CLEAR-HPV: Interpretable concept discovery for human-papillomavirus-associated morphology in whole-slide histology
Researchers have developed CLEAR-HPV, a new framework designed to improve the interpretability of AI models used in analyzing whole-slide histopathology images for human papillomavirus (HPV) detection. This method restructures the latent space of attention-based multiple instance learning models to automatically discover and map morphologic concepts like keratinizing, basaloid, and stromal features. The framework reduces high-dimensional data to a compact vector of interpretable concepts, maintaining predictive accuracy across different cancer datasets. AI
IMPACT Enhances interpretability of AI in medical diagnostics, potentially improving clinician trust and understanding of model predictions.