SGP-SAM: Self-Gated Prompting for Transferring 3D Segment Anything Models to Lesion Segmentation
SGP-SAM addresses challenges in transferring 3D segmentation models like SAM to medical lesion segmentation by using a self-gated prompting framework. It introduces a module that conditionally enhances features at multiple scales when needed, combined with a Zoom Loss function to focus supervision on small lesions and improve performance in 3D volumes.