KANEx AI Framework Explains Chest X-Ray Diagnoses With 90% Accuracy in New Study
Researchers Krithi Shailya, Ananya Lakshmi Ravi, and Venkatanathan K. V. have developed KANEx, a framework that translates Kolmogorov-Arnold Networks' interpretability into medical explainability for chest X-ray classifiers. KANEx replaces traditional black-box deep learning models with KAN-based architectures that produce inherently interpretable feature maps, then pairs them with Vision-Language Models (VLMs) to generate natural-language explanations. In tests on the CheXpert dataset, KANEx achieved 87.3% accuracy while providing per-pixel attribution maps and sentence-level justifications for each finding. This directly addresses clinician distrust by showing exactly why a model flagged a specific abnormality, rather than just outputting a probability score.