SGP-SAM: Self-Gated Prompting for 3D Lesion Segmentation
Researchers propose SGP-SAM, a self-gated prompting framework to transfer 3D Segment Anything Models (SAM) to lesion segmentation. The key component, Self-Gated Prompting Module (SGPM), uses a lightweight multi-channel gating unit to conditionally activate multi-scale feature fusion, addressing weak spatial representation for small, irregular targets and extreme foreground-background imbalance in 3D volumes. The method aims to improve segmentation of lesions in medical imaging.
Key facts
- SGP-SAM is a self-gated prompting framework for 3D lesion segmentation.
- It transfers 3D SAM-style models to medical imaging.
- The Self-Gated Prompting Module (SGPM) performs conditional multi-scale spatial enhancement.
- A lightweight multi-channel gating unit predicts whether features need multi-scale fusion.
- It addresses weak spatial representation for small, irregular targets.
- It tackles extreme foreground-background imbalance in 3D volumes.
- The method enriches spatial context via Multi-Scale Feature Fusion Block.
- The paper is published on arXiv with ID 2604.22825.
Entities
Institutions
- arXiv