What Are Patients' Perceptions and Attitudes Regarding the Use of Artificial Intelligence in DR. MEHRDAD FARROKHI 270 Skin Cancer Screening and Diagnosis? Narrative Review. J Invest Dermatol. 2025;145(8):1858-65. 128. Lei C, Sun W, Wang K, Weng R, Kan X, Li R. Artificial intelligence-assisted diagnosis of early gastric cancer: present practice and future prospects. Ann Med. 2025;57(1):2461679. 129. Leonard S, Patel MA, Zhou Z, Le H, Mondal P, Adams SJ. Comparing Artificial Intelligence and Traditional Regression Models in Lung Cancer Risk Prediction Using A Systematic Review and MetaAnalysis. J Am Coll Radiol. 2025;22(6):675-90. 130. Leszczyńska A, Obuchowicz R, Strzelecki M, Seweryn M. The Integration of Artificial Intelligence into Robotic Cancer Surgery: A Systematic Review. J Clin Med. 2025;14 )17(. 131. Li G, Shi Q, Wu Q, Sui X. Target identification of natural products in cancer with chemical proteomics and artificial intelligence approaches. Cancer Biol Med. 2025;22(6):549-97. 132. Li K, Wu S, Zhang Y, Zhu B, Qi Z, Hou S, et al. The Usability and Experience of Artificial Intelligence-Based Conversational Agents in Health Education for Cancer Patients: A Scoping Review. J Clin Nurs. 2025. 133. Li R, Li J, Wang Y, Liu X, Xu W, Sun R, et al. The artificial intelligence revolution in gastric cancer management: clinical applications. Cancer Cell Int. 2025;25(1):111. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 271 134. Li W, Hu R, Zhang Q, Yu Z, Deng L, Zhu X, et al. Artificial intelligence in prostate cancer. Chin Med J (Engl). 2025;138(15):1769-82. 135. Liao W, Xu X. Progress in the application research of cervical cancer screening developed by artificial intelligence in large populations. Discov Oncol. 2025;16(1):1282. 136. Lin X, Zhang Z, Zhou T, Li J, Jin Q, Li Y, et al. The Role of Computed Tomography and Artificial Intelligence in Evaluating the Comorbidities of Chronic Obstructive Pulmonary Disease: A OneStop CT Scanning for Lung Cancer Screening. Int J Chron Obstruct Pulmon Dis. 2025;20:1395-406. 137. Lin Y, Cheng M, Wu C, Huang Y, Zhu T, Li J, et al. MRI-based artificial intelligence models for post-neoadjuvant surgery personalization in breast cancer: a narrative review of evidence from Western Pacific. Lancet Reg Health West Pac. 2025;57:101254. 138. Liu J, Fu R, Su Y, Li Z, Huang X, Wang Q, et al. Applications of artificial intelligence in cancer immunotherapy: a frontier review on enhancing treatment efficacy and safety. Front Immunol. 2025;16:1676112. 139. Liu J, Zhang H, Woon DTS, Perera M, Lawrentschuk N. Predicting Biochemical Recurrence of Prostate Cancer Post-Prostatectomy Using Artificial Intelligence: A Systematic Review. Cancers (Basel). 2024;16 )21(. DR. MEHRDAD FARROKHI 272 140. Liu JY, Sun RH, Li C. [Advances in the application of artificial intelligence technology in the surgical diagnosis and treatment of thyroid cancer]. Zhonghua Er Bi Yan Hou Tou Jing Wai Ke Za Zhi. 2025;60(1):80-6. 141. Liu L, Pei Q, Qadir J, Chen Y, Li J, Luo Y, et al. Application of artificial intelligencebased stemness index in cancer. Front Oncol. 2025;15:1608712. 142. Liu M, Li L, Rais-Bahrami S, Hemal AK, Atala A, Zhang Y. Artificial intelligence in prostate cancer: navigating the new frontier of precision uro-oncology. Am J Clin Exp Urol. 2025;13(5):348-59. 143. Lococo F, Ghaly G, Flamini S, Campanella A, Chiappetta M, Bria E, et al. Artificial intelligence applications in personalizing lung cancer management: state of the art and future perspectives. J Thorac Dis. 2024;16(10):7096-110. 144. Long X, Sun K, Lai S, Liu Y, Su J, Chen W, et al. Artificial intelligence and anti-cancer drugs' response. Acta Pharm Sin B. 2025;15(7):3355-71. 145. Luan WY, Zhao Q, Zhang Z, Xu ZX, Lin SX, Miao YD. Multidimensional decoding of colorectal cancer heterogeneity: Artificial intelligenceenabled precision exploration of single-cell and spatial transcriptomics. World J Gastrointest Oncol. 2025;17(10):110661. 146. Lukac S, Putz F, De Micheli G, Corti C, ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 273 emerging integrations-a narrative review. Discov Oncol. 2025;16(1):697. 213. Sampieri C, Peretti G. Democratizing cancer detection: artificial intelligence-enhanced endoscopy could address global disparities in head and neck cancer outcomes. Eur Arch Otorhinolaryngol. 2025;282(5):2739-43. 214. Sarvepalli S, Vadarevu S. Role of artificial intelligence in cancer drug discovery and development. Cancer Lett. 2025;627:217821. 215. Scalia IG, Pathangey G, Abdelnabi M, Ibrahim OH, Abdelfattah FE, Pietri MP, et al. Applications of Artificial Intelligence for the Prediction and Diagnosis of Cancer TherapyRelated Cardiac Dysfunction in Oncology Patients. Cancers (Basel). 2025;17 )4(. 216. Schito L, Rey-Keim S. Transcriptional regulation of hypoxic cancer cell metabolism and artificial intelligence. Trends Cancer. 2025;11(2):88-90. 217. Šebestová G, Klinger T, Švajdler M, Jr., Daum O, Jirásek T. Utilization of Artificial Intelligence Algorithms for the Diagnosis of Breast, Lung, and Prostate Cancer. Cesk Patol. 2025;61(2):70-90. 218. Selvaraj C, Cho WC, Langeswaran K, Alothaim AS, Vijayakumar R, Jayaprakashvel M, et al. Artificial intelligence in cancer care: revolutionizing diagnosis, treatment, and DR. MEHRDAD FARROKHI 286 precision medicine amid emerging challenges and future opportunities. 3 Biotech. 2025;15(10):355. 219. Seth L, Ladbury C, Amini A. Artificial Intelligence and Machine Learning Approaches in Designing Immunotherapy in Cancer. Cancer Treat Res. 2025;129:17-32. 220. Sguanci M, Palomares SM, Cangelosi G, Petrelli F, Sandri E, Ferrara G, et al. Artificial Intelligence in the Management of Malnutrition in Cancer Patients: A Systematic Review. Adv Nutr. 2025;16(7):100438. 221. Shahzad K, Abu-Zanona M, Elzaghmouri BM, AbdelRahman SM, Fadol Osman AA, Al-Khateeb A, et al. USING ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING APPROACHES TO ENHANCE CANCER THERAPY AND DRUG DISCOVERY: A NARRATIVE REVIEW. J Ayub Med Coll Abbottabad. 2024;36(1):183-9. 222. Shaktah LA, Carrero ZI, Hewitt KJ, Gustav M, Cecchini M, Foersch S, et al. Application of artificial intelligence and digital tools in cancer pathology. Lancet Digit Health. 2026:100933. 223. Shen K, Tan M, Liu Y, Xu X, Yang S. From data to Diagnosis: How artificial intelligence is revolutionizing preoperative assessment of thyroid nodules and cancer. Eur J Surg Oncol. 2025;51(9):110191. 224. Shi J, Chen J, He G, Peng Q. Artificial intelligence in high-dose-rate brachytherapy ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 287 treatment planning for cervical cancer: a review. Front Oncol. 2025;15:1507592. 225. Shirzad M, Salahvarzi A, Razzaq S, JavidNaderi MJ, Rahdar A, Fathi-Karkan S, et al. Revolutionizing prostate cancer therapy: Artificial intelligence - Based nanocarriers for precision diagnosis and treatment. Crit Rev Oncol Hematol. 2025;208:104653. 226. Silveira JA, da Silva AR, de Lima MZT. Harnessing artificial intelligence for predicting breast cancer recurrence: a systematic review of clinical and imaging data. Discov Oncol. 2025;16(1):135. 227. Silvestre-Barbosa Y, Castro VT, Di Carvalho Melo L, Reis PED, Leite AF, Ferreira EB, et al. Worldwide research trends on artificial intelligence in head and neck cancer: a bibliometric analysis. Oral Surg Oral Med Oral Pathol Oral Radiol. 2025;140(1):64-78. 228. Smiley A, Reategui-Rivera CM, VillarrealZegarra D, Escobar-Agreda S, Finkelstein J. Exploring Artificial Intelligence Biases in Predictive Models for Cancer Diagnosis. Cancers (Basel). 2025;17 )3(. 229. Soghli N, Khormali A, Mahboubi D, Peng A, Miguez PA. Recent advancements in artificial intelligence-powered cancer prediction from oral microbiome. Periodontol 2000. 2025. 230. Song B, Liang R. Integrating artificial DR. MEHRDAD FARROKHI 288 intelligence with smartphone-based imaging for cancer detection in vivo. Biosens Bioelectron. 2025;271:116982. 231. Song J, Gao Y, Liu W, Sun X, Chen C, Wu IX. Effectiveness of Artificial Intelligence-Assisted Examination for Cancer Detection in Medical Imaging: A Systematic Review and Meta-Analysis. J Am Coll Radiol. 2025. 232. Stevenson E, Esengur OT, Zhang H, Simon BD, Harmon SA, Turkbey B. An overview of utilizing artificial intelligence in localized prostate cancer imaging. Expert Rev Med Devices. 2025;22(4):293-310. 233. Sultany A, Chikatimalla R, Rao A, Omar MA, Shaar A, Ali H, et al. Real-Time Artificial Intelligence Versus Standard Colonoscopy in the Early Detection of Colorectal Cancer: A Systematic Review and Meta-Analysis. Healthcare (Basel). 2025;13 )19(. 234. Sun K, Wang Y, Qu R, Yang Q, Luo R, Jiang Z, et al. Comprehensive application of artificial intelligence in colorectal cancer: A review. iScience. 2025;28(7):112980. 235. Sun L, Zhang R, Gu Y, Huang L, Jin C. Application of Artificial Intelligence in the diagnosis and treatment of colorectal cancer: a bibliometric analysis, 2004-2023. Front Oncol. 2024;14:1424044. 236. Sun Y, Guo X, Liu X, Liu H, Wang ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 289 X. Advances in artificial intelligence-based radiogenomics for lung cancer precision medicine. Prog Biomed Eng (Bristol). 2025;8 )1(. 237. Tabataba Vakili S, Haywood D, Kirk D, Abdou AM, Gopalakrishnan R, Sadeghi S, et al. Application of Artificial Intelligence in Symptom Monitoring in Adult Cancer Survivorship: A Systematic Review. JCO Clin Cancer Inform. 2024;8:e2400119. 238. Taha HA, Zeilani RS, Haddad RH, Abdalrahim MS. Artificial intelligence and machine learning techniques for predicting neuropathic pain in patients with cancer: A systematic review. Digit Health. 2025;11:20552076251358315. 239. Takahashi T, Kobayashi Y, Sakurai R, Matsuoka K, Akatsuka J, Kisu I, et al. A Systematic Review of the Application of Artificial Intelligence in Colposcopy: Diagnostic Accuracy for Cervical Intraepithelial Neoplasia and Cervical Cancer. Clin Med Insights Oncol. 2025;19:11795549251374908. 240. Takamatsu M. Transforming histologic assessment: artificial intelligence in cancer diagnosis and personalized treatment. Br J Cancer. 2025. 241. Tang C, Xu Z, Duan H, Zhang S. Advancements in artificial intelligence for ultrasound diagnosis of ovarian cancer: DR. MEHRDAD FARROKHI 290 a comprehensive review. Front Oncol. 2025;15:1581157. 242. Tang X, Zhou H, Liu Y, Gao S, Zhou Y. Diagnostic performance of the ultrasound -based artificial intelligence diagnostic system in predicting cervical lymph node metastasis in patients with thyroid cancer: A systematic review and meta-analysis. Sci Prog. 2025;108(2):368504251346906. 243. Tarhini AA, Dave P, El Naqa I. General artificial intelligence for the diagnosis and treatment of cancer: the rise of foundation models. BJR Artif Intell. 2025;2(1):ubaf015. 244. Tayyil Purayil AL, Joseph RM, Raj A, Kooriyattil A, Jabeen N, Beevi SF, et al. Role of Artificial Intelligence in MRI-Based Rectal Cancer Staging: A Systematic Review. Cureus. 2024;16(12):e76185. 245. Thalambedu N, Balla M, Sivasubramanian BP, Sadaram P, Malla KP, Vasipalli KP, et al. Integrating artificial intelligence with circulating tumor DNA for non-small cell lung cancer: opportunities, challenges, and future directions. Front Med (Lausanne). 2025;12:1612376. 246. Tiwari A, Ghose A, Hasanova M, Faria SS, Mohapatra S, Adeleke S, et al. The current landscape of artificial intelligence in computational histopathology for cancer diagnosis. Discov Oncol. 2025;16(1):438. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 291 247. Toumaj S, Heidari A, Jafari Navimipour N. Leveraging explainable artificial intelligence for transparent and trustworthy cancer detection systems. Artif Intell Med. 2025;169:103243. 248. Tun HM, Rahman HA, Naing L, Malik OA. Artificial intelligence utilization in cancer screening program across ASEAN: a scoping review. BMC Cancer. 2025;25(1):703. 249. Uchikov P, Khalid U, Dedaj-Salad GH, Ghale D, Rajadurai H, Kraeva M, et al. Artificial Intelligence in Breast Cancer Diagnosis and Treatment: Advances in Imaging, Pathology, and Personalized Care. Life (Basel). 2024;14 )11(. 250. Urso L, Badrane I, Manco L, Castello A, Lancia F, Collavino J, et al. The Role of Radiomics and Artificial Intelligence Applied to Staging PSMA PET in Assessing Prostate Cancer Aggressiveness. J Clin Med. 2025;14 )10(. 251. Usmani S, Al Riyami K, Kheruka S, Numani SP, Al Sukaiti R, Ahmed M, et al. Deep learning (DL)- based advancements in prostate cancer imaging: Artificial intelligence (AI)-based segmentation of (68)Ga-PSMSA PET for tumor volume assessment. Precis Radiat Oncol. 2025;9(2):120-32. 252. Utti V, Bikia V, Agarwal AA, Daneshjou R. Integrating Artificial Intelligence in Dermatological Cancer Screening and Diagnosis: Efficacy, Challenges, and Future Directions. Annu Rev Biomed Data Sci. 2025;8(1):591-603. DR. MEHRDAD FARROKHI 292 253. Uwimana A, Gnecco G, Riccaboni M. Artificial intelligence for breast cancer detection and its health technology assessment: A scoping review. Comput Biol Med. 2025;184:109391. 254. Vardasca R, Mendes JG, Magalhaes C. Skin Cancer Image Classification Using Artificial Intelligence Strategies: A Systematic Review. J Imaging. 2024;10 )11(. 255. Vasquez VM, Jr., McCabe M, McKee JC, Siby S, Hussain U, Faizuddin F, et al. Transforming Cancer Care: A Narrative Review on Leveraging Artificial Intelligence to Advance Immunotherapy in Underserved Communities. J Clin Med. 2025;14 )15(. 256. Verma H, Mistry S, Jayam KV, Shrestha P, Adkins L, Liang M, et al. Artificial Intelligence in Oncology: A 10-Year ClinicalTrials.gov-Based Analysis Across the Cancer Control Continuum. Cancers (Basel). 2025;17 )21(. 257. Vidiyala N, Parupathi P, Sunkishala P, Muppavarapu CS, Gujja A, Kanagala P, et al. Artificial intelligence: a new era in prostate cancer diagnosis and treatment. Int J Pharm. 2025;683:126024. 258. Vinay V, Jodalli P, Chavan MS, Buddhikot CS, Luke AM, Ingafou MSH, et al. Artificial Intelligence in Oral Cancer: A Comprehensive Scoping Review of Diagnostic and Prognostic Applications. Diagnostics (Basel). 2025;15 )3(. ARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 293 259. Viviana Cortlana M, Kennedy Itodo B, Yan Leyfman M, Jenna Ghazal B, Vraj JigarKumar Rangrej M, Adhith T, et al. Artificial Intelligence in Cancer Care: Addressing Challenges and Health Equity. Oncology (Williston Park). 2025;39(3):105-10. 260. Vyas A, Kumar K, Sharma A, Verma D, Bhatia D, Wahi N, et al. Advancing the frontier of artificial intelligence on emerging technologies to redefine cancer diagnosis and care. Comput Biol Med. 2025;191:110178. 261. Farrokhi M, Taheri F, Moeini A, Farrokhi M, Rabiei S, Farahmandsadr M, et al. Sex and gender differences in the pathogenesis and treatment of diseases. Kindle. 2023;3(1):1-168. 262. Karimian S, Taheri F, Farrokhi M, Farrokhi M, Bayat Z, Zadeh SAM, et al. Digital Health and Wearable Technologies. Kindle. 2024;4(1):1-240. 263. Taheri F, Farrokhi M, Fard AM, Rahmani E, Soltani R, Shamsedanesh S, et al. Role of Micronutrients and Nutrition in Prevention and Treatment of Cancers. Kindle. 2023;3(1):1-102. 264. Farrokhi M, Taheri F, Farrokhi M, Tosi YEK, Ghadirzadeh E, Bagheri M, et al. Nanomedicine: Technologies and Applications. Kindle. 2024;4(1):1-196. 265. Motavaselian M, Farrokhi M, Khouzani PJ, Fard AM, Daeizadeh F, Pourrahimi M, et al. Diagnostic performance of ultrasonography DR. MEHRDAD FARROKHI 294 for identification of small bowel obstruction; a systematic review and meta-analysis. Archives of Academic Emergency Medicine. 2024;12(1):e33. 266. Rahmani E, Amani-Beni R, Hekmatnia Y, Yaseri AF, Ahadiat S-A, Boroujeni PT, et al. Diagnostic accuracy of ultrasonography for detection of intussusception in children; a systematic review and Meta-Analysis. Archives of Academic Emergency Medicine. 2023;11(1):e24. 267. Farrokhi M, Moeini A, Taheri F, Farrokhi M, Khodashenas M, Babaei M, et al. AI-assisted Screening and Prevention Programs for Diseases. Kindle. 2023;3(1):1-209. 268. Rahmani E, Bayat Z, Farrokhi M, Karimian S, Zahedpasha R, Sabzehie H, et al. Monkeypox: a comprehensive review of virology, epidemiology, transmission, diagnosis, prevention, treatment, and artificial intelligence applications. Archives of Academic Emergency Medicine. 2024;12(1):e70. 269. Shayganfar A, Farrokhi M, Shayganfar S, Ebrahimian S. Associations between bone mineral density, trabecular bone score, and body mass index in postmenopausal females. Osteoporosis and sarcopenia. 2020;6(3):111-4. 270. Farrokhi M, Yarmohammadi B, Mangouri A, Hekmatnia Y, Bahramvand Y, Kiani M, et al. Screening performance characteristics of ultrasonography in confirmation of endotracheal intubation; a systematic review and metaARTIFICIAL INTELLIGENCE FOR HARD-TO-TREAT AND U... 295 analysis. Archives of Academic Emergency Medicine. 2021;9(1):e68. 271. Rahmani E, Fayyazishishavan E, Afzalian A, Varshochi S, Amani-Beni R, Ahadiat S-A, et al. Point-of-care ultrasonography for identification of skin and soft tissue abscess in adult and pediatric patients; a systematic review and meta-analysis. Archives of Academic Emergency Medicine. 2023;11(1):e49. 272. Kazemi S-M, Khorram R, Fayyazishishavan E, Amani-Beni R, Haririan Y, Khameneh SMH, et al. Diagnostic Accuracy of Ottawa Knee Rule for Diagnosis of Fracture in Patients with Knee Trauma; a Systematic Review and Meta-analysis. Archives of Academic Emergency Medicine. 2023;11(1):e30. DR. MEHRDAD FARROKHI 296 Proof