Research develops segmentation and representation learning for medical images and multimodal foundation encoders; focuses on adversarial robustness, privacy-preserving federated training, domain adaptation, semi-supervision, and efficient inference to enable scalable, clinically usable systems.