DistMedVL Probabilistic Framework for Medical Image Segmentation

DistMedVL introduces a Probabilistic Cross-Modal Adapter (PCM-Adapter) to handle aleatoric and epistemic uncertainty in medical vision-language tasks. The framework uses a Mahalanobis Alignment Module to model textual tokens as Gaussian distributions, improving segmentation robustness against domain shifts.