An Explainable AI System for the Diagnosis of High Dimensional Biomedical Data

AI can match expert performance in medical diagnosis, but transparency is essential: The paper demonstrates that AI systems can diagnose complex biomedical data (flow cytometry samples with over 100,000 cells) with accuracy comparable to human experts, yet emphasizes that such high-stakes medical decisions require explainability because they directly affect patient health and life.

Explainability can be achieved through domain-expert language: ALPODS generates "fuzzy reasoning rules expressed in the typical language of domain experts," meaning the AI's diagnostic logic is translated into medical terminology that clinicians already understand, rather than requiring them to interpret black-box algorithms.

Identifying patient subpopulations helps justify AI decisions: Rather than treating all patients identically, the system classifies cases based on clusters or subpopulations within the data, allowing doctors to see which patient groups the AI identified and why specific diagnoses were recommended for each group.

Visualization of AI reasoning improves clinical trust: The paper includes a visualization method based on the generated rules that allows human experts to directly observe and verify the AI system's reasoning process, supporting the responsible deployment principle that clinicians should understand—not just accept—AI recommendations.

Explainable AI can be practical for routine clinical use: ALPODS performs efficiently on both benchmark datasets and "everyday routine case data," suggesting that explainability doesn't require sacrificing real-world applicability in clinical settings.