GeoCFNet: Geometry-Aware Confidence Field Network for Robot-Assisted Endoscopic Submucosal Dissection
Researchers have developed GeoCFNet, a novel geometry-aware confidence field network designed to enhance visual guidance for robot-assisted endoscopic submucosal dissection (ESD). This network addresses challenges in dynamic endoscopic environments, such as smoke and tissue deformation, by estimating dissection corridors and safe tissue margins with improved accuracy and geometric stability. GeoCFNet integrates a Token-Differentiated Fusion module and Geometry-Aware Spatial Regularization (GASR) to preserve spatial coherence, achieving strong performance metrics like an RMSE of 0.0480. AI
IMPACT This AI model could improve precision and safety in robot-assisted surgical procedures, potentially leading to better patient outcomes.