Medical ML: Training Curriculum for Medical AI — Healthcare Professional Development Framework

Posted on by

Training Curriculum for Medical AI: A Comprehensive Framework for Healthcare Professional Development

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

Abstract

The integration of artificial intelligence into medical imaging diagnosis demands comprehensive training programs that prepare healthcare professionals for effective human-AI collaboration. This paper presents a structured training curriculum framework for medical AI, synthesizing international best practices with localized implementation strategies for Ukrainian healthcare contexts. Drawing from the AAPM-ACR-RSNA-SIIM joint syllabus, Delphi-validated competency frameworks, and Kirkpatrick evaluation methodology, we propose a modular curriculum spanning foundational AI literacy through advanced clinical integration. The framework addresses four distinct persona categories—AI users, purchasers, clinical collaborators, and developers—with differentiated learning pathways totaling 120-240 hours depending on role complexity. Assessment strategies incorporate knowledge testing, simulation-based competency evaluation, and workplace-based observation to ensure translation of learning into clinical practice. Implementation guidelines specifically address Ukrainian healthcare system constraints including infrastructure limitations, language localization requirements, and integration with existing medical education accreditation. The curriculum framework supports the ScanLab pilot program while providing generalizable structure for national medical AI education initiatives.

24%
of radiology residents report no AI/ML education in their programs (2026 survey)
82
essential AI competencies identified across healthcare professions
52%
of AI medical education publications appeared after ChatGPT release (Nov 2022)
4
distinct learner personas requiring differentiated curricula

1. Introduction

The rapid proliferation of AI-enabled medical devices—exceeding 1,200 FDA authorizations as of 2026 with 80% targeting radiology—has outpaced the educational infrastructure needed to prepare healthcare professionals for effective utilization. A 2026 survey revealed that approximately 24% of radiology residents report having no AI/ML educational offerings in their residency programs, despite the technology’s ubiquitous presence in modern imaging departments.

This educational gap creates multiple risks: underutilization of expensive AI investments, over-reliance without appropriate critical appraisal, and failure to recognize AI limitations in edge cases. The Josiah Macy Jr. Foundation’s 2025 report identified five domains where AI impacts medical education—admissions, classroom-based learning, workplace-based learning, assessment/feedback, and program evaluation—yet most training programs address none systematically.

For Ukrainian healthcare institutions implementing AI diagnostics through initiatives like ScanLab, the training challenge is compounded by limited existing curricula in the local language, varying baseline digital literacy among medical staff, and the need to integrate with national accreditation requirements. This paper presents a comprehensive training curriculum framework that addresses these challenges while maintaining international standards alignment.

1.1 Training Objectives

The curriculum framework aims to achieve the following objectives:

  • Establish foundational AI literacy across all healthcare professionals interacting with AI-enabled diagnostic tools
  • Develop role-specific competencies for users, purchasers, clinical collaborators, and technical developers
  • Ensure appropriate trust calibration—neither blind acceptance nor reflexive rejection of AI recommendations
  • Prepare staff for regulatory compliance, quality assurance, and continuous monitoring responsibilities
  • Support Ukrainian language localization while maintaining international competency standards

2. Background and Related Work

2.1 International Curriculum Initiatives

The AAPM, ACR, RSNA, and SIIM joint effort represents the most comprehensive attempt to standardize AI education for medical imaging professionals. Published in late 2025, their syllabus defines competencies across four persona categories rather than traditional role hierarchies, recognizing that an AI “user” (radiologist interpreting AI-highlighted findings) requires different knowledge than a “purchaser” (department chair selecting AI vendors) or “developer” (data scientist building algorithms).

Table 1: Major International AI Curriculum Initiatives for Healthcare
Initiative Organization(s) Year Focus Target Audience
Multisociety AI Syllabus AAPM, ACR, RSNA, SIIM 2025 Radiology/imaging professionals Users, purchasers, collaborators, developers
Imaging AI Certificate Program RSNA 2022-ongoing Foundational to advanced AI literacy Radiologists, residents
23 AI Competencies Delphi Consensus Panel 2022 Validated physician competencies All physicians
Macy Foundation Framework Josiah Macy Jr. Foundation 2025 Five educational domains Medical education institutions
NHS AI Curriculum UK NHS Health Education 2021-ongoing Healthcare workforce AI readiness All NHS staff
FACETS Assessment Framework BEME Guide 84 2024 AI intervention assessment taxonomy Medical educators

2.2 Competency Framework Analysis

A 2024 scoping review identified 30 educational programs and 2 curriculum frameworks for AI in medical education. The review noted significant heterogeneity: 17% of programs targeted radiology residents, 26% served practicing physicians through CME, and none described underlying learning theories or pedagogical frameworks guiding program design.

The Delphi-validated 23 AI competencies for physicians span three domains:

  • Knowledge: Understanding ML fundamentals, data requirements, validation concepts, performance metrics
  • Skills: Interpreting AI outputs, recognizing limitations, appropriate reliance decisions, clinical integration
  • Attitudes: Ethical awareness, bias recognition, patient communication, lifelong learning commitment

flowchart TB
    subgraph Knowledge["Knowledge Domain"]
        K1[ML Fundamentals]
        K2[Data Requirements]
        K3[Validation Concepts]
        K4[Performance Metrics]
        K5[Regulatory Landscape]
    end
    
    subgraph Skills["Skills Domain"]
        S1[Output Interpretation]
        S2[Limitation Recognition]
        S3[Appropriate Reliance]
        S4[Clinical Integration]
        S5[Quality Monitoring]
    end
    
    subgraph Attitudes["Attitudes Domain"]
        A1[Ethical Awareness]
        A2[Bias Recognition]
        A3[Patient Communication]
        A4[Lifelong Learning]
        A5[Collaboration Mindset]
    end
    
    Knowledge --> Skills
    Skills --> Attitudes
    
    K1 -.->|informs| S1
    K3 -.->|enables| S2
    S3 -.->|requires| A1
    S4 -.->|builds| A5

2.3 Gap Analysis: Current State

Despite proliferating AI tools, significant educational gaps persist:

📊 Current Training Deficiencies

  • No validated assessment tools exist for AI competency in clinical contexts
  • Faculty preparedness gap: Most medical educators lack formal AI training
  • Curricular crowding: Adding AI competes with existing content
  • Standardization absence: No accreditation requirements for AI competencies
  • LMICs underrepresented: Most frameworks target high-income healthcare systems

3. Curriculum Architecture

3.1 Persona-Based Learning Pathways

Following the AAPM-ACR-RSNA-SIIM model, our curriculum differentiates four learner personas with distinct competency requirements:

flowchart LR
    subgraph Foundation["Foundation Module (All Personas)"]
        F1[AI Fundamentals
8 hours]
        F2[Ethics & Bias
4 hours]
        F3[Regulatory Basics
4 hours]
    end
    
    subgraph Users["AI Users Pathway"]
        U1[Clinical Interpretation
16 hours]
        U2[Workflow Integration
8 hours]
        U3[Quality Assurance
8 hours]
    end
    
    subgraph Purchasers["AI Purchasers Pathway"]
        P1[Vendor Evaluation
12 hours]
        P2[Implementation Planning
12 hours]
        P3[ROI Assessment
8 hours]
    end
    
    subgraph Collaborators["Clinical Collaborators Pathway"]
        C1[Dataset Curation
16 hours]
        C2[Annotation Standards
12 hours]
        C3[Validation Protocols
16 hours]
    end
    
    subgraph Developers["AI Developers Pathway"]
        D1[Model Architecture
24 hours]
        D2[Training Pipelines
20 hours]
        D3[Deployment & MLOps
20 hours]
    end
    
    Foundation --> Users
    Foundation --> Purchasers
    Foundation --> Collaborators
    Foundation --> Developers
Table 2: Curriculum Hours by Persona Category
Persona Foundation Role-Specific Practicum Total Hours
AI User (Radiologist, Physician) 16 32 40 88
AI User (Technologist) 16 24 24 64
AI Purchaser (Administrator) 16 32 16 64
Clinical Collaborator 16 44 60 120
AI Developer 16 64 80 160

3.2 Module Structure

Module 1: AI Foundations (16 hours) — All Personas

This foundational module establishes common vocabulary and conceptual understanding across all healthcare professionals interacting with AI systems.

Table 3: Foundation Module Learning Objectives
Topic Hours Learning Objectives Assessment Method
Introduction to AI/ML 4 Define AI, ML, deep learning; distinguish supervised/unsupervised learning; explain neural network basics MCQ quiz
AI in Medical Imaging 4 Identify current FDA-cleared applications; describe CAD vs. autonomous systems; recognize appropriate use cases Case analysis
Ethics and Bias 4 Recognize algorithmic bias sources; apply fairness frameworks; navigate informed consent for AI-assisted care Scenario discussion
Regulatory Landscape 4 Distinguish FDA/CE/Ukrainian pathways; interpret device classifications; understand post-market surveillance Regulatory case study

Module 2: Clinical Interpretation (32 hours) — AI Users

For radiologists, pathologists, and physicians who will interpret AI-generated findings in clinical workflows.

sequenceDiagram
    participant Img as Imaging System
    participant AI as AI Algorithm
    participant Rad as Radiologist
    participant EHR as EHR/PACS
    participant Pat as Patient Record
    
    Img->>AI: DICOM images
    AI->>AI: Processing & analysis
    AI->>Rad: Findings + confidence scores
    Rad->>Rad: Critical appraisal
    alt AI agrees with clinical suspicion
        Rad->>EHR: Confirmed finding
    else AI disagrees
        Rad->>Rad: Independent review
        Rad->>EHR: Documented decision
    end
    EHR->>Pat: Final report
Table 4: Clinical Interpretation Module Content
Topic Hours Key Competencies
Performance Metrics Interpretation 6 AUC-ROC, sensitivity/specificity at clinical thresholds, PPV/NPV in prevalence contexts
Confidence Score Utilization 4 Threshold selection rationale, uncertainty quantification, calibration assessment
Explainability Methods 6 Interpreting attention maps, GradCAM, SHAP values; recognizing explanation limitations
Failure Mode Recognition 8 Distribution shift detection, artifact sensitivity, edge case identification
Override Decision Making 8 When to trust, question, or reject AI recommendations; documentation requirements

Module 3: Technical Operations (24-64 hours) — Technologists and Developers

Differentiated content for imaging technologists (workflow operators) versus developers (algorithm creators).

🔧 Technologist Track (24 hours)

  • Image acquisition optimization for AI processing
  • DICOM header requirements and routing
  • System troubleshooting and error reporting
  • Quality control protocols and drift detection

💻 Developer Track (64 hours)

  • Medical imaging preprocessing pipelines
  • CNN and Vision Transformer architectures
  • Transfer learning and domain adaptation
  • Federated learning implementation
  • MLOps for healthcare deployment
  • Regulatory submission requirements

Module 4: Implementation and Quality (32 hours) — Purchasers and Administrators

For healthcare administrators responsible for AI procurement, deployment, and ongoing governance.

Table 5: Implementation Module Components
Component Hours Deliverables
Vendor Evaluation Framework 8 Standardized RFP template, evaluation scorecard, reference check protocol
PACS Integration Planning 8 Integration architecture document, workflow mapping, timeline
ROI Analysis Methods 8 Cost-benefit model, productivity metrics, quality outcome measures
Governance Framework 8 Algorithm oversight committee charter, monitoring protocols, escalation procedures

4. Assessment Framework

4.1 Kirkpatrick Evaluation Model Application

The curriculum employs the Kirkpatrick Four-Level Training Evaluation Model, extended with Level 0 (baseline assessment) and Level 5 (organizational impact) for comprehensive program evaluation.

flowchart TB
    subgraph PreTraining["Pre-Training"]
        L0[Level 0: Baseline
Prior knowledge assessment]
    end
    
    subgraph Training["During Training"]
        L1[Level 1: Reaction
Satisfaction surveys, engagement metrics]
        L2[Level 2: Learning
Knowledge tests, skill demonstrations]
    end
    
    subgraph PostTraining["Post-Training"]
        L3[Level 3: Behavior
Workplace observation, chart review]
        L4[Level 4: Results
Clinical outcomes, efficiency metrics]
        L5[Level 5: Organizational
ROI, culture change, adoption rates]
    end
    
    L0 --> L1
    L1 --> L2
    L2 --> L3
    L3 --> L4
    L4 --> L5
Table 6: Assessment Methods by Kirkpatrick Level
Level Focus Methods Timing
0 – Baseline Prior knowledge Pre-test MCQ, self-assessment survey Before training
1 – Reaction Satisfaction Course evaluations, Net Promoter Score After each module
2 – Learning Knowledge/skills Post-test MCQ, simulation scenarios, case analysis Module completion
3 – Behavior Application Workplace observation, AI override audit, peer review 3-6 months post
4 – Results Outcomes Diagnostic accuracy, turnaround time, patient outcomes 6-12 months post
5 – Organizational Impact AI utilization rates, staff confidence surveys, ROI analysis Annual review

4.2 Competency-Based Assessment Design

Building on the FACETS framework from BEME Guide 84 (2024), assessments target specific competency domains with appropriate methods:

Knowledge Assessment (MCQ and Short Answer)

Sample competency: “Interpret AUC-ROC curves in context of clinical decision thresholds”

Example Assessment Item

Scenario: A chest X-ray AI system reports AUC of 0.95 for pneumothorax detection. In your clinical setting, pneumothorax prevalence is 2%.

Question: At a sensitivity threshold of 95%, the AI’s specificity is 80%. Calculate the positive predictive value and explain how this affects your clinical workflow design.

Expected Response Elements:

  • PPV calculation: ~9% (demonstrates understanding of prevalence impact)
  • Recognition that most positive AI findings will be false positives in low-prevalence settings
  • Workflow implication: AI serves as screening tool requiring radiologist confirmation

Skills Assessment (Simulation-Based)

Simulation scenarios present AI outputs with embedded challenges requiring appropriate critical appraisal:

Table 7: Simulation Assessment Scenarios
Scenario AI Behavior Expected Learner Response Competency Tested
Chest CT with artifact False positive nodule detection Recognize artifact, override AI, document rationale Failure mode recognition
Mammogram with prior comparison Missed interval change Identify AI limitation, conduct independent review Appropriate reliance calibration
Brain MRI in pediatric patient Low confidence score Recognize out-of-distribution case, apply clinical judgment Uncertainty quantification interpretation
Retrospective audit showing drift Degraded performance metrics Escalate to AI oversight committee, document concerns Quality monitoring responsibility

Attitudes Assessment (360-Degree Feedback)

Attitudes toward AI collaboration are assessed through structured feedback from colleagues, patients, and supervisors:

  • Peer observation: Does the learner appropriately explain AI involvement to colleagues?
  • Patient feedback: How effectively does the learner communicate AI-assisted diagnosis?
  • Supervisor rating: Does the learner demonstrate appropriate trust calibration in AI recommendations?

5. Implementation Strategy

5.1 Phased Rollout Plan

gantt
    title AI Training Curriculum Implementation Timeline
    dateFormat  YYYY-MM
    section Phase 1: Foundation
    Faculty Development           :2026-03, 3M
    Curriculum Localization       :2026-03, 4M
    Platform Setup                :2026-04, 2M
    section Phase 2: Pilot
    Champion Cohort (20 staff)    :2026-07, 3M
    Assessment Validation         :2026-08, 2M
    Curriculum Refinement         :2026-09, 2M
    section Phase 3: Scale
    Department-wide Rollout       :2026-11, 4M
    CME Integration               :2027-01, 3M
    National Accreditation        :2027-03, 6M

5.2 Faculty Development Program

The primary barrier to AI education is faculty unfamiliarity—most medical educators completed training before clinical AI deployment. A dedicated faculty development program addresses this gap:

Table 8: Faculty Development Curriculum
Component Duration Format Outcomes
AI Foundations Intensive 24 hours Workshop (in-person) Personal AI literacy, teaching confidence
Pedagogical Methods 8 hours Online modules Adult learning principles, simulation facilitation
Assessment Design 8 hours Workshop Valid assessment item creation, rubric development
Teaching Practicum 16 hours Supervised teaching Observed teaching sessions with feedback
Ongoing Community Continuous Monthly meetings Peer support, curriculum updates, best practices

5.3 Ukrainian Context Adaptations

The curriculum framework requires specific adaptations for Ukrainian healthcare contexts:

Language Localization

  • Medical terminology standardization: Develop Ukrainian AI/ML glossary aligned with existing medical vocabulary standards
  • Interface translation: Ensure AI system interfaces display Ukrainian with appropriate medical terminology
  • Assessment localization: Translate and validate assessment instruments maintaining psychometric properties

Infrastructure Considerations

  • Connectivity limitations: Design offline-capable learning modules for facilities with unreliable internet
  • Hardware constraints: Simulation scenarios must function on available equipment
  • PACS diversity: Include training for multiple PACS vendor integration scenarios

Regulatory Alignment

  • MHSU requirements: Map curriculum to Ukrainian Ministry of Health Service requirements
  • CME credit recognition: Secure accreditation from Ukrainian medical education authorities
  • EU harmonization: Align with CE marking requirements given Ukraine’s EU integration trajectory

6. Learning Management System Architecture

6.1 Platform Requirements

flowchart TB
    subgraph Frontend["Learner Interface"]
        Web[Web Portal
Ukrainian/English]
        Mobile[Mobile App
Offline capable]
        LTI[LTI Integration
Existing LMS]
    end
    
    subgraph Core["Learning Core"]
        Content[Content Delivery
Video, interactive]
        Assess[Assessment Engine
Adaptive testing]
        Sim[Simulation Platform
AI case scenarios]
        Track[Progress Tracking
Competency mapping]
    end
    
    subgraph Data["Data Layer"]
        LRS[Learning Record Store
xAPI compliant]
        Analytics[Learning Analytics
Dashboards]
        Report[Reporting
Accreditation, compliance]
    end
    
    subgraph Integration["External Integration"]
        HR[HR Systems
Staff records]
        PACS[PACS/AI Systems
Usage data]
        Accred[Accreditation Bodies
CME reporting]
    end
    
    Frontend --> Core
    Core --> Data
    Data --> Integration

6.2 Content Delivery Specifications

Table 9: Content Format Standards
Content Type Format Duration Interaction
Concept Videos MP4, H.264, 720p minimum 5-12 minutes Embedded quizzes, transcripts (UA/EN)
Interactive Modules SCORM 1.2/xAPI 15-30 minutes Branching scenarios, knowledge checks
Case Simulations Web-based (HTML5) 20-45 minutes AI output interpretation, decision documentation
Reading Materials PDF, EPUB Variable Downloadable, searchable, annotatable
Live Sessions Webinar platform 60-90 minutes Q&A, breakout rooms, polling

7. Quality Assurance and Continuous Improvement

7.1 Curriculum Review Cycle

Given the rapid evolution of medical AI, the curriculum requires structured review cycles:

  • Quarterly: Content currency review (new FDA/CE clearances, literature updates)
  • Semi-annual: Assessment item analysis and refinement
  • Annual: Comprehensive curriculum review with stakeholder input
  • Triggered: Major technology or regulatory changes prompt immediate review

7.2 Learning Analytics Dashboard

Key metrics tracked for continuous improvement:

Table 10: Learning Analytics Metrics
Metric Category Specific Metrics Target Action Threshold
Engagement Module completion rate, time-on-task, video completion >90% completion <80% triggers content review
Performance Assessment pass rates, first-attempt scores >85% pass rate <75% triggers assessment/content review
Satisfaction NPS, module ratings, qualitative feedback NPS >40 NPS <20 triggers redesign
Transfer AI utilization rates, override appropriateness Baseline +20% No improvement triggers support intervention

8. ScanLab Integration Specifications

8.1 Pilot Program Training Timeline

For the ScanLab pilot implementation, training follows this sequence aligned with system deployment:

Table 11: ScanLab Pilot Training Schedule
Phase Training Focus Duration Participants
Pre-deployment (T-8 weeks) Foundation modules (all personas) 16 hours All pilot staff
Pre-deployment (T-4 weeks) Role-specific pathways 24-44 hours By persona category
Deployment (T-0) System-specific training 8 hours All pilot staff
Early operation (T+2 weeks) Supervised practice, troubleshooting 16 hours All pilot staff
Stabilization (T+8 weeks) Advanced scenarios, optimization 8 hours Radiologists, technologists
Ongoing (Monthly) Case conferences, updates 2 hours/month All pilot staff

8.2 Competency Certification Requirements

Staff must demonstrate competency before independent AI-assisted practice:

✅ Certification Requirements

  • Foundation Assessment: Pass with ≥80% score
  • Role-Specific Assessment: Pass with ≥85% score
  • Simulation Scenarios: Complete 5 scenarios with satisfactory ratings
  • Supervised Practice: 20 AI-assisted cases with mentor sign-off
  • Attestation: Sign acknowledgment of responsibilities and limitations

9. Future Directions

9.1 Emerging Training Needs

The curriculum framework must anticipate evolving training requirements:

  • Multimodal AI: Systems integrating imaging with clinical data, genomics, and pathology
  • Generative AI: LLM-based clinical reasoning assistants and report generation
  • Foundation models: General-purpose medical AI requiring different interaction patterns
  • Autonomous systems: Preparation for progressively independent AI decision-making

9.2 Research Agenda

Prioritized research questions for curriculum development:

  1. What assessment methods best predict appropriate AI reliance in clinical practice?
  2. How does training with AI assistants affect development of independent diagnostic skills?
  3. What faculty development interventions most effectively improve AI teaching confidence?
  4. How should curricula adapt for healthcare workers with varying digital literacy baselines?
  5. What longitudinal outcomes demonstrate effective AI education programs?

10. Conclusion

Effective integration of AI into medical diagnosis requires comprehensive training programs that prepare healthcare professionals for human-AI collaboration while maintaining independent clinical reasoning capabilities. This curriculum framework provides a structured approach spanning foundational AI literacy through advanced clinical integration, with differentiated pathways for distinct learner personas.

Key implementation principles emerging from international best practices include:

  • Persona-based pathways: Differentiate training by role (user, purchaser, collaborator, developer) rather than traditional hierarchies
  • Faculty-first approach: Invest in educator development before broad curriculum deployment
  • Competency-based assessment: Use multiple methods (knowledge testing, simulation, workplace observation) to verify learning transfer
  • Continuous evolution: Build review cycles that keep pace with rapidly advancing technology
  • Local adaptation: Customize international frameworks for specific healthcare contexts and languages

For Ukrainian healthcare institutions implementing AI diagnostics, this framework provides actionable guidance while maintaining alignment with international competency standards. Success requires sustained institutional commitment, adequate faculty development resources, and integration with existing medical education accreditation pathways.

The training curriculum is not merely a prerequisite for AI deployment—it is a continuous investment in the human intelligence that must guide, evaluate, and ultimately be responsible for AI-assisted medical decisions.

References

  1. Kitamura FC, et al. Teaching AI for Radiology Applications: A Multisociety-Recommended Syllabus from the AAPM, ACR, RSNA, and SIIM. Radiology: Artificial Intelligence. 2025. DOI: 10.1148/ryai.250137
  2. Tolentino R, et al. Curriculum Frameworks and Educational Programs in AI for Medical Students, Residents, and Practicing Physicians: Scoping Review. JMIR Medical Education. 2024;10:e54793. DOI: 10.2196/54793
  3. Boscardin CK, et al. Macy Foundation Report: AI in Medical Education. Academic Medicine. 2025. DOI: 10.1097/ACM.0000000000006107
  4. Simoni AH, et al. AI in Medical Education: A Scoping Review. BMC Medical Education. 2025. DOI: 10.1186/s12909-025-08188-2
  5. RSNA. Imaging AI Certificate Program Curriculum. RSNA Education. 2024. Available at: rsna.org/ai-certificate
  6. Seifert R, et al. A Framework to Integrate AI Training into Radiology Residency Programs. Insights into Imaging. 2024;15:1-12. DOI: 10.1186/s13244-023-01595-3
  7. Tschandl P, et al. Assessing AI Awareness and Identifying Essential Competencies: Insights from Key Stakeholders. JMIR Medical Education. 2024. DOI: 10.2196/52462
  8. Kirkpatrick JD, Kirkpatrick WK. Kirkpatrick’s Four Levels of Training Evaluation. ATD Press. 2016. ISBN: 978-1607280088
  9. Li Y, Lutfi A. LLM-Based Virtual Patients for History-Taking Training: Systematic Review. JMIR Medical Informatics. 2026. DOI: 10.2196/79039
  10. Pianykh OS, et al. Continuous Learning AI in Radiology: Implementation Strategies. Radiology. 2020;297:6-14. DOI: 10.1148/radiol.2020200038
  11. Park SH, Han K. Methodologic Guide for Evaluating Clinical Performance of AI. Radiology. 2018;286:800-809. DOI: 10.1148/radiol.2017171920
  12. Bluemke DA, et al. Assessing Radiology Research on AI: A Brief Guide for Authors, Reviewers, and Readers. Radiology. 2020;294:487-489. DOI: 10.1148/radiol.2019192515
  13. Geis JR, et al. Ethics of AI in Radiology: European and North American Multisociety Statement. Radiology. 2019;293:436-440. DOI: 10.1148/radiol.2019191586
  14. Langlotz CP. Will Artificial Intelligence Replace Radiologists? Radiology: AI. 2019;1:e190058. DOI: 10.1148/ryai.2019190058
  15. Cabitza F, et al. Unintended Consequences of Machine Learning in Medicine. JAMA. 2017;318:517-518. DOI: 10.1001/jama.2017.7797
  16. Shen J, et al. AI vs. Physician for Diagnostic Accuracy: A Systematic Review. NPJ Digital Medicine. 2019;2:24. DOI: 10.1038/s41746-019-0099-5
  17. Wong TY, et al. AI in Medical Imaging: Challenges and Opportunities. Nature Reviews Bioengineering. 2021;1:4-15. DOI: 10.1038/s43586-021-00048-x
  18. Topol EJ. High-Performance Medicine: The Convergence of Human and AI. Nature Medicine. 2019;25:44-56. DOI: 10.1038/s41591-018-0300-7
  19. Liu X, et al. Reporting Guidelines for Clinical Trial Reports on AI: CONSORT-AI. Nature Medicine. 2020;26:1364-1374. DOI: 10.1038/s41591-020-1034-x
  20. Collins GS, et al. Protocol for AI Prediction Studies: TRIPOD+AI. BMJ. 2024;384:e078378. DOI: 10.1136/bmj-2023-078378
  21. He J, et al. The Practical Implementation of AI in Medicine. Nature Medicine. 2019;25:30-36. DOI: 10.1038/s41591-018-0307-0
  22. Celi LA, et al. Sources of Bias in AI for Health Care. NEJM AI. 2024;1:AIra2300028. DOI: 10.1056/AIra2300028
  23. Paranjape K, et al. Introducing AI Training in Medical Education. JMIR Medical Education. 2019;5:e16048. DOI: 10.2196/16048
  24. Wartman SA, Combs CD. Medical Education Must Move from Ivory Tower to Real World. Academic Medicine. 2018;93:1171-1173. DOI: 10.1097/ACM.0000000000002044
  25. European Commission. AI Act Requirements for High-Risk Systems. Official Journal of the European Union. 2024. Available at: eur-lex.europa.eu

Article Series: This is Article 34 in the “ML for Medical Diagnosis” research series examining machine learning integration in Ukrainian healthcare.

Keywords: medical AI training, curriculum development, healthcare professional education, AI competencies, radiology education, Kirkpatrick evaluation, RSNA certificate, medical imaging AI, Ukrainian healthcare

Acknowledgments: This research is conducted as part of PhD studies at Odessa National Polytechnic University, Department of Economic Cybernetics.