XAI-MeD: Explainable Knowledge Guided Neuro-Symbolic Framework for Domain Generalization and Rare Class Detection in Medical Imaging

Expert knowledge improves AI reliability: XAI-MeD encodes clinical expertise as logical rules within the AI system, achieving a 6% performance improvement in cross-domain generalization—demonstrating that incorporating doctor knowledge helps AI systems work better across different hospitals and patient populations.

Rare diseases get better detection: The framework achieved a 10% improvement in detecting rare clinical conditions (measured by F1 score), addressing a critical gap where standard AI often misses infrequent but serious diagnoses like uncommon seizure patterns or advanced diabetic retinopathy.

Explainability enables clinical trust: By providing transparent, clinically-aligned interpretations through symbolic reasoning, the system allows doctors to understand why the AI made a diagnosis—essential for responsible deployment in medical settings where clinicians need to verify and trust AI recommendations.

Robustness under real-world conditions: The system maintains performance when tested across different medical imaging datasets and conditions (including seizure detection from brain scans and diabetic retinopathy grading), addressing the practical problem that AI trained in one hospital often fails in another.

Addressing AI bias systematically: The framework specifically targets class imbalance and uncertainty issues that cause AI systems to exhibit bias against underrepresented patient groups, using adaptive mechanisms to ensure fair performance across all patient populations.