-
1
MemSAM-2.5D is a unified 2.5D segmentation framework designed to improve liver tumor segmentation from 3D CT volumes.
-
2
The framework incorporates a Hybrid Mamba-Adapter, Z-axis State Flow module, and Confidence-Gated Prototype Memory for enhanced performance.
-
3
MemSAM-2.5D addresses challenges such as multi-scale variation, volumetric discontinuity, and ambiguous tumor boundaries in liver tumor segmentation.
-
4
Extensive evaluations show that MemSAM-2.5D outperforms CNN-based, Transformer-based, and other baseline models in segmentation accuracy.
-
5
The proposed framework demonstrates effective and transferable solutions for clinically relevant segmentation of hepatocellular carcinoma.
