A XAI-based Framework for Frequency Subband Characterization of Cough Spectrograms in Chronic Respiratory Disease

AI transparency matters for medical diagnosis: The study uses explainable AI (XAI) techniques—specifically "occlusion maps"—to show which parts of cough sound recordings the AI actually uses to make decisions, rather than treating the AI as a "black box." This transparency is crucial for clinicians to trust and validate AI-assisted diagnostic tools.

Cough sounds contain disease-specific patterns: The research found that different respiratory diseases produce distinct patterns across five frequency ranges within cough spectrograms, meaning chronic conditions like COPD leave measurable acoustic "fingerprints" that could potentially support non-invasive diagnosis.

Interpretability enables clinical insight: By breaking down the AI's decisions into specific frequency subbands, the framework reveals how different parts of the sound spectrum relate to underlying disease mechanisms—moving beyond prediction to actual understanding of why the AI makes its recommendations, which is essential for responsible clinical deployment.

Distinguishing disease severity is possible: The framework successfully differentiated between chronic and non-chronic respiratory conditions, suggesting that acoustic analysis could help clinicians assess disease progression or severity, though further validation would be needed before clinical use.