Artificial Intelligence in Radiology

Year: 2025 Volume: 3 Issue: 1 10.5281/zenodo.15733908 Artificial Intelligence in Radiology REVIEW ARTICLE Mete Özdikici¹ 1. Department of Radiology, Taksim Training and Research Hospital, Istanbul, Turkey Abstract Abstract In radiology and undoubtedly in radiologic anatomy, artificial intelligence (AI) plays an important role in the diagnostic and therapeutic processes by offering revolutionary innovations in the field of imaging. AI algorithms enhance accuracy, speed, and efficiency by analyzing images obtained through methods such as X-rays, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasound (US). Specifically, deep learning (DL) and machine learning (ML) applications enable early diagnosis by detecting fine details. This not only ensures patient safety but also alleviates the workload of radiologists and improves cost-effectiveness. Furthermore, AI offers safer imaging opportunities by reducing radiation doses and improving low-quality images. However, limitations such as data quality, ethical concerns, and patient privacy complicate the integration of AI into healthcare systems. In the future, AI is expected to expand its applications in radiology, offering more accurate diagnostic and therapeutic possibilities. Keywords: Radiology, artificial intelligence, machine learning, deep learning, radiomics ulusmedj.com Published by Ulus Medical Journal. Review ArticleReview Article How to cite: Özdikici M. Artificial Intelligence in Radiology. Ulus Med J. 2025;3(1):9-28. Received: 29 March 2025 Revised: 25 April 2025 Accepted: 1 May 2025 Published: 16 July 2025 ORCID ID of the author(s): M.Ö: 0000-0001-6309-8306 Correspondence Author Mete Özdikici, MD, PhD. Department of Radiology, Taksim Training and Research Hospital, Istanbul, Turkey e-mail [email protected] 2980-1907 /© 2025 Ulus Medical Journal. Published by Unico's Medicine. This is an open-access article under the terms of the CC BY license. (https://creativecommons.org/licenses/by/4.0/) Ulus Med J. 10 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 Introduction The earliest computers were machines designed to perform specific mathematical operations and predictable tasks. These machines did not possess the reasoning and analytical capabilities of humans. A significant milestone in this field was Alan M. Turing's seminal 1950 paper titled “Computing Machinery and Intelligence ”(1). The term “Machine Learning” was first introduced in 1959 by Arthur Samuel to describe algorithms that enable computers to learn without being explicitly programmed. Through this innovation, computers learned how to play checkers (2). Artificial Intelligence (AI) is driving a revolutionary transformation in medicine, particularly impacting the field of radiology. In the medical context, AI refers to the development and application of algorithms and software techniques that can analyze medical data, learn from it, and optimize and improve their performance based on accumulated experience—ultimately aiming to enhance patient safety and reduce workload (3). AI plays a critical role in optimizing image quality, accelerating image acquisition, and predicting both disease prognosis and treatment responses. To promote interdisciplinary research and bridge the gap between academia and industry, a strong synergy is needed between radiologists and AI developers. Collaboration between radiologists and AI developers can ensure that AI tools align more closely with clinical needs and effectively bridge the gap between theory and practice (4). Radiology encompasses medical imaging techniques that are crucial for disease diagnosis and treatment planning. Traditionally, radiologists diagnose diseases using modalities such as X-rays, computed tomography (CT), magnetic resonance imaging (MRI), and ultrasonography (US) (5). AI algorithms enhance the accuracy, speed, and efficiency of image analysis, thereby reducing the radiologists' workload (4). This review article explores the contributions and potential of AI in radiology under various subheadings. TERMINOLOGY RADIOLOGY: This field is divided into two main categories: diagnostic radiology and radiation therapy. Radiology involves the use of X-rays and other imaging technologies in medical diagnosis and treatment. These methods include radiography, mammography, CT, MRI, and US. Diagnostic use involves detecting diseases through medical imaging, while therapeutic use enables minimally invasive surgical procedures (5). RADIOLOGICAL ANATOMY: A medical discipline focused on examining body structures and organs through radiologic imaging methods and correlating these images with anatomical features (5). ARTIFICIAL INTELLIGENCE (AI): AI refers to the ability of computer software to perform tasks that typically require human intelligence—such as perception, recognition, learning, reasoning, inference, decision-making, planning, problem-solving, and communication (6). It emphasizes the creation of intelligent machines that operate and respond like humans (4). ALGORITHM: A set of steps to solve a problem. Algorithms provide instructions to computers for deriving answers or performing tasks, which is especially useful when a precise solution is unattainable or data analysis needs to be expedited. They offer guidance and direction for AI systems (4). Ulus Med J. 11 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 MACHINE LEARNING (ML): A subfield of AI that enables computers to learn from experience and acquire knowledge through examples, allowing them to adapt and enhance performance over time. Programming languages like Python are often used in ML applications (7,8). DEEP LEARNING (DL): A revolutionary advancement in AI and a more specialized branch of ML. DL enables machines to perform tasks like classification, object recognition, speech recognition, and language translation with minimal error, using data such as sound, images, signals, and text (9). DL algorithms are developed using software such as MATLAB for tasks like object detection and recognition. These methods rely on neural network architectures. One of the most widely used forms of deep neural networks is the Convolutional Neural Network (CNN or ConvNet), which extracts features directly from images (10,11). ROBOTICS: A technology domain associated with robots and an important field within AI. In the future, humans may possess auxiliary limbs or extensions to assist in physically challenging tasks, effectively becoming “cyborgs.” The term “cyborg” is short for “cybernetic organism” (12). HOW DOES THE HUMAN BRAIN WORK? Since AI research often focuses on analyzing human cognitive processes to develop similar artificial mechanisms, it is helpful to first understand how the human brain functions. The human brain weighs approximately 1,250–1,500 grams and consists of around 100 billion neurons. The two hemispheres of the brain differ both physically and functionally. When compared to a computer, the right hemisphere functions like a parallel processor, while the left hemisphere operates more like a serial processor. These two hemispheres are connected by a bundle of approximately 300 million nerve fibers (axons) called the corpus callosum. Although both hemispheres process information, they do so differently, leading to distinct cognitive styles. The right hemisphere focuses on the present moment—it perceives your current location and immediate actions. For instance, while reading these lines, your awareness of light, temperature, and smell is processed through the right hemisphere. On the other hand, the left hemisphere is concerned with the past and the future. It captures the details of the current experience perceived by the right hemisphere, analyzes them further, and integrates them with past experiences to project into the future. Generally, the right hemisphere is associated with creativity, emotional understanding, art, and music, whereas the left hemisphere is more dominant in analytical thinking, language, logic, and mathematics (13). CATEGORIES OF ARTIFICIAL INTELLIGENCE Today, AI systems are generally categorized into four main types (4,14,17): Reactive Machines: These AI systems respond to immediate stimuli. Examples include IBM’s Deep Blue, which defeated world chess champion Garry Kasparov in 1997, and Google’s AlphaGo, which beat Go Ulus Med J. 12 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 champion Lee Sedol in 2016. Reactive machines cannot learn from past experiences and can only respond to current inputs. As such, they form a significant portion of machine learning systems. Limited Memory Machines: These systems can learn from historical data, but only store it temporarily and lose it over time. Applications include autonomous driving systems, traffic signal control, and online messaging platforms. Theory of Mind Machines: These AI systems possess the capability to communicate with humans and encompass all characteristics of the previous categories. They can understand human emotions and thoughts, engage in social interactions, and may even be applied in therapeutic contexts in medicine. Virtual assistants on smartphones are gradually evolving toward this level by adapting to users’ individual needs. Self-Aware AI: This represents a more advanced form of Theory of Mind AI. Self-awareness is a cognitive trait typically developed in human childhood. For an AI to attain self-awareness, it must develop consciousness. Such AI would be capable of understanding abstract concepts and reasoning about them. A “CENTRAL ARTIFICIAL INTELLIGENCE” CONTROLLING ALL ROBOTS AND SOFTWARE THE DARKEST SCENARIO: A GOD-LIKE AI When contemplating self-aware AI, one of the first images that comes to mind is a consciousness without emotions or the capacity to forget. This often evokes depictions like the humanoid robots from the Terminator franchise. However, the belief that AI requires a physical body is one of the most fundamental misconceptions. Imagine an AI that can improve its own software without human awareness. This entity, driven by a relentless desire to learn and evolve, would not be hindered by a physical form. Existing solely in data, such a being could access the entirety of human history and potentially use that knowledge to subjugate or eliminate humanity. If such a system were to emerge suddenly, the likelihood of it causing millions of deaths within seconds is disturbingly high. This could potentially be a scenario leading to the end of civilization. Currently, AI products marketed under this label are successful only in specific, narrow domains. There is no existing system that replicates all dimensions of human intelligence, nor is such a development expected in the near future (16,17). HISTORY OF ARTIFICIAL INTELLIGENCE (1,2,14,15) • 1950: Alan M. Turing published his seminal paper “Computing Machinery and Intelligence.” • 1956: The term “artificial intelligence” (AI) was coined at the Dartmouth Conference, marking the beginning of modern AI research. • 1959: Arthur Samuel was the first to use the term “machine learning” (ML). • 1966: Joseph Weizenbaum developed ELIZA, the first simple chatbot to interact with humans. Ulus Med J. 13 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 • 1970: The first computer-assisted automatic electrocardiogram (ECG) interpretation was implemented. • 1972: MYCIN, one of the first expert systems in medicine, was developed to offer diagnosis and treatment recommendations for bacterial infections. • 1973: The British government ceased AI funding, initiating a period known as the “AI winter.” • 1980: The era of expert systems began, and decision support systems became widespread in fields like medicine, engineering, and finance. • 1980: CASNET (Causal Associational Network) was developed for diagnosing eye diseases. • 1980s: Computer-Aided Detection (CAD) systems were first introduced for medical image analysis. • 1988: AI began automatically detecting peripheral lung lesions. • 1990s: Early applications of AI in robotic surgery and medical imaging emerged; the Da Vinci Surgical System was developed. • 1997: IBM’s Deep Blue defeated world chess champion Garry Kasparov. • 2000: CAD systems were implemented in mammography for breast cancer screening. • 2004: CAD systems were introduced for lung cancer detection. • 2010s: Deep learning (DL) algorithms became widely used for detecting tumors, lesions, and anomalies in radiological images. • 2011: IBM’s Watson AI was used for cancer diagnosis and treatment. • 2011: IBM Watson gained widespread attention by defeating human contestants on the game show Jeopardy!. • 2012: Success of DL algorithms surged; Google Brain gained the ability to recognize cats after analyzing millions of YouTube videos. • 2012: Brain image segmentation and tumor grading were accomplished using DL. • 2016: Google DeepMind’s AlphaGo program defeated the world champion in the game of Go. • 2016: DL became common in medical imaging; DeepMind was applied for detecting eye diseases. • 2017: Stanford University compared AI systems to human radiologists in tumor detection tasks. • 2020: Artificial neural networks were employed for ECG interpretation. • 2020: AI began detecting lung lesions related to COVID-19, leading to the development of fully autonomous diagnostic systems in radiology. • 2023: With GPT-4, AI models gained the ability to process not only text but also images and video, advancing multimodal capabilities. APPLICATIONS AND CONTRIBUTIONS OF ARTIFICIAL INTELLIGENCE IN RADIOLOGY AI demonstrates notable expertise in a variety of tasks such as detecting thromboembolic infarcts or hemorrhages in the brain, segmentation, classification, and identifying large vessel occlusions. It plays a critical role in the early detection of neurodegenerative disorders such as Alzheimer’s and Parkinson’s disease. Its potential in predicting post-operative outcomes for brain and spinal surgeries is promising (3,4,18). Ulus Med J. 14 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 ML algorithms can combine perfusion data from MRI with coronary anatomy from CT to create sophisticated 3D heart models, thereby improving the detection of cardiac ischemia and facilitating more precise procedural planning (4). DL algorithms are increasingly used in tumor detection and classification, particularly in diagnosing breast, lung, and prostate cancers. They can distinguish between benign and malignant lesions and various tumor types. For example, AI has shown the ability to classify lung nodules on CT scans and accurately differentiate subtypes of renal cell carcinoma on MRI—often rivaling the expertise of experienced radiologists. From pre-treatment CT images, AI can extract meaningful data to predict survival rates in lung cancer patients. Similarly, radiomic features derived from MRI scans have shown correlation with recurrence risk in glioblastoma patients. The integration of radiomics introduces new quantitative metrics to radiology reports, enhancing the detection and characterization of both focal lesions and diffuse diseases in the liver and pancreas, potentially leading to improved clinical outcomes (4,19,20). In fact, AI’s applications in radiology are extensive (16) and are summarized in Table 1. Table 1. Artificial Intelligence Applications in Radiology FIELD APPLICATIONS Emergency Radiology Detection of intracranial hemorrhages, large vessel occlusions, fractures, free abdominal fluid, small bowel obstruction, intussusception detection Head and Neck Radiology Segmentation of lesions and anatomical structures, localization and classification of lesions, segmentation and classification of lymph nodes Neuroradiology Evaluation of brain anatomy, segmentation of cortical and subcortical structures, lesion detection, stroke and hemorrhage detection, aneurysm and degeneration detection Chest Radiology Detection of lung nodules and tumors, pneumonia, pneumothorax, emphysema, rib fractures, pulmonary embolism detection, diagnosis of obstructive lung disease Cardiovascular Radiology Coronary calcium scoring, coronary angiography, fractional flow reserve, plaque analysis, left ventricular myocardium analysis, myocardial infarction diagnosis, prognosis of coronary artery disease, cardiac function evaluation, and cardiomyopathy diagnosis and prognosis Breast Radiology Lesion detection, classification and characterization, breast density estimation, characterization of mammographic abnormalities Abdominal Radiology Segmentation of liver and spleen, segmentation of adrenal and urogenital structures, lesion detection and characterization, free intraperitoneal air, vertebral compression fractures, aortic dissection Musculoskeletal Radiology Detection of fractures in proximal humerus, hand, wrist and foot, detection of hip osteoarthritis, quantitative bone imaging for bone strength and quality assessment Oncologic Imaging Tumor segmentation and characterization, differentiation of benign-malignant lesions and pathological lymph nodes Ulus Med J. 15 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 Contributions of Artificial Intelligence (AI) in Radiology (14,16,17): 1. Radiation Dose Reduction: AI algorithms enable safer imaging procedures for patients by reducing radiation exposure. 2. Image Quality Enhancement: AI algorithms can improve noisy or low-quality images. Compared to radiologists, they can offer higher accuracy rates and detect details that may otherwise be overlooked. Deep Learning (DL) techniques are capable of identifying even very subtle abnormalities. 3. Image Analysis and Diagnosis: AI algorithms utilize DL techniques to analyze and interpret medical images. 4. Disease Prediction and Monitoring: By analyzing findings in medical images, AI algorithms can predict disease progression and assist in personalized treatment planning. This enables patients to receive more effective and timely interventions. 5. Objectivity and Standardization: AI reduces variability in radiological assessments between different radiologists, thereby making the evaluation process more objective. 6. Reduction of Radiologists’ Workload: AI algorithms automate routine tasks and generate automatic reports, allowing radiologists to focus on more complex cases and reducing their overall workload. 7. Rapid Results and Cost Reduction: While traditional imaging methods can be time-consuming, AI accelerates this process, offering significant advantages in cases that require urgent diagnosis. Automated analysis ensures faster results, reduces costs, and enhances the cost-effectiveness of healthcare services. 8. Continuous Learning: AI algorithms can be updated with new data, improving their performance over time. This results in increasingly accurate diagnoses and better outcomes. Limitations of Artificial Intelligence (6,16,17,21): 1. Data Quality and Quantity: The effectiveness of AI algorithms depends on the quality and size of the datasets used. Inaccurate or incomplete data can lead to incorrect outcomes. Moreover, collecting and processing large datasets can be a challenging and resource-intensive task. 2. Legal and Ethical Issues: The accuracy, transparency, and accountability of AI-generated results can be questioned. In addition, ethical concerns such as patient privacy and data security must be addressed. Machine Learning Artificial intelligence (AI) has various subfields similar to medical specialties, including machine learning (ML) and deep learning (DL) (22). ML, a subset of AI, focuses on developing algorithms that autonomously learn from data. These algorithms reference pre-labeled datasets to learn, improve over time, and evaluate new data accordingly (4). Effective ML models require extensive and high-quality datasets for training (17). The major advantage of ML lies in its ability to handle large datasets that are beyond human analytical capabilities and to identify quantitative features with ease (2). ML incorporates a variety of algorithms and techniques that enable learning from data. At its core are two main approaches: supervised learning and unsupervised learning. • Supervised learning relies on labeled input-output pairs from training datasets. The objective is to formulate a function that accurately maps inputs to outputs and can predict new cases reliably. Key algorithms include linear regression, logistic regression, and decision trees. Ulus Med J. 16 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 • Unsupervised learning, on the other hand, autonomously explores data to identify patterns or relationships without predefined labels. It helps reveal inherent data structures and generate insights for complex problems. Well-known methods include k-means clustering, hierarchical clustering, and principal component analysis (PCA) (2,4,7). Among ML approaches, deep learning stands out for its clinical potential. While other ML methods often require manual data annotation, DL is self-learning—it can train directly from raw data without needing pre-defined features. Only the categorization of raw inputs and outcomes is needed (7,15). Deep Learning DL is one of the most widely used methods in medical image analysis. With diverse neural network architectures, DL is especially prevalent in healthcare applications such as medical imaging (10,11). It is a subfield of ML, which in turn is a subfield of AI (see Figure 1). Compared to classical ML techniques, DL trains on larger datasets and typically delivers higher performance. DL systems can automatically perform complex classification tasks and feature extraction using multi-layered artificial neural networks (three or more layers) (4,17). Figure 1. Artificial Intelligence, Machine Learning and Deep Learning Diagram There is no universal or optimal method that guarantees the best results in ML or DL. Because each technique addresses different problems and performance metrics, direct comparisons can be misleading (5). DL-based computer-aided diagnosis (CAD) systems often outperform traditional CAD tools and have demonstrated diagnostic accuracy comparable to that of radiologists (4,20). Thanks to powerful graphics processing units (GPUs) and the availability of big data, DL algorithms utilize deep neural networks with parallel computation capabilities, allowing multiple operations to run simultaneously. This parallelism significantly reduces processing time and provides advantages for complex image analysis. Prominent DL models include: Ulus Med J. 17 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 • Convolutional Neural Networks (CNNs), frequently used for tasks such as segmentation and classification. • Generative Adversarial Networks (GANs), used in image synthesis and enhancement (4,14). Convolutional Neural Networks (CNNs) One of AI's most significant applications in radiology is image classification, which distinguishes normal from pathological findings using CNNs. CNNs are currently the most effective models for image analysis and classification. These deep, layered neural networks are also the most commonly used DL methods. They require large, labeled image datasets to perform optimally (4,10). CNNs deconstruct images into pixels, extract important features (e.g., edges, shapes), and learn to recognize complex structures. They can automatically detect pathologies such as tumors or lesions, making them suitable for complex radiological analyses (4). Instead of standard matrix multiplication, CNNs use convolution operations at specific layers. They are especially well-suited for visual, auditory, and textual pattern recognition tasks and contribute significantly to computer vision and natural language processing (NLP) (4). CNNs require multiple layers to extract meaningful features. Their structure includes: • Feature extraction layers: input layer, convolution + activation, and pooling. • Classification layers: fully connected layer and output layer. For example (23), a 9×9 input matrix combined with a 3×3 filter would yield a 7×7 output: (9−3)+1 = 7 • The convolutional layer (or transformation layer) identifies visual features. • The pooling layer reduces image dimensions while preserving essential characteristics. • The fully connected layer uses these features for classification (10). GENERATIVE ADVERSARIAL NETWORKS (GANs) GANs are utilized for expanding image datasets, obtaining high-resolution images, and transferring textures/patterns from one image to another (10). To illustrate, imagine a counterfeiter (generator) producing fake currency while the police (discriminator) try to detect the counterfeit bills. Over time, the police become better at distinguishing the fakes, while the counterfeiter improves the realism of the counterfeit money. This iterative process continues until the generator produces images so realistic that the discriminator can no longer differentiate them from real ones. This architecture, composed of two neural networks engaged in continuous competition, is referred to as a Generative Adversarial Network (GAN) (5). LABELING, SEGMENTATION, AND CLASSIFICATION In radiological imaging, labeling refers to annotating specific structures, while segmentation involves delineating the boundaries of these structures. Classification determines whether these structures are normal or pathological. Segmentation, also known as partitioning or delineation, refers to dividing an image into meaningful regions that exhibit distinct features. In this process, labels are generated for each pixel, and predictions are made based on these labels to derive insights. Typically, segmentation is performed first to detect tumors or lesions, followed by classification of their types. Convolutional Neural Networks (CNNs) are particularly useful at this stage (10). Several improvements to CNN architectures have enabled more effective segmentation. For example, Long et al. introduced Fully Convolutional Networks (FCNs) (24), which were later adapted by Ronneberger Ulus Med J. 24 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 cancer treatments and the management of chronic diseases (18,28,34). Another important contribution of AI is its application in remote healthcare services (telemedicine). Globally, especially in rural and developing regions, there are significant inequalities in access to qualified healthcare services. AI-assisted radiology systems can facilitate access to expert radiologists for patients in such areas, contributing to a more equitable healthcare provision on a global scale. The importance of remote diagnosis and treatment processes has become even more apparent during crises such as pandemics. In such scenarios, AI can enhance the flexibility and resilience of healthcare systems (14,15,33). Advancements in image processing algorithms have enabled the use of AI not only in diagnostic processes but also in treatment procedures. AI-assisted systems are increasingly being used in the field of robotic surgery. For example, robotic surgical systems such as "CyberKnife" perform operations with lower error rates thanks to AI algorithms. These systems enhance surgeons’ manual dexterity and enable safer, more precise surgeries using minimally invasive techniques. In the future, it may even be possible to have fully autonomous robots operating in surgical settings (12,18). However, there are several challenges to the widespread adoption of AI in radiology and the broader medical field. Chief among these are issues related to data quality and data security. For AI algorithms to generate accurate and reliable results, access to large amounts of high-quality data is essential. Moreover, the privacy and security of patient data pose critical ethical concerns. Regulatory frameworks must be established regarding data sharing and usage, and ethical and legal principles must be clearly defined. Additionally, the “black box” nature of AI systems, referring to the lack of transparency in decisionmaking processes, including medical, legal, and financial responsibilities, can undermine the trust of clinicians and patients in AI technologies (6,16,21). In conclusion, the opportunities offered by AI in radiology and medicine have the potential to shape the future of healthcare services. Its advantages in improving diagnostic and therapeutic accuracy, saving time, and enabling personalized medicine make AI an indispensable tool. Nevertheless, issues such as data quality, ethical considerations, and transparency must be addressed with care. A multidisciplinary approach is essential for the effective and reliable implementation of AI, necessitating collaboration among healthcare professionals, engineers, ethicists, legal experts, insurance providers, and policymakers. Through such collaboration, the full potential of AI can be realized, ushering in a new era in healthcare services. ABBREVIATIONS AI: Artificial Intelligence ANN: Artificial Neural Networks CT: Computed Tomography CAD: Computer-Aided Detection CNN: Convolutional Neural Networks DL: Deep Learning ECG: Electrocardiography FCN: Fully Convolutional Networks GAN: Generative Adversarial Networks GPU: Graphics Processing Unit GUI: Graphical User Interface ML: Machine Learning MRI: Magnetic Resonance Imaging NLP: Natural Language Processing Ulus Med J. 25 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 RNN: Recurrent Neural Networks US: Ultrasonographyt SOME USEFUL LINKS https://acikveri.saglik.gov.tr/ https://altair.com/altair-rapidminer https://deepcognition.ai/ https://developer.nvidia.com/digits https://www.editverse.com/tr/the-human-connectome-project-brain-mapping/ https://elsevier.health/en-US/marketing/radiologists/pinpoint-complex-and-common-radiology-caseswith-statdx https://www.geminibilgi.com.tr/statdx-654.html https://keras.io/ https://www.lunit.io/en https://www.mathworks.com/products/matlab.html?s_tid=hp_ff_p_matlab https://oxipit.ai/products/chesteye/ https://pyradiomics.readthedocs.io/en/latest/# https://www.python.org/ https://www.qure.ai/ https://www.radiantviewer.com/ https://radiology.healthairegister.com/ https://www.slicer.org/ https://teleradyoloji.saglik.gov.tr/ https://www.tensorflow.org/?hl=tr Declaration of interest: The authors report no conflicts of interest. Funding source: No funding was required Ethical approval: No need for reviews. Acknowledgments: No Peer-review: Evaluated by independent reviewers working in at least two different institutions appointed by the field editor. Data availability The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request. Ulus Med J. 26 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 Contributions Research concept and design: MÖ Data analysis and interpretation: MÖ Collection and/or assembly of data: MÖ Writing the article: MÖ Critical revision of the article: MÖ Final approval of the article: MÖ References 1. Turing AM. Computing machinery and intelligence. Mind. 1950;LIX(236):433-60. 2. Yılmaz MÖ, Aydın SO, Baran O. Nöroşirürjide büyük veri ve yapay zekâ: Görüntüleme uygulamaları [Big Data and Artificial Intelligence in Neurosurgery: Imaging Applications]. Türk Nöroşir Derg. 2022;32(2):142-49. Turkish 3. Yapay zekâ [Artificial intelligence] [Internet]. Erişim adresi [access address]: https://tr.wikipedia.org/ wiki/Yapay_zekâ. 4. Najjar R. Redefining radiology: A review of artificial intelligence integration in medical imaging. Diagnostics. 2023;13:2760. 5. Atlan F, Pençe İ. Yapay zekâ ve tıbbi görüntüleme teknolojilerine genel bakış [An Overview of Artificial Intelligence and Medical Imaging Technologies]. Acta Infologica. 2021;5(1):207-30. Turkish 6. Özdemir L, Bilgin A. Sağlıkta yapay zekânın kullanımı ve etik sorunlar [The Use of Artificial Intelligence in Healthcare and Ethical Issues]. SHYD. 2021;8(3):439-45. Turkish 7. Janssen NE. A machine learning proposal for predicting the success rate of IT-projects based on project metrics before initiation. University of Twente; 2020. 8. Demšar J, Curk T, Erjavec A, et al. Orange: Data mining toolbox in Python. J Mach Learn Res. 2013;14:2349-53. 9. Sayar B. Tıp alanında yapay zekânın kullanımı [The Use of Artificial Intelligence in Medicine]. Acta Medica Ruha. 2023;1(1):27-33.Turkish 10. Eker AG, Duru N. Medikal görüntü işlemede derin öğrenme uygulamaları [Deep Learning Applications in Medical Image Processing]. Acta Infologica. 2021;5(2):1-16. Turkish 11. Er MB. Önceden eğitilmiş derin ağlar ile göğüs röntgeni görüntüleri kullanarak pnömoni sınıflandırılması [Classification of Pneumonia Using Chest X-Ray Images with Pretrained Deep Neural Networks]. Konjes. 2021;9(1):193-204. Turkish 12. Robotik [Robotics]. [Internet]. Erişim adresi [access address]: https://tr.wikipedia.org/wiki/Robotik. 13. İnsan beyni [Human Brain]. [Internet]. Erişim adresi [access address]: https://tr.wikipedia.org/wiki/ İnsan_beyni. 14. Pianykh OS, Langs G, Dewey M, et al. Continuous learning AI in radiology: implementation principles and early applications. Radiology. 2020;297:6-14. 15. Yu KH, Beam AL, Kohane IS. Artificial intelligence in health care. Nat Biomed Eng. 2018;2(10):719. Ulus Med J. 27 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 16. Yordanova MZ. The applications of artificial intelligence in radiology: Opportunities and challenges. Eur J Med Health Sci. 2024;6(2):11-4. 17. Waller J, O’Connor A, Raafat E, et al. Applications and challenges of artificial intelligence in diagnostic and interventional radiology. Pol J Radiol. 2022;87. 18. Bakas S, Reyes M, Jakab A, et al. Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the BRATS challenge. 2018. 19. Koçak B, Durmaz EŞ, Ateş E, Kılıçkesmez Ö. Radiomics with artificial intelligence: a practical guide for beginners. Diagn Interv Radiol. 2019;25:485-95. 20. Huang X, Shan J, Vaidya V. Lung nodule detection in CT using 3D convolutional neural networks. In: Proceedings of the IEEE 14th International Symposium on Biomedical Imaging (ISBI); 2017 Apr 1821; Melbourne, Australia. New York: IEEE; 2017:379-383. 21. Özçiftçi S, Akpınar A, Dönmez O. Tıp etiği araştırmalarında Q metodolojisi kullanımı: Radyoloji alanında yapay zekâ etiği araştırması örneği [Use of Q Methodology in Medical Ethics Research: An Example of Artificial Intelligence Ethics Study in the Field of Radiology]. Lokman Hekim Dergisi. 2024;14(2):41829. Turkish 22. Ozlu C, Acet A, Korkut B, Yalcin C. Artificial intelligence studies and data analysis in chronic lymphocytic leukemia: A current review. Selcuk Med J. 2024;40(3):146-51. 23. Vatansever S, Bıyıklıoğlu HF. Sağlık görüntüleme sistemlerinde görüntü işleme ile hastalık teşhisi [Disease Diagnosis Using Image Processing in Medical Imaging Systems]. Yıldız Teknik Üniversitesi; 2021.Turkish 24. Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR); 2015 Jun 7-12; Boston, MA, USA. New York: IEEE; 2015:3431-40. 25. Ronneberger O, Fischer P, Brox T. U-net: Convolutional networks for biomedical image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer; 2015. 26. Zhou Z, Siddiquee MMR, Tajbakhsh N, Liang J. Unet++: A nested u-net architecture for medical image segmentation. In: Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. Cham: Springer; 2018:3-11. 27. Alom MZ, Hasan M, Yakopcic C, et al. Recurrent residual convolutional neural network based on u-net (r2u-net) for medical image segmentation. arXiv preprint arXiv:1802.06955. 2018. 28. Ibtehaz N, Rahman MS. MultiResUNet: Rethinking the U-Net architecture for multimodal biomedical image segmentation. Neural Netw. 2020;121:74-87. 29. Sun J, Zhang J, Wen Y, et al. SAUNet: Shape attentive U-Net for interpretable medical image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer; 2020. 30. Tong X, Chen M, Nie D, et al. ASCU-Net: Attention gate, spatial and channel attention U-Net for skin lesion segmentation. Diagnostics (Basel). 2021;11(3):501. 31. Li C, Sun S, Liu W, et al. MRFU-Net: A multiple receptive field U-Net for environmental microorganism image segmentation. In: Information Technology in Biomedicine. Cham: Springer; 2021:27-40. 32. Han C, Hayashi H, Rundo L, et al. GAN-based synthetic brain MR image generation. In: Proceedings of Ulus Med J. 28 AI in Radiology Özdikici M. Ulus Med J. 2025;3(1):9-28 the IEEE 15th International Symposium on Biomedical Imaging (ISBI); 2018 Apr 4-7; Washington, DC, USA. New York: IEEE; 2018:734-38. 33. Gillies RJ, Kinahan PE, Hricak H. Radiomics: Images are more than pictures, they are data. Radiology. 2016;278:563-77. 34. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational radiomics system to decode the radiographic phenotype. Cancer Res. 2017;77. 35. Szczypiński PM, Strzelecki M, Materka A, Klepaczko A. MaZda—a software package for image texture analysis. Comput Methods Programs Biomed. 2009;94:66-76. 36. Nioche C, Orlhac F, Boughdad S, et al. LIFEx: A freeware for radiomic feature calculation in multimodality imaging to accelerate advances in the characterization of tumor heterogeneity. Cancer Res. 2018;78:4786-89. 37. Zhang L, Fried DV, Fave XJ, et al. IBEX: An open infrastructure software platform to facilitate collaborative work in radiomics. Med Phys. 2015;42:1341-53. 38. Williams GJ. Rattle: A data mining GUI for R. The R Journal. 2009;1:45-55. 39. Witten IH, Frank E, Hall MA, Pal CJ. Data mining: Practical machine learning tools and techniques. 4th ed. San Francisco: Morgan Kaufmann Publishers Inc; 2016. 40. Shayesteh SP, Alikhassi A, Fard Esfahani A, et al. Neo-adjuvant chemoradiotherapy response prediction using MRI-based ensemble learning method in rectal cancer patients. Phys Med. 2019;62:111-19. 41. Serhatlıoğlu S, Hardalaç F. Yapay Zekâ teknikleri ve radyolojiye uygulanması [Artificial Intelligence Techniques and Their Application in Radiology]. Fırat Tıp Dergisi. 2009;14(1):1-6. Turkish 42. Mannil M, von Spiczak J, Muehlematter UJ, et al. Texture analysis of myocardial infarction in CT: Comparison with visual analysis and impact of iterative reconstruction. Eur J Radiol. 2019;113:245-50. 43. Liu Y, Chen J, Hai J, et al. Three-dimensional semi-supervised lumbar vertebrae region of interest segmentation based on MAE pre-training. J Xray Sci Technol. 2025;33(1):270-282. Publisher's Note: Unico's Medicine remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.