VITAL: Visual-Semantic Dual Supervision for Enhanced and Interpretable Latent Reasoning in Medical MLLMs
VITAL introduces a latent-space reasoning framework for medical multimodal large language models (MLLMs) using visual-semantic dual supervision to prevent modality collapse. It uses a recurrent loop for continuous state reasoning, ensuring train-inference parity and zero inference overhead. The framework achieves state-of-the-art performance on medical benchmarks while enabling post-hoc dual interpretability.