Surgical Scene Segmentation

Surgical scene segmentation aims to automatically identify and delineate different objects (e.g., instruments, organs, tissues) within intraoperative images and videos, facilitating improved surgical workflow understanding and potentially autonomous robotic surgery. Current research focuses on adapting and improving existing deep learning models, including transformers and foundation models like Segment-Anything, to handle the challenges of limited annotated data, geometric variations in surgical scenes, and the domain gap between natural and surgical images. These advancements are crucial for developing safer and more efficient surgical procedures, enabling computer-assisted interventions, and improving surgical training and assessment.