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

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

**Article #1 in Medical ML for Ukrainian Doctors Series**

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

—

—

—

## 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]
“`

—

## 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.

Project	Focus	Status
Flaidata.com	Data analytics platform	Production since 2022
Gromus.ai	AI platform	Active
medai-hack.com	Medical AI hackathons	Ongoing events
ScanLab	Medical imaging (open source)	Active development

### 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
– **Team** — 70+ engineers with experience across enterprise, finance, telecom, and healthcare domains

—

## Why This Research Matters for Ukraine

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

“`mermaid
graph LR
C1[Physician Shortage]
C2[Infrastructure Constraints]
C3[Conflict Impact]
C4[Cost Pressures]
C5[Digital Transformation]
S1[Remote Diagnostics]
“`

—

## Research Phases and Timeline

Phase	Duration	Focus	Deliverables
1. Foundation	Weeks 1-2	Research methodology, ML taxonomy, data standards	6 articles, methodology framework
2. Best Practices	Weeks 3-4	Global successes and failures analysis	6 articles, lessons learned database
3. Technical	Weeks 5-6	CNN, ViT, hybrid architectures, XAI	6 articles, model comparison matrix
4. Clinical	Weeks 7-8	Workflow integration, protocols, training	6 articles, protocol templates
5. Ukrainian	Weeks 9-10	Localization, legal framework, pilot design	6 articles, adaptation guidelines
6. Framework	Weeks 11-12	Integration, consolidation, publication	5 articles, comprehensive framework

—

## Core Research Questions

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

—

## 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.
2. Chen, R.J. et al. “A pathologist–AI collaboration framework.” *Nature Biomedical Engineering*, 2024.
3. Ly, N. et al. “Recent Advances in Medical Image Classification.” arXiv:2506.04129, 2025.
4. “Enhanced early skin cancer detection through EViT-DenseNet169.” *Scientific Reports*, 2025.
5. “Artificial intelligence based classification using inverted self-attention DNN.” *Scientific Reports*, 2025.

—

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