Medical ML: Comprehensive Framework for ML-Based Medical Imaging Diagnosis — Ukrainian Implementation Guide

Posted on by






Comprehensive Framework for ML-Based Medical Imaging Diagnosis: A Ukrainian Implementation Guide


Comprehensive Framework for ML-Based Medical Imaging Diagnosis: A Ukrainian Implementation Guide

By Oleh Ivchenko, PhD Candidate | Odesa National Polytechnic University | Stabilarity Hub | February 11, 2026

Abstract

This paper presents the UMAID Framework (Ukrainian Medical AI Deployment) — a comprehensive, evidence-based implementation guide for machine learning-based medical imaging diagnosis systems tailored specifically for the Ukrainian healthcare context. Synthesizing insights from 30 prior research articles spanning international best practices, technical architectures, clinical workflow integration, and regulatory requirements, this framework addresses the unique challenges facing Ukrainian healthcare institutions, including infrastructure limitations, regulatory alignment with both EU standards and Ukrainian MHSU requirements, and integration with the national eZdorovya digital health system.

The UMAID Framework is structured around six foundational principles — Fairness, Universality, Traceability, Usability, Robustness, and Explainability (FUTURE) — adapted for Ukrainian implementation through 42 actionable best practices organized across four implementation phases. Validation through expert review with Ukrainian radiologists and pilot planning with ScanLab diagnostic centers demonstrates the framework’s practical applicability. The framework provides healthcare administrators, radiologists, and IT professionals with a structured roadmap for deploying medical AI systems that meet international quality standards while respecting local constraints and maximizing diagnostic accuracy for Ukrainian patient populations.

1. Introduction

The global deployment of artificial intelligence in medical imaging has reached an inflection point. As of early 2026, the U.S. FDA has authorized over 1,200 AI-enabled medical devices, with radiology comprising approximately 75% of all authorizations (ACR Data Science Institute, 2026). The European Union’s AI Act has established the world’s most comprehensive regulatory framework for high-risk AI applications in healthcare, while China has deployed AI diagnostic systems across thousands of hospitals serving hundreds of millions of patients.

Yet a stark implementation gap persists: while technology advances rapidly, 81% of hospitals worldwide report zero operational AI deployments in clinical workflows (Spatharou et al., 2020). This gap is particularly acute in emerging healthcare markets, where resource constraints, regulatory uncertainty, and infrastructure limitations create unique barriers to adoption.

📊 The Ukrainian Context

Ukraine’s healthcare system has demonstrated remarkable resilience, with the eZdorovya platform processing nearly 5 billion medical data records by the end of 2025 and receiving the GovTech Award 2025. However, AI-specific diagnostic implementations remain nascent, creating both challenges and opportunities for systematic deployment.

This paper addresses the need for a comprehensive, actionable implementation framework specifically designed for Ukrainian healthcare institutions. Drawing on systematic analysis of international experiences — including successes at NHS AI Lab, lessons from China’s massive deployment, and failures across multiple health systems — the UMAID Framework provides healthcare leaders with a structured approach to navigating the complex landscape of medical AI deployment.

1.1 Framework Objectives

The UMAID Framework is designed to achieve five primary objectives:

  1. Clinical Safety: Ensure patient outcomes improve or remain unchanged, with robust mechanisms for detecting and preventing AI-induced errors
  2. Regulatory Compliance: Meet requirements across Ukrainian MHSU regulations, EU Medical Device Regulation (MDR), and international standards
  3. Workflow Integration: Enable seamless integration with existing PACS infrastructure, RIS systems, and the national eZdorovya platform
  4. Economic Viability: Demonstrate positive return on investment within 24-36 months for typical Ukrainian diagnostic centers
  5. Scalability: Support expansion from pilot implementations to regional and national deployment

2. Theoretical Foundations

The UMAID Framework synthesizes principles from multiple international guidelines while adapting them for the Ukrainian context. The theoretical foundation rests on three pillars: the FUTURE-AI consensus principles (BMJ, 2025), WHO guidance on AI ethics and governance (WHO, 2024), and the Joint Commission’s Responsible Use of AI in Healthcare (RUAIH) framework.

2.1 Existing Frameworks Comparison

Framework Origin Focus Ukrainian Applicability Key Gap
FUTURE-AI International Consortium (BMJ) Trustworthy AI development High (principles) No implementation guidance
WHO AI Ethics World Health Organization Ethical governance Medium Limited technical specificity
FDA AI/ML Framework United States Regulatory pathway Low (US-specific) Not applicable to Ukraine
EU AI Act European Union Risk-based regulation High (EU alignment) Complex for small institutions
RUAIH Joint Commission/CHAI Hospital implementation Medium US-centric assumptions
NHS AI Playbook United Kingdom Practical deployment Medium NHS-specific infrastructure

The UMAID Framework addresses identified gaps by providing Ukraine-specific implementation guidance that bridges high-level principles with practical deployment steps, incorporating lessons learned from both successful and failed implementations internationally.

3. Framework Development

3.1 Development Methodology

The UMAID Framework was developed through a three-stage process:

  1. Literature Synthesis: Systematic review of 30 research articles covering global best practices, technical architectures, clinical integration, and Ukrainian healthcare context
  2. Expert Consultation: Interviews with Ukrainian radiologists (n=12), healthcare IT specialists (n=8), and regulatory affairs professionals (n=4) across five regions
  3. Pilot Design: Collaboration with ScanLab diagnostic network to develop implementation specifications for initial deployment

3.2 Framework Overview

flowchart TD
    subgraph Foundation["🏛️ FOUNDATION LAYER"]
        F1[Fairness
Demographic Equity]
        F2[Universality
Broad Applicability]
        F3[Traceability
Audit Trail]
        F4[Usability
Clinical Workflow Fit]
        F5[Robustness
Reliability Under Variation]
        F6[Explainability
Clinical Interpretability]
    end
    
    subgraph Components["⚙️ COMPONENT LAYER"]
        C1[Governance
Structure]
        C2[Technical
Infrastructure]
        C3[Clinical
Protocols]
        C4[Training &
Change Mgmt]
        C5[Monitoring &
QA]
    end
    
    subgraph Phases["📋 IMPLEMENTATION PHASES"]
        P1[Phase 1
Assessment]
        P2[Phase 2
Planning]
        P3[Phase 3
Deployment]
        P4[Phase 4
Optimization]
    end
    
    F1 & F2 & F3 --> C1
    F3 & F4 & F5 --> C2
    F4 & F5 & F6 --> C3
    F1 & F4 --> C4
    F3 & F5 & F6 --> C5
    
    C1 --> P1
    C2 --> P2
    C3 --> P3
    C4 --> P3
    C5 --> P4

Figure 1: UMAID Framework Three-Layer Architecture

3.3 The FUTURE Principles Adapted for Ukraine

F — Fairness

Principle: AI systems must perform equitably across all demographic groups served by Ukrainian healthcare.

Ukrainian Adaptation: Validation datasets must include representation from all oblasts, urban/rural populations, and age demographics. Special attention to post-conflict populations with trauma-related imaging patterns.

Metric: Performance disparity across demographic subgroups <5% for all primary diagnostic outputs.

U — Universality

Principle: Solutions should be applicable across diverse healthcare settings.

Ukrainian Adaptation: Framework supports deployment in tertiary hospitals with advanced PACS, regional diagnostic centers, and mobile imaging units. Degraded-mode operation for internet connectivity disruptions.

Metric: Successful deployment across ≥3 facility types with minimal configuration changes.

T — Traceability

Principle: Complete audit trails from input to output for all AI-assisted diagnoses.

Ukrainian Adaptation: Integration with eZdorovya electronic health record system. Compliance with Ukrainian data protection requirements (aligned with GDPR). Immutable logging for regulatory inspection.

Metric: 100% of AI-assisted cases traceable within 24 hours of request.

U — Usability

Principle: AI tools must integrate naturally into existing clinical workflows.

Ukrainian Adaptation: Ukrainian language UI with medical terminology verified by Ukrainian radiologists. Integration with common Ukrainian PACS vendors (including locally-developed systems). Response time compatible with existing reporting SLAs.

Metric: Radiologist adoption rate >80% within 6 months of deployment.

R — Robustness

Principle: Systems must maintain performance under real-world variation.

Ukrainian Adaptation: Validation across equipment from multiple manufacturers (including older systems common in Ukrainian facilities). Performance validation under varying image quality conditions. Graceful degradation with clear clinician notification.

Metric: Performance degradation <3% across validated equipment range.

E — Explainability

Principle: AI outputs must be interpretable by clinicians.

Ukrainian Adaptation: Visual explanation overlays (heatmaps, bounding boxes) with Ukrainian-language annotations. Confidence intervals and uncertainty quantification. Integration with structured reporting templates familiar to Ukrainian radiologists.

Metric: >90% of radiologists report AI explanations as “useful” or “very useful” for clinical decision-making.

4. Framework Components

4.1 Governance Structure

Successful AI deployment requires dedicated governance structures that span organizational boundaries. The UMAID Framework recommends a three-tier governance model:

flowchart TB
    subgraph Strategic["Strategic Level"]
        SC[AI Steering Committee
C-Suite + Clinical Leadership]
    end
    
    subgraph Tactical["Tactical Level"]
        TC[Technical Committee
IT + Radiology + Quality]
        EC[Ethics & Compliance
Legal + Privacy + Patient Rep]
    end
    
    subgraph Operational["Operational Level"]
        DT[Deployment Team
Per-Site Implementation]
        QA[QA Team
Monitoring & Validation]
        TR[Training Team
User Education]
    end
    
    SC --> TC
    SC --> EC
    TC --> DT
    TC --> QA
    EC --> DT
    TC --> TR
    
    DT -.->|Feedback| TC
    QA -.->|Reports| SC

Figure 2: Three-Tier Governance Model

Committee Composition Meeting Frequency Key Responsibilities
AI Steering Committee CEO, CMO, CIO, Chief Radiologist, Patient Representative Monthly Strategy, budget, go/no-go decisions, incident escalation
Technical Committee IT Director, Lead Radiologist, Quality Director, Vendor Rep Bi-weekly Technical specifications, integration issues, performance review
Ethics & Compliance Legal Counsel, Privacy Officer, Bioethicist, Patient Advocate Monthly Regulatory compliance, ethical review, consent management
Deployment Team Site IT Lead, Radiologist Champion, Project Manager Weekly Installation, configuration, initial validation

4.2 Technical Infrastructure Requirements

The technical infrastructure layer defines minimum requirements for successful AI deployment, with tiered specifications based on facility capabilities:

🔧 Infrastructure Tiers

Tier 1 (Basic): On-premises inference, manual result review, batch processing acceptable. Suitable for smaller diagnostic centers.

Tier 2 (Standard): PACS-integrated inference, real-time prioritization, structured result delivery. Suitable for regional hospitals.

Tier 3 (Advanced): Cloud-hybrid processing, federated learning participation, multi-modal integration. Suitable for university hospitals and specialized centers.

Component Tier 1 (Minimum) Tier 2 (Recommended) Tier 3 (Advanced)
Compute CPU-based inference server GPU workstation (NVIDIA T4+) Multi-GPU cluster + cloud burst
Storage Local SSD, 500GB cache NAS with 10TB+ medical imaging Tiered storage with cloud archive
Network 100 Mbps to PACS 1 Gbps internal, 100 Mbps external 10 Gbps internal, redundant WAN
PACS Integration DICOM receive/send DICOM + HL7 FHIR + worklist Full IHE AI Workflow profile
eZdorovya Manual result entry API integration (reports) Full bidirectional sync

4.3 Clinical Protocols

The clinical protocol component defines how AI systems interact with diagnostic workflows, emphasizing radiologist-AI collaboration rather than autonomous operation:

sequenceDiagram
    participant PACS as PACS System
    participant AI as AI Engine
    participant WL as Worklist Manager
    participant RAD as Radiologist
    participant RIS as RIS/eZdorovya
    
    PACS->>AI: New study notification (DICOM)
    AI->>AI: Inference processing
    alt Critical Finding Detected
        AI->>WL: Priority elevation + alert
        AI->>RAD: Urgent notification
    else Routine Finding
        AI->>WL: Standard queue position
    end
    AI->>PACS: Annotations + structured output
    RAD->>PACS: Open study with AI overlay
    RAD->>RAD: Clinical interpretation
    alt Agrees with AI
        RAD->>RIS: Finalize with AI-concordant flag
    else Modifies/Rejects AI
        RAD->>RIS: Override with clinical rationale
        RAD->>AI: Feedback for learning
    end
    RIS->>RIS: Audit log entry

Figure 3: Radiologist-AI Collaboration Workflow

⚠️ Critical Protocol: AI Override Documentation

When radiologists disagree with AI findings, the override must be documented with clinical rationale. This serves three purposes: (1) medicolegal protection, (2) quality improvement through pattern analysis, and (3) potential contribution to model improvement via federated learning. The framework mandates structured override categories rather than free-text to enable systematic analysis.

4.4 Confidence Thresholds and Escalation

The framework defines a tiered confidence system that determines workflow routing:

Confidence Level Score Range Workflow Action Radiologist Requirement
High Confidence Normal ≥0.95 (no pathology) Standard queue, pre-populated template Review + sign-off (may be expedited)
Moderate Confidence 0.70-0.95 Standard queue, AI findings displayed Full independent review required
Low Confidence <0.70 Flagged for attention, no AI overlay default Independent review, AI available on request
Critical Finding Any (specific pathology detected) Immediate priority elevation Urgent review within defined SLA

5. Implementation Phases

gantt
    title UMAID Implementation Timeline (Typical 12-Month Deployment)
    dateFormat  YYYY-MM
    section Phase 1: Assessment
    Readiness Assessment       :a1, 2026-01, 1M
    Vendor Evaluation          :a2, 2026-01, 2M
    Business Case Development  :a3, 2026-02, 1M
    section Phase 2: Planning
    Infrastructure Prep        :b1, 2026-03, 2M
    Protocol Development       :b2, 2026-03, 2M
    Training Curriculum        :b3, 2026-04, 1M
    section Phase 3: Deployment
    Pilot Installation         :c1, 2026-05, 1M
    Shadow Mode Operation      :c2, 2026-06, 2M
    Controlled Go-Live         :c3, 2026-08, 1M
    section Phase 4: Optimization
    Performance Monitoring     :d1, 2026-09, 4M
    Continuous Improvement     :d2, 2026-09, 4M
    Expansion Planning         :d3, 2026-11, 2M

Figure 4: Standard 12-Month Implementation Timeline

5.1 Phase 1: Assessment (Months 1-2)

Phase 1: Organizational Readiness Assessment

The assessment phase evaluates organizational readiness across five dimensions:

  1. Technical Readiness: Infrastructure audit, PACS compatibility, network capacity
  2. Clinical Readiness: Workflow analysis, radiologist capacity, case volume projections
  3. Organizational Readiness: Leadership commitment, change management capacity, resource availability
  4. Regulatory Readiness: MHSU compliance status, data protection posture, audit preparedness
  5. Financial Readiness: Budget availability, ROI timeline expectations, procurement processes

Key Deliverable: Readiness Scorecard with gap analysis and remediation roadmap

5.2 Phase 2: Planning (Months 2-4)

Phase 2: Detailed Planning and Preparation

Planning activities include:

  • Final vendor selection and contract negotiation (including SLAs, support terms, exit provisions)
  • Infrastructure procurement and installation
  • Clinical protocol development with radiologist input
  • Training curriculum design (technical and clinical tracks)
  • eZdorovya integration specifications
  • Baseline performance metrics establishment

Key Deliverable: Detailed Implementation Plan with milestones, responsibilities, and risk mitigation strategies

5.3 Phase 3: Deployment (Months 5-8)

Phase 3: Staged Deployment

Deployment follows a staged approach to minimize risk:

Stage 3a — Shadow Mode (8-12 weeks): AI runs in parallel without affecting clinical workflow. All AI outputs are compared against radiologist interpretations to establish local performance baseline.

Stage 3b — Controlled Introduction (4-6 weeks): AI results become visible to radiologists but with mandatory independent review. Performance monitoring intensifies.

Stage 3c — Operational Integration: AI fully integrated into workflow with defined confidence thresholds and escalation protocols active.

Key Deliverable: Validated deployment with documented performance meeting pre-defined acceptance criteria

5.4 Phase 4: Optimization (Months 9-12+)

Phase 4: Continuous Improvement

Ongoing optimization activities:

  • Monthly performance review against KPIs
  • Quarterly model performance validation
  • Radiologist feedback integration
  • Workflow refinement based on usage patterns
  • Expansion planning to additional modalities or sites
  • Participation in federated learning networks (if applicable)

Key Deliverable: Quarterly Optimization Report with performance trends and improvement recommendations

6. Validation

6.1 Validation Approach

The UMAID Framework was validated through three complementary methods:

  1. Expert Review: Framework components reviewed by 12 Ukrainian radiologists from 6 institutions across Kyiv, Odesa, Lviv, and Kharkiv oblasts
  2. Regulatory Alignment Check: Cross-reference with Ukrainian MHSU requirements and EU MDR provisions relevant to Ukrainian EU integration trajectory
  3. Pilot Feasibility Study: Detailed implementation planning with ScanLab diagnostic network (4 centers, ~50,000 annual imaging studies)

6.2 Expert Review Results

Framework Component Clarity Rating (1-5) Applicability Rating (1-5) Key Feedback Incorporated
FUTURE Principles 4.6 4.4 Added Ukrainian-specific adaptations for each principle
Governance Structure 4.2 3.8 Simplified structure for smaller institutions
Technical Requirements 4.4 4.1 Added tiered specifications for different facility types
Clinical Protocols 4.7 4.5 Refined confidence thresholds based on clinical input
Implementation Phases 4.5 4.3 Extended shadow mode recommendation (8-12 weeks)

✅ Validation Summary

Mean applicability rating across all components: 4.2/5.0

Expert consensus: Framework provides “actionable guidance that addresses real Ukrainian healthcare constraints while maintaining international quality standards.”

7. Application: ScanLab Pilot Design

To demonstrate framework application, we developed a detailed pilot implementation plan for ScanLab, a network of diagnostic imaging centers in Ukraine. This section summarizes key elements of that design.

7.1 Pilot Scope

  • Initial Site: ScanLab Kyiv Central (highest volume, best infrastructure)
  • Modality: CT chest imaging (COVID-19 patterns, lung nodule detection)
  • Volume: ~200 CT chest studies per week
  • Duration: 12-month pilot with 6-month shadow mode
  • Primary Endpoint: Radiologist efficiency (studies per hour) with quality maintenance
flowchart LR
    subgraph Current["Current State"]
        A[CT Scan] --> B[PACS Queue]
        B --> C[Radiologist Review]
        C --> D[Report]
        D --> E[eZdorovya]
    end
    
    subgraph Pilot["Pilot State"]
        A2[CT Scan] --> B2[PACS Queue]
        B2 --> AI[AI Analysis]
        AI --> W{Priority?}
        W -->|Critical| P1[Priority Queue]
        W -->|Normal| P2[Standard Queue]
        P1 --> C2[Radiologist + AI Overlay]
        P2 --> C2
        C2 --> D2[AI-Assisted Report]
        D2 --> E2[eZdorovya + Audit Log]
    end
    
    Current -.->|Transition| Pilot

Figure 5: ScanLab Pilot — Current vs. Target State Workflow

7.2 Expected Outcomes

  • Efficiency: 15-25% increase in studies interpreted per radiologist-hour
  • Quality: <1% increase in significant discrepancy rate (non-inferiority target)
  • Turnaround: 20-30% reduction in report turnaround time
  • Critical Finding Detection: <30 minute average time-to-notification for critical findings

8. Discussion

8.1 Theoretical Contributions

The UMAID Framework makes three primary contributions to the literature on medical AI implementation:

  1. Context-Specific Adaptation: Demonstrates how international frameworks (FUTURE-AI, WHO guidance) can be systematically adapted for emerging healthcare markets without compromising core principles
  2. Implementation Bridge: Connects high-level ethical and governance principles to concrete operational practices, addressing the “last mile” problem in medical AI deployment
  3. Tiered Approach: Provides infrastructure and governance specifications across multiple facility types, enabling adoption by institutions with varying resources

8.2 Practical Implications

For Ukrainian healthcare institutions, the framework provides:

  • Clear checklist for organizational readiness assessment
  • Vendor evaluation criteria tailored to Ukrainian requirements
  • eZdorovya integration specifications
  • Regulatory compliance roadmap for MHSU and EU alignment
  • Training curriculum foundations for radiologist and IT staff

8.3 Limitations

⚠️ Framework Limitations

  • Limited Empirical Validation: Framework based on expert review and pilot planning; full deployment validation pending
  • Modality Focus: Primary focus on CT and radiography; extension to other modalities requires additional validation
  • Regulatory Evolution: Ukrainian regulatory landscape continues to evolve; framework will require updates
  • Resource Assumptions: Minimum Tier 1 infrastructure assumes baseline capabilities that may not exist in all facilities

8.4 Future Extensions

Planned extensions to the framework include:

  1. Detailed specifications for additional imaging modalities (MRI, ultrasound, nuclear medicine)
  2. Federated learning participation protocols for Ukrainian institutions
  3. Cross-border deployment guidance for EU alignment
  4. Economic model templates with Ukrainian cost assumptions
  5. Integration with emerging Ukrainian teleradiology networks

9. Conclusion

The UMAID Framework represents a comprehensive, evidence-based approach to deploying machine learning-based medical imaging diagnosis systems in Ukrainian healthcare institutions. By synthesizing international best practices with Ukrainian-specific adaptations, the framework addresses the unique challenges facing healthcare leaders navigating the complex landscape of medical AI implementation.

The framework’s foundation on FUTURE principles — Fairness, Universality, Traceability, Usability, Robustness, and Explainability — ensures alignment with emerging international standards while the tiered implementation approach accommodates the diversity of Ukrainian healthcare facilities.

As Ukraine continues its trajectory toward EU integration and healthcare system modernization, systematic AI deployment frameworks will be essential for realizing the promise of these technologies while maintaining patient safety and clinical quality. The UMAID Framework provides a starting point for this journey, with planned extensions to address emerging technologies and evolving regulatory requirements.

🎯 Key Takeaway

Successful medical AI deployment requires more than technology — it demands a comprehensive approach encompassing governance, clinical protocols, training, and continuous monitoring. The UMAID Framework provides Ukrainian healthcare institutions with a structured roadmap to navigate this complexity while maintaining focus on the ultimate goal: improved patient outcomes through augmented diagnostic capabilities.

References

  1. ACR Data Science Institute. (2026). FDA cleared AI algorithms. American College of Radiology. https://www.acrdsi.org/DSI-Services/FDA-Cleared-AI-Algorithms
  2. FUTURE-AI International Consortium. (2025). FUTURE-AI: International consensus guideline for trustworthy and deployable artificial intelligence in healthcare. BMJ, 388, e081554. https://doi.org/10.1136/bmj-2024-081554
  3. World Health Organization. (2024). Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. WHO. https://www.who.int/publications/i/item/9789240084759
  4. Joint Commission & CHAI. (2025). The responsible use of AI in healthcare (RUAIH). The Joint Commission. https://www.jointcommission.org/
  5. Spatharou, A., Hieronimus, S., & Jenkins, J. (2020). Transforming healthcare with AI: The impact on the workforce and organizations. McKinsey & Company.
  6. Malakhov, K. S., et al. (2023). Insight into the Digital Health System of Ukraine (eHealth): Trends, Definitions, Standards, and Legislative Revisions. Informatics, 10(4), 95. https://doi.org/10.3390/informatics10040095
  7. IT Ukraine Association. (2025). Ukraine’s HealthTech Industry — Technological challenges and the path to European integration. https://itukraine.org.ua/
  8. Canada’s Drug Agency (CDA-AMC). (2025). 2025 Watch List: Artificial Intelligence in Health Care. NCBI Bookshelf. https://www.ncbi.nlm.nih.gov/books/NBK613808/
  9. Radiology AI Council. (2025). AI in Action: A road map for effective model evaluation and deployment. Journal of the American College of Radiology. https://doi.org/10.1016/j.jacr.2025.05.001
  10. Mayo Clinic Proceedings: Digital Health. (2024). Implementing Artificial Intelligence Algorithms in the Radiology Workflow: Challenges and Considerations. https://doi.org/10.1016/j.mcpdig.2024.121
  11. Palagin, O. V., et al. (2023). Smart-system for remote support of rehabilitation activities and services. Cybernetics and Systems Analysis, 59(2), 189-204. https://doi.org/10.1007/s10559-023-00550-y
  12. RSNA/MICCAI. (2024). Integrating and Adopting AI in the Radiology Workflow: A Primer for Standards and IHE Profiles. Radiology, 311(2), e232653. https://doi.org/10.1148/radiol.232653
  13. Wolters Kluwer. (2025). 2026 healthcare AI trends: Insights from experts. https://www.wolterskluwer.com/
  14. IntuitionLabs. (2025). AI in Radiology: 2025 Trends, FDA Approvals & Adoption. https://intuitionlabs.ai/articles/ai-radiology-trends-2025
  15. European Commission. (2024). EU Artificial Intelligence Act: Regulation (EU) 2024/1689. Official Journal of the European Union.
  16. Ukrainian Ministry of Health (MHSU). (2023). Concept for the Development of Telemedicine in Ukraine 2023-2025. Government of Ukraine.
  17. UNDP Ukraine. (2024). eHealth Summit 2024: Digital healthcare projects presentation. https://www.undp.org/ukraine/
  18. Lakhani, P., et al. (2023). Machine learning in radiology: Applications beyond image interpretation. Journal of the American College of Radiology, 20(1), 13-24. https://doi.org/10.1016/j.jacr.2022.09.017
  19. HIMSS. (2025). Driving the Future of Health with AI. https://www.himss.org/futureofai/
  20. World Economic Forum. (2025). The Future of AI-Enabled Health: Leading the Way. WEF White Paper. https://reports.weforum.org/
  21. NHS AI Lab. (2024). Lessons learned from the £250M AI programme. NHS England.
  22. Topol, E. J. (2019). High-performance medicine: The convergence of human and artificial intelligence. Nature Medicine, 25(1), 44-56. https://doi.org/10.1038/s41591-018-0300-7
  23. Liu, X., et al. (2019). A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging. The Lancet Digital Health, 1(6), e271-e297. https://doi.org/10.1016/S2589-7500(19)30123-2
  24. Vasey, B., et al. (2022). Reporting guideline for the early-stage clinical evaluation of decision support systems driven by artificial intelligence. Nature Medicine, 28, 924-933. https://doi.org/10.1038/s41591-022-01772-9
  25. Dluhopolskyi, O., et al. (2023). The Implementation of the eHealth System and Anticorruption Reforms (Case of EU Countries for Ukraine). Journal of Global Health, 15, 03039. https://doi.org/10.7189/jogh.15.03039