XCoOp: Explainable Prompt Learning for Computer-Aided Diagnosis via Concept-guided Context Optimization

Explainability is achievable without sacrificing performance: XCoOp demonstrates that medical AI systems can provide interpretable explanations while maintaining or improving diagnostic accuracy, challenging the common assumption that explainability requires trading off model performance.

Clinical concepts make AI more trustworthy: By aligning AI decisions with established medical concepts rather than opaque learned patterns, the framework creates a bridge between how AI systems and clinicians think about diagnoses, potentially increasing trust in high-stakes healthcare settings.

Large language models can fill knowledge gaps: The research shows that when concept annotations are unavailable, large language models can be leveraged to generate the medical knowledge needed to make AI systems more interpretable, making explainable AI more practical to implement.

Multiple forms of explanation matter: The framework provides both visual explanations (showing which image regions matter) and textual explanations (describing the clinical reasoning), recognizing that different stakeholders may need different types of evidence to understand and validate AI recommendations.

Foundation models enable responsible AI development: The work demonstrates that powerful pre-trained vision-language models like CLIP can be adapted to support explainability requirements in healthcare, suggesting that foundation models may be valuable tools for building AI systems that meet regulatory and ethical standards in regulated industries.