Healthcare 4.0: Requirements and Challenges for E2E Slicing & Orchestration

Healthcare 4.0: Requirements and Challenges for E2E Slicing & Orchestration Khushboo Sahu∗⋄, Lechosław Tomaszewski∗, Agapi Mesodiakaki⋄, Amalia Miliou⋄ ∗Orange Polska, Warsaw, Poland ⋄Aristotle University of Thessaloniki, Thessaloniki, Greece Abstract—The evolution from beyond 5G to 6G networks has envisioned the adoption of various healthcare 4.0 applications, including telemedicine, e-Health, robotic surgery, and many more. These applications require stringent KPIs, including ultra-low latency, massive device connectivity, reliability, and robust security, which are not feasible in the traditional network architecture. The softwarisation of networks by SDN and NFV has enabled the logical partitioning of the physical infrastructure across the RAN, CN, and TN domains as an E2E network slice for user-specific services, which are tailored, isolated, managed, and orchestrated across virtual networks using Zero-Touch Management (ZTM). This paper presents the KPI requirements for healthcare 4.0 use cases, a brief survey on recent AL/ML-based slicing and orchestration frameworks, and then discusses challenges and future research directions for automated healthcare network slices. It also discusses open research directions towards ZTM of slices, including semantic intent translation specific to domains, cross-domain observability of QoS, AI-enabled closed-loop assurance, and the integration of emerging 6G enablers. It aims to guide the healthcare services in a scalable, flexible, secure, and dynamic way over mobile networks. Index Terms—Healthcare 4.0, KPI, E2E Slicing, Management and Orchestration, QoS, ZTM, AI, ML. I. INTRODUCTION Advances in network technology are transforming clinical and patient care in the healthcare area. The information and communications technology (ICT) promises to provide ontime and secure medical assistance from healthcare experts in a personalised and ubiquitous way. The emerging medical technologies, such as telepresence, robotic telesurgery, connected ambulances, and many more, play a crucial role in it [1]–[3]. These require reliable and secure network connectivity with high bandwidth and near real-time data transmission, having minimal delay to perform tasks promptly and efficiently. The beyond-5th Generation (5G) provides network services of Enhanced Mobile Broadband (eMBB), Ultra-Reliable LowLatency Communication (URLLC), and Massive Machine Type Communications (mMTC). As per the medical use cases, the requirements regarding the Key Performance Indicator (KPI) vary and demand multiple services simultaneously. So, This work received funding from Horizon Europe under the Marie Sklodowska Curie actions: ELIXIRION (rEaLIzing healthcare 4.0 eXploIting the 6G netwoRk evolutION) (GA 101120135). Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union. Neither the European Union nor the granting authority can be held responsible for them. obtaining an E2E Network Slice (NS) that can fulfil the KPIs with good Quality of Service (QoS) is essential. Network slicing is a key concept in 5G technology. The NS enables the logical partitioning of the communication infrastructure tailored to the demands of the use case. Its key enablers are the softwarisation of the network utilising Software-Defined Networking (SDN) and Virtual Network Function (VNF). The NSs are managed and orchestrated across the network in four levels, namely by the Communication Service Management Function (CSMF), Network Slice Management Function (NSMF), Network Slice Subnet Management Function (NSSMF), and then Network Function Management Function (NFMF). The CSMF obtains the service-specific needs of the user from the Operations Support System (OSS)/Business Support System (BSS), and translates them into the domain-specific configurations for NSMF. Then Network Slice Instance (NSI) is created and managed by NSMF in cooperation with NSSMF of each domain. Each NSI consists of a Network Slice Subnet Instance (NSSI) from each domain, which its respective NSSMF manages. Thereafter, NFMF provides management at the Network Functions (NFs) level. At the same time, the network orchestrator coordinates with management functions to automate and manage various services and resources across the domain, maintaining KPI demands and ensuring Service Level Agreement (SLA) [4]. Here, we contribute to accumulating important KPIs for healthcare applications, such as tele-robotic surgery, connected ambulances and remote triage, in-hospital pervasive monitoring, holographic/eXtended Reality (XR), wearables, pointof-care imaging assisted with Artificial Intelligence (AI) and autonomous mobile services. Thereafter, a brief survey is conducted on AI/Machine Learning (ML) enabled management and orchestration. At last, a discussion about the challenges and future research direction for scalable, secure, and automated slicing and Zero-Touch Management (ZTM). The organisation of this manuscript is as follows: Section II describes the related work concerning E2E NS for healthcare, and Section III presents the KPI requirements for different healthcare use cases. Following this, Sections IV and V discuss the current AI/ML-enabled Management and Orchestration (MANO) framework for E2E NS, challenges, and future research directions for healthcare applications. Finally, Section VI concludes the findings and discusses open research challenges. Fig. 1. Healthcare use-cases and interaction with network II. RELATED WORK Researchers are continually involved in developing clinical use-case-specific network slicing applications to enhance healthcare services. Smart hospital appliances and healthcare applications require strict communication protocols for realtime patient care and operations. To accommodate these services seamlessly, medical network slicing for different use cases leveraging mobile edge computing, and heterogeneous network management is essential [3]. The cross-vertical application with horizontal E2E NS requirements, such as remote robotic surgery and real-time patient monitoring, is analysed [5]. Their simulation results show that the hospital acting as a tenant needs strict admission control and slicing policy coordination in domain-level orchestration. Bandwidth utilisation is enhanced while reducing network connection blockage for E2E slicing. However, [6] introduces a Transport Network (TN)-aware E2E NS orchestration framework, presenting the dynamic resource mapping over all the domains of Radio Access Network (RAN), TN, and Core Network (CN), according to the KPI requirements of the verticals. It also highlights the AI-assisted admission control for services and dynamic QoS adaptations according to their performances. These massive slice services require efficient management with minimal human intervention. The E2E slicing, with efficient and dynamic resource allocation methods, also requires seamless orchestration across vendors and multi-domain boundaries to meet the required KPIs. However, on-demand slicing is necessary, especially for critical clinical applications. So, leveraging European Telecommunications Standards Institute (ETSI) Network Function Virtualisation (NFV) MANO and real-time slice instantiation, the CURATE model in [7] presents dynamic slicing and slice lifecycle management as an SLA compliant and cost-effective service for healthcare emergencies. In addition to the network itself, both management and orchestration frameworks should be scalable, which, in the case of a decentralised health sector composed of units independently acting as tenants of NSs communication services, will be of particular importance as a factor multiplying the number of NSs. Scalable management and orchestration approaches may include embedding the management layer in deployed NSs [8]. Recent model-based approaches for network slice management and NFV placement provide foundational work for Healthcare 4.0 orchestration. Multi-tier heterogeneous network analysis under DDoS attacks reveals network selection strategies relevant for secure healthcare infrastructures, as discussed in [9]. Similarly, for bandwidth-intensive medical applications, optimal video bitrate selection and an edge caching mechanism in Multi-access Edge Computing (MEC)-empowered, sliceenabled networks have been presented in [10]. [11] presents a meta-heuristic approach for VNF scheduling to minimise service creation time for stringent healthcare latency requirements. In contrast, [12] demonstrates that demand-aware cooperative content caching strategies support data-intensive clinical scenarios across heterogeneous cellular networks. III. KPIS FOR HEALTHCARE 4.0 USE-CASES Healthcare 4.0 imposes stringent QoS requirements for critical and heterogeneous applications, as per the clinical use-cases. These KPIs consist of ultra-low end-to-end delay (from sub-ms to 100 ms), extremely high reliability (≥99.999%), high throughput (from Mbps to Gbps), very low jitter, and robust security and privacy compliant with General Data Protection Regulation (GDPR) and healthcare regulations [13]. Here, Table I shows consolidated KPIs for some use-cases, such as – mandatory sub-5 ms E2E latency for tele-robotic surgery, 99.999% reliability for connected ambulance, in-hospital pervasive monitoring, wearables or remote monitoring, holographic or XR, point-of-care imaging assisted with AI, and autonomous mobile service robots. These applications require precise and quantitative values of KPIs for designing and managing an E2E NSs. It ensures use-casespecific, tailored, and isolated virtual network resource allocation across domains. Also, by incorporating these KPIs into NS descriptor and orchestration policy, AI/ML-enabled ZTM system can dynamically monitor, reconfigure, adjust, scale, secure and allocate resources to maintain SLA compliance and optimise performance while guaranteeing patient safety and data confidentiality [14]. A. Mapping KPIs for E2E-Orchestration Achieving a good QoS while having all KPIs requires well-coordination and domain-specific resource management and orchestration across RAN, TN, CN, and Edge/Cloud, as shown in figure 1. The following describes domain-specific descriptions of how the KPIs are mapped and enforced to get an E2E slicing and management. 1) Radio Access Network (RAN): RAN provides wireless connectivity to the healthcare applications or end-user devices. RAN enables logical partitioning of radio resources, and maps the KPIs into a slice descriptor for RAN-specific slice services. According to requirements, RAN-domain allocates dedicated radio spectrum, prioritises and performs deterministic scheduling, and supports dynamic resource allocation while isolating it from other interfering services and making it secure. It also applies slice-specific security, encryption, and authentication for sensitive patient data over the air interface [21]. 2) Transport Network (TN): TN is responsible for connecting RAN, CN, and Edge/Cloud resources to transmit a large volume of data with a strict performance guarantee. It consists of fronthaul, midhaul, and backbone links. TN-aware slice requirements are mapped into transport links, preventing overprovisioning resources while assuring E2E performance. TN allows dynamic wavelength allocation, optical Virtual Private Networks (VPNs) for multi-Gbps throughputs, low delay and jitter, and a fast re-routing mechanism to ensure high service availability. An E2E encryption and network traffic isolation prevent data leakage during transmission [22]. 3) Core Network (CN): CN handles NS-specific service function chaining, session management, policy enforcement, and inter-domain coordination. Single – Network Slice Selection Assistance Information (S-NSSAI) parameter directs incoming flow to a dedicated AMF, SMF, or UPF instance to ensure minimum user plane latency and packet loss. The Network Data Analytics Function (NWDAF) predicts the KPI drift for closed-loop automation. Then, it also manages NS instantiation, resource allocation, and lifecycle management, translating high-level healthcare intents into actionable configurations across all domains [21], [22]. 4) Edge/Cloud: Edge and cloud resources provide localised or distributed processing, analytics, and storage, reducing latency while supporting scalable data-intensive healthcare applications. NS orchestration dynamically places workloads at the edge/cloud as per the real-time KPI requirements while optimising latency, throughput, and privacy of NSI [22]. B. Slice Lifecycle Management (LCM) As shown in figure 1, each NSI has its lifecycle composed of four stages: preparation, commissioning, operation, and decommissioning. During the first stage, NS design, onboarding, and environment preparation are done, as well as admission control and resource allocation across the domain based on the usages. Then, a blueprint description defines resource requirements, network functions, topologies, and policies. Network orchestrators reserve the required resources on virtual and physical networks across all the domains. Then the commissioning stage is responsible for instantiating an NS, configured as per the QoS policies, and provides security and service chaining. In the operation stage, the slice gets activated to handle the user traffic. Thereafter, this slice performance is constantly monitored across domains, checking for any change in the policy to ensure that the requested performance metrics are maintained. It also ensures closedloop assurance utilising the AI/ML-based methods for the NS lifecycle. As per the demand, VNFs are migrated, resources are scaled, and configured. It also handles deactivation of these NSIs at the tenant’s or the Mobile Network Operator (MNO) discretion. During the decommissioning, the allocated resources and VNFs resources are released, which are not shared with other NSIs, and are cleaned up. IV. ZERO-TOUCH ORCHESTRATION AND AI/ML-ENABLED MANAGEMENT A. Zero-Touch Management (ZTM) Paradigm The ZTM represents a complete autonomous and selfmanageable network solution without direct human intervention. It is a critical paradigm for Healthcare 4.0, and essential for managing complex and dynamic healthcare NSs requirements on-demand. ZTM utilises AI/ML-based decision algorithms to automate E2E execution while maintaining human validations for safety-critical operations. The primary responsibility of ZTM is to automate the translation of high-level clinical intents, in this case for healthcare application/tenants, into technical slice-level configurations, ML-based workload prediction, dynamic resource allocation, fault detection, selfhealing without service disruption, while providing closedloop assurance [19]. B. AI/ML Paradigms for Healthcare 4.0 Slicing For ZTM, AI/ML-based algorithms are essential for autonomous, adaptive, predictive, scalable, and explainable control over NSs. It optimises dynamic resource allocation, predicts KPI violations, and automates decision-making across a multi-domain or cross-domain orchestration stack. The following describes the list of present AI/ML paradigms: TABLE I LIST OF KPIS FOR HEALTHCARE 4.0 APPLICATIONS Use-cases Latency Reliability Throughput Jitter Security Other KPIs Reference Tele-robotic surgery ≤5 ms E2E latency ≥99.999% ≥1 Gbps ≤50 µs Slice isolation ≈10 km surgeonpatient distance [15], [16] Connected ambulance & Remote triage ≤10 ms ≥99.999% 0.5-1 Gpbs uplink ≤2 ms E2E IPsec, and dynamic anonymisation 120 km/h mobility support [15], [17] In-hospital pervasive monitoring ≤250 ms ≥99% ≈1 Mbps per patient ≤25 ms Device attestation and zerotouch key rotation Connection density ≈104[18] Wearables or Remote monitoring ≤500 ms (alarm ≤50 ms) ≥99% 10 kbps – 1 Mbps per device ≤30 ms E2E encryption; federated analytics Battery life ≥1 year, indoor coverage guarantee [6], [18] Holographic, XR ≤5 ms ≥99.999% 1-4 Gbps ≤1 ms Digital rights management, watermarking, and integrity proofs multi-user playback [15], [16] AI-assisted Point-ofCare Imaging ≤50 ms ≥99.99% ≥1 Gbps ≤5 ms Slice-aware digital imaging and communication transfer, data-loss prevention Energy efficiency ≤ 5 J per GB [17], [19] Autonomous Mobile Service Robots (hospital logistics) ≤20 ms ≥99.999% 100 Mbps ≤1 ms Zero-trust operational technology segmentation ≤10 cm localisation error [15], [20] 1) Reinforcement Learning (RL) & Deep Reinforcement Learning (DRL): RL has agents that make decisions on a trialand-error basis, receiving either positive or negative rewards for their actions. Here, RL or DRL interacts with the dynamic environment of the network and learns to allocate resources optimally, including admission control and slice management policies. [20] and [23] have used multi-agent DRL for RAN and CN resource allocations and slice admission. It achieved high QoS and cost efficiency in vehicular and healthcare networks. The DRL also optimises the placement of VNFs according to changes in network traffic and KPI values. 2) Federated Learning (FL): FL-based approach utilises distributed training of an AI model, which only shares learned vectors instead of raw data. The FL-based methods are used mostly for distributed learning across RAN, TN, CN, and edge domains. It allows local network data analytics and predicts network requirements while preserving users’ privacy and security. Statistical FL-based analytics are used to predict the slice-level KPIs and required resources. It also predicts the possibility of SLAs violations and minimises energy savings for decentralised computing [16]. FL agents coordinate with different domains for resource management, providing scalability and privacy for beyond 5G networks. 3) Graph Neural Network (GNN): GNNs ability to learn from complex, non-Euclidean graph-structured data makes them well-suited for getting the complex relationships between network nodes, links, and service requirements in slicingenabled architectures. It rapidly emerged for modelling, optimising, and automating the network. GNNs can accurately predict E2E NS performance metrics such as latency, reliability, and throughput by learning from the topology and operational state of the network [24]. Then it has been used to estimate hidden nodes, optimise resource allocation, and solve NP-hard combinatorial problems, for dynamic networks utilising MEC [25]. It also optimises the placement of VNFs, routing network function, policy, and security enforcement on the networks. 4) Supervised Learning (SL): SL is a branch of ML where models get labelled datasets for training and learn from input features to target outputs. Deep Neural Networks (DNNs) is a subset of SL, consisting of multiple layers of interconnected neurons to capture complex, non-linear relationships in a dataset. For automated network slicing and orchestration [26], SL has been widely used for network traffic classification, resource allocation, and E2E NS closed-loop automation. 5) Explainable AI (XAI): The ML models are used as “black boxes”, making it difficult for network operators, regulators, and vertical stakeholders to trust, audit, or certify automated decisions. Therefore, XAI are being actively researched to enhance transparency in healthcare slice management, providing auditable and interpretable insights for admission control, KPI violation analysis, and regulatory compliance. For instance, attribution-based methods can identify the most influential features within a dataset that drive the model’s decisions, thereby facilitating trust and accountability in AIenabled network operations. Its variant is a confidence metricbased attribution method, quantifying the dataset attribute’s faithfulness of the used ML-models. The ML-models also use predefined constraints during the training for explanationguided learning. Therefore, XAI is also used in closed-loopbased requirement prediction across domains with proper reasoning. Then the inference-causal-based models provide auditable and interpretable reasons for SLA and regulatory compliance [27]. 6) Multi-Agent Systems and Distributed Optimisation: The Multi-agent RL-based dynamic resource allocation while optimising the communication overhead over time is used in [28]. The base station is considered an independent RL agent. It manages spectrum allocation and computes resources for vehicular and healthcare scenarios. V. E2E SLICE MANAGEMENT CHALLENGES FOR HEALTHCARE 4.0 These diverse healthcare applications and Internet of Medical Things (IoMT) devices may require dynamic or massive E2E slicing for hospitals, patients/doctors, and connected ambulances. These slices are automated and orchestrated across network domain technologies. However, challenges remain to achieve automated and dynamic slicing satisfying the KPIs with good QoS. Some of these challenges are as follows: 1) E2E Resource Allocation with QoS: Tele-robotic surgery and holographic/ XR based applications have stringent KPI requirements such as ultra-low latency, high reliability, high throughput, and strict privacy. These heterogeneous requirements are challenging to guarantee across multi-domain, multivendor networks. Especially when these network domains coexist and are managed by different administrators. Hence, efficient resource allocation and orchestration of resources across multiple domains becomes difficult under dynamic traffic loads and service demands. 2) Privacy and Security: Transmitting sensitive medical or patient data across domains while ensuring privacy and compliance is another challenge, particularly in a multiadministrator and distributed environment. Therefore, rigorous NS isolation guarantees are necessary for healthcare and other services, primarily to protect against data leakage and performance degradation. NSs also poses security threats from inter-slices and unauthorised access, necessitating robust and adaptive security requirements. 3) Scalability: Healthcare applications have various deployments that should support a heterogeneous and massive number of devices. These applications can demand resources according to the variation in traffic. Hence, an efficient utilisation of the present resources with low power consumption is required, without over-sizing the current capability. An E2E slicing utilises AI-enabled hierarchical and distributed orchestration methods to scale slicing and management functions across multiple domains within different administration boundaries. However, it remains challenging to scale and dynamically allocate resources in real-time across various domains while reducing the CAPEX and OPEX of MNOs. 4) Interoperability: The network domains share administrative responsibilities with multiple vendors and operators. Making trustworthy coordination and cooperation for slicing among the operators is difficult, especially for NS LCM. NSI’s performance is continuously monitored and deployed in the best condition concerning QoS, energy consumption, and overall cost. To achieve an E2E NS within multiple technological, administrative, and owner domains requires standardised interfaces, a shared information model, and interoperable orchestration Application Programming Interfaces (APIs). A. Future Research Directions for ZTM These challenges are faced for scalable, reliable, secure E2E slicing for various healthcare applications. To address these and make efficient on-demand slicing across all the domains, this paper discusses its research directions as follows: 1) Semantic Intent Translation for Clinical KPIs: Present network orchestrators provide a predefined set of network slice templates according to the ETSI-NFV standard. The translation of high-level service requirements or tenants’ intents into network-specific configurations remains challenging. It becomes particularly critical when these applications require a stringent network performance guarantee. Retrieving configuration parameters for cross-domain management (e.g, E2E latency ≤5 ms) into slice-specific, cross-domain policies is needed. However, it happens on an ad-hoc basis, which is slow and prone to error. So, models based on GNN or Large Language Model (LLM) can be utilised for retrieving crossdomain intent language decomposition of service parameters. 2) Cross-domain Observability and Coordination: A wellcoordinated management of the network slices and their resources requires domain-specific observability and adaptation in real-time. The closed-loop automation of a slice demands real-time input of telemetry data from RAN, TN, CN, and edge. Due to vendor-specific control over resources and privacy policy, data sharing becomes difficult. However, distributed analytics-based, FL-based methods are being adopted to address it [29]. However, the stability and scalability of nested closed loops are still required to handle massive NSs concurrently. The inter-loop slice interference creates KPI changes and SLA violations as well [19]. Therefore, hierarchical and game-theory-based loop admission and instantiation could be utilised to ensure the stability of the respective slice. 3) Trustworthy and Explainable AI/ML for Slice Automation: The current method, based on Federated Learning or reinforcement learning-based methods, is being used as a black-box tool to achieve the desired performance through model tuning. They meet the desired expectation; however, this creates trust issues among industry verticals [30], especially in the healthcare sector. So, causal-inference-driven network orchestration promises to provide the reasons for the industry verticals. Then, a counterfactual explanation for policy changes can be explored, and online confidence scores can be created to get human interventions on this automation. Moreover, if required, it can withdraw its agreement [19]. 4) Energy-aware Slicing and ZTM: Incorporation of softwarised network and AI/ML-based models for automation of ZTM comes with a cost of utilisation of compute, storage, and energy resources. [16] proposed energy-aware FL model for RAN saving 10 times more energy. However, optimisation of energy utilisation along with KPI fulfilment across the domain still needs to be addressed. VI. CONCLUSION This paper presents an aggregated contribution for the precise KPIs for healthcare 4.0 use cases, spanning telerobotic surgery, connected ambulances, and other applications. It maps them across the RAN, transport, core, and edge domains. Then we further analysed the integration of AI/ML paradigms, such as RL, DNN, and FL, for NS optimisation, intent translation, and closed-loop assurance for E2E slicing and orchestration. Our findings emphasise that healthcare NS orchestration is a matter of performance optimisation and an interdisciplinary coordination of latency, reliability, privacy, and trust. Our findings emphasise that healthcare NS orchestration is a matter of performance optimisation and a multidisciplinary coordination of latency, reliability, privacy, and trust. While recent frameworks show promising gains in predictive orchestration, intelligent network function placement, and KPI compliance, key challenges regarding semantic intent expression, federated assurance, explainable AI, and cross-domain interoperability remain unresolved. This research contributes towards the AI/ML-native network slicing in 5G-Advanced and 6th Generation (6G) systems, highlighting the rising need for trustworthy, adaptive networks which can autonomously guarantee clinical-grade service delivery and safety, ultimately paving the way for real-time, resilient, and patient-aware digital healthcare ecosystems. ACKNOWLEDGMENT This work received funding from Horizon Europe under the Marie Skłodowska Curie Actions: ELIXIRION (GA 101120135). REFERENCES [1] M. Wang, H. Ji, M. Jia, Z. Sun, J. Gu, and H. Ren, “Method and application of information sharing throughout the emergency rescue process based on 5G and AR wearable devices,” Scientific Reports, vol. 13, no. 1, p. 6353, 2023. [2] M. William, S. Sharif, and W. Ejaz, “Enabling Communication and Networking Technologies for 6G in Healthcare Sector,” in 2023 Second International Conference On Smart Technologies For Smart Nation (SmartTechCon), 2023, pp. 699–704. [3] G. Cisotto, E. Casarin, and S. Tomasin, “Requirements and Enablers of Advanced Healthcare Services over Future Cellular Systems,” IEEE Communications Magazine, vol. 58, no. 3, pp. 76–81, 2020. [4] 3GPP, “Management and orchestration; Concepts, use cases and requirements,” 3GPP, Technical Standard TS 28.530, ver. 19.0.0, Mar. 2025. [5] S. Wijethilaka, P. Porambage, C. de Alwis, and M. Liyanage, “A Comprehensive Analysis on Network Slicing for Smart Hospital Applications,” in 2022 IEEE 19th Annual Consumer Communications & Networking Conference (CCNC), 2022, pp. 276–279. [6] P. Iovanna, M. Svensson, A. Shapin, G. Bottari, F. Ubaldi, F. Ponzini, and M. Puleri, “End-to-end network slicing orchestration – A Key Enabler for Industry-Vertical use Cases,” Ericsson Technology Review, vol. 2022, no. 2, pp. 2–10, 2022. [7] L. Sanabria-Russo, J. Serra, D. Pubill, and C. Verikoukis, “CURATE: On-Demand Orchestration of Services for Health Emergencies Prediction and Mitigation,” IEEE Journal on Selected Areas in Communications, vol. 39, no. 2, pp. 438–445, 2021. [8] S. Kukli´ nski and L. Tomaszewski, “DASMO: A scalable approach to network slices management and orchestration,” in NOMS 2018 - 2018 IEEE/IFIP Network Operations and Management Symposium, 2018, pp. 1–6. [9] A. Tsiota, D. Xenakis, N. Passas, and L. Merakos, “Multi-Tier HetNets With Random DDoS Attacks: Service Probability and User Load Analysis,” IEEE Transactions on Information Forensics and Security, vol. 20, pp. 6190–6204, 2025. [10] D. Xenakis, “To DASH, or Not to DASH? Optimal Video Bitrate Selection and Edge Network Caching in MEC-Empowered Slice-Enabled Networks,” IEEE Transactions on Vehicular Technology, vol. 73, no. 4, pp. 5556–5571, 2024. [11] A.-S. Charismiadis, D. Tsolkas, N. Passas, D. Xenakis, and L. Merakos, “Metaheuristics as enablers for VNF scheduling in the network slice set up process,” Journal of Communications and Networks, vol. 24, no. 6, pp. 742–753, 2022. [12] T. Mihretu Ayenew, D. Xenakis, L. Alonso, N. Passas, and L. Merakos, “Demand-Aware Cooperative Content Caching in 5G/6G Networks With MEC-Enabled Edges,” IEEE Networking Letters, vol. 4, no. 3, pp. 118– 122, 2022. [13] 3GPP, “Service requirements for the 5G system,” 3GPP, Technical Standard TS 22.261, ver. 20.3.0, Jun. 2025. [14] A. A. Abdellatif, A. Abo-Eleneen, A. Mohamed, A. Erbad, N. V. Navkar, and M. Guizani, “Intelligent-slicing: An AI-assisted network slicing framework for 5G-and-beyond networks,” IEEE Transactions on Network and Service Management, vol. 20, no. 2, pp. 1024–1039, 2023. [15] I. Afolabi, T. Taleb, K. Samdanis, A. Ksentini, and H. Flinck, “Network Slicing & Softwarization: A Survey on Principles,” Enabling Technologies & Solutions, p. 24, 2018. [16] H. Chergui, L. Blanco, L. A. Garrido, K. Ramantas, S. Kukli´ nski, A. Ksentini, and C. Verikoukis, “Zero-Touch AI-Driven Distributed Management for Energy-Efficient 6G Massive Network Slicing,” IEEE Network, vol. 35, no. 6, pp. 43–49, 2021. [17] A. A. Barakabitze, A. Ahmad, R. Mijumbi, and A. Hines, “5G network slicing using SDN and NFV: A survey of taxonomy, architectures and future challenges,” Computer Networks, vol. 167, p. 106984, 2020. [18] S. Wijethilaka and M. Liyanage, “Survey on network slicing for Internet of Things realization in 5G networks,” IEEE Communications Surveys & Tutorials, vol. 23, no. 2, pp. 957–994, 2021. [19] E. Coronado, R. Behravesh, T. Subramanya, A. Fern` andez-Fern` andez, M. S. Siddiqui, X. Costa-P´ erez, and R. Riggio, “Zero Touch Management: A Survey of Network Automation Solutions for 5G and 6G Networks,” IEEE Communications Surveys & Tutorials, vol. 24, no. 4, pp. 2535–2578, 2022. [20] S. Jiang, J. Zheng, F. Yan, and S. Zhao, “Reinforcement-Learning-Based Network Slicing and Resource Allocation for Multi-Access Edge Computing Networks,” IEEE Transactions on Cognitive Communications and Networking, vol. 10, no. 3, pp. 1132–1145, 2024. [21] W. Rafique, J. Rani Barai, A. O. Fapojuwo, and D. Krishnamurthy, “A Survey on Beyond 5G Network Slicing for Smart Cities Applications,” IEEE Communications Surveys & Tutorials, vol. 27, no. 1, pp. 595–628, 2025. [22] M. A. Habibi, B. Han, A. Fellan, W. Jiang, A. G. S´ anchez, I. L. Pavon, A. Boubendir, and H. D. Schotten, “Toward an Open, Intelligent, and End-to-End Architectural Framework for Network Slicing in 6G Communication Systems,” IEEE Open Journal of the Communications Society, vol. 4, pp. 1615–1658, 2023. [23] Y. Ji, Y. Wang, H. Zhao, G. Gui, H. Gacanin, H. Sari, and F. Adachi, “Multi-Agent Reinforcement Learning Resources Allocation Method Using Dueling Double Deep Q-Network in Vehicular Networks,” IEEE Transactions on Vehicular Technology, vol. 72, no. 10, pp. 13 447– 13 460, 2023. [24] H. Wang, Y. Wu, G. Min, and W. Miao, “A Graph Neural NetworkBased Digital Twin for Network Slicing Management,” IEEE Transactions on Industrial Informatics, vol. 18, no. 2, pp. 1367–1376, 2022. [25] A. Asheralieva, D. Niyato, and Y. Miyanaga, “Efficient Dynamic Distributed Resource Slicing in 6G Multi-Access Edge Computing Networks With Online ADMM and Message Passing Graph Neural Networks,” IEEE Transactions on Mobile Computing, vol. 23, no. 4, pp. 2614–2638, 2024. [26] T. Zhang, Y. Bian, Q. Lu, J. Qi, K. Zhang, H. Ji, W. Wang, and W. Wu, “Supervised Learning Based Resource Allocation with Network Slicing,” in 2020 Eighth International Conference on Advanced Cloud and Big Data (CBD), 2020, pp. 25–30. [27] S. Roy, H. Chergui, and C. Verikoukis, “Toward Bridging the FL Performance-Explainability Tradeoff: A Trustworthy 6G RAN Slicing Use-Case,” IEEE Transactions on Vehicular Technology, vol. 73, no. 7, pp. 10 529–10 538, 2024. [28] Y. Cui, H. Shi, R. Wang, P. He, D. Wu, and X. Huang, “Multi-Agent Reinforcement Learning for Slicing Resource Allocation in Vehicular Networks,” IEEE Transactions on Intelligent Transportation Systems, vol. 25, no. 2, pp. 2005–2016, 2024. [29] K. Khalafi and N. Lu, “Network Slicing for Edge-Cloud Orchestrated Networks via Online Convex Optimization,” in IEEE INFOCOM 2024 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2024, pp. 1–6. [30] A. A. Abdellatif, N. Mhaisen, A. Mohamed, A. Erbad, and M. Guizani, “Reinforcement Learning for Intelligent Healthcare Systems: A Review of Challenges, Applications, and Open Research Issues,” IEEE Internet of Things Journal, vol. 10, no. 24, pp. 21 982–22 007, 2023.