XAI-CLIP: ROI-Guided Perturbation Framework for Explainable Medical Image Segmentation in Multimodal Vision-Language Models

Researchers developed XAI-CLIP, a new method to explain how AI models segment medical images like CT scans. It uses language-guided cues to focus on relevant anatomy, generating clearer visual explanations. The system is 60% faster than current methods and significantly improves explanation accuracy, with a 44.6% better dice score. This addresses a key barrier to clinical trust by making AI decisions more transparent and interpretable for doctors.