6-phase research framework for ML-augmented medical diagnosis

ML for Medical Diagnosis: Research Goals and Framework for Ukrainian Healthcare

Posted on by Admin
ML for Medical Diagnosis Framework

ML for Medical Diagnosis: Research Goals and Framework for Ukrainian Healthcare

📚 Academic Citation: Ivchenko, O. (2026). ML for Medical Diagnosis: Research Goals and Framework for Ukrainian Healthcare. Medical ML Diagnosis Series. Odesa National Polytechnic University.

Article #1 in Medical ML for Ukrainian Doctors Series

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

35

Planned Research Articles

12

Week Research Program

5-15%

Human+AI Accuracy Gain

📋 Key Questions Addressed

  1. What is the optimal methodology for systematic review of medical AI literature?
  2. How can research findings be validated through practical application (ScanLab)?
  3. What organizational structure enables sustainable, high-quality research output?

Context: Why This Research Matters

Can machine learning transform medical diagnosis in Ukraine? This article launches a systematic 12-week research program to answer that question. We will examine global best practices, analyze failures, and build a practical framework for integrating ML-powered image classification into Ukrainian clinical workflows—with all findings validated through our open-source ScanLab application.

Research Program Architecture

“`mermaid graph TD A1[Research Methodology] A2[ML Taxonomy] A3[Data Standards] B1[Global Successes] B2[Failure Analysis] B3[Lessons Learned] subgraph Foundation[“Phase 1: Foundation”] A1 A2 A3 end subgraph BestPractices[“Phase 2: Best Practices”] B1 B2 B3 end Foundation –> BestPractices “`

Who We Are: The Stabilarity Ecosystem

Stabilarity Hub is a Ukrainian/cross-national community dedicated to making AI affordable and explainable. We provide hardware, software, engineering support, and research assistance to ML researchers worldwide.

Achievements & Credentials

  • Alpha Startup WebSummit — Global recognition for innovation
  • ONPU Partnership — Collaboration with Odesa National Polytechnic University
  • PhD Research — Ongoing dissertation on data sufficiency for ML in pharmaceutical portfolio optimization
  • Stabilarity OU — Estonian company providing legal and financial structure

Why This Research Matters for Ukraine

Ukrainian healthcare faces unique challenges that make ML-assisted diagnosis particularly valuable:

“`mermaid graph LR C1[Physician Shortage] –> S1[Remote Diagnostics] C2[Infrastructure Constraints] –> S1 C3[Conflict Impact] –> S1 C4[Cost Pressures] –> S2[AI-Assisted Efficiency] C5[Digital Transformation] –> S2 “`

⚠️ The Adaptation Challenge

Most medical AI research originates from well-resourced Western institutions. Direct application of these findings to Ukrainian context requires careful adaptation. That’s what this research program aims to provide.

Research Phases and Timeline

  • Phase 1: Foundation (Weeks 1-2): Research methodology, ML taxonomy, data standards — 6 articles
  • Phase 2: Best Practices (Weeks 3-4): Global successes and failures analysis — 6 articles
  • Phase 3: Technical (Weeks 5-6): CNN, ViT, hybrid architectures, XAI — 6 articles
  • Phase 4: Clinical (Weeks 7-8): Workflow integration, protocols, training — 6 articles
  • Phase 5: Ukrainian (Weeks 9-10): Localization, legal framework, pilot design — 6 articles
  • Phase 6: Framework (Weeks 11-12): Integration, consolidation, publication — 5 articles

Core Research Questions

  • Q1: What is the current state of ML adoption in medical imaging globally? (Article #2)
  • Q2: How do different ML architectures compare for specific imaging modalities? (Articles #4, #13-15)
  • Q3: What data quality and quantity is required for reliable medical ML? (Article #5)
  • Q4: How should physicians interact with AI predictions? (Articles #20-21)
  • Q5: What are the most common failure modes of medical AI? (Article #11)

Research Methodology

“`mermaid sequenceDiagram participant L as 📚 Literature participant R as 🔬 Researcher participant S as 💻 ScanLab participant C as 👨‍⚕️ Clinicians L->>R: Tier 1-5 Source Selection R->>R: Critical Analysis R->>S: Prototype Implementation S->>C: Clinical Validation C–>>R: Feedback Loop R->>L: Updated Research Questions Note over L,C: Continuous iteration ensures practical relevance “`

Source Selection Criteria

  1. Tier 1: Systematic reviews and meta-analyses (Cochrane, JAMA)
  2. Tier 2: Randomized controlled trials and large observational studies
  3. Tier 3: Peer-reviewed technical papers (arXiv with citations, IEEE, ACM)
  4. Tier 4: Industry reports and regulatory documents (FDA, CE, MHSU)
  5. Tier 5: Expert opinion and case studies (with explicit caveat)

Key Preliminary Findings

🔬 Finding #1

Doctor + AI > Either Alone

Human-AI collaboration outperforms both human-only and AI-only diagnosis by 5-15%

🔬 Finding #2

Experience ≠ AI Benefit

Physician experience does not predict who benefits most from AI assistance

🔬 Finding #3

Hybrid Models Win

CNN + Vision Transformer hybrids consistently achieve highest accuracy

🔬 Finding #4

XAI is Essential

Without explainability, physician trust and adoption suffer dramatically

Unique Conclusions

🔬 Conclusion 1: The Research-Practice Gap

Most medical AI research is conducted in high-resource settings with different demographics, equipment, and clinical workflows than Ukraine. Direct transplantation of research findings will fail. A systematic adaptation layer—which this research program provides—is essential for successful deployment.

🔬 Conclusion 2: The Open Source Imperative

Commercial medical AI solutions are often unaffordable for Ukrainian healthcare institutions and lack transparency. Open source alternatives like ScanLab provide a path to AI-augmented diagnosis that is affordable, auditable, and adaptable to local needs.

🔬 Conclusion 3: The Academic-Industry Bridge

Our model—academic partnership (ONPU) + engineering capability (Stabilarity) + open source distribution (ScanLab)—bridges the gap between research and accessible clinical tools.

References

  1. Agarwal, N. et al. “Heterogeneity and predictors of the effects of AI assistance on radiologists.” Nature Medicine, 2024. https://doi.org/10.1038/s41591-024-02850-w
  2. Chen, R.J. et al. “A pathologist–AI collaboration framework.” Nature Biomedical Engineering, 2024. https://doi.org/10.1038/s41551-024-01223-5
  3. Ly, N. et al. “Recent Advances in Medical Image Classification.” arXiv:2506.04129, 2025. https://arxiv.org/abs/2506.04129
  4. “Enhanced early skin cancer detection through EViT-DenseNet169.” Scientific Reports, 2025. https://doi.org/10.1038/s41598-025-85234-7
  5. “Artificial intelligence based classification using inverted self-attention DNN.” Scientific Reports, 2025. https://doi.org/10.1038/s41598-025-87456-3

Author: Oleh Ivchenko, PhD Candidate
Affiliation: Odessa Polytechnic National University | Stabilarity Hub