Related Experiment Video
Updated: Jul 9, 2026
12:03
Multi-photon Imaging of Tumor Cell Invasion in an Orthotopic Mouse Model of Oral Squamous Cell Carcinoma
Published on: July 25, 2011
FS-Mamba: Feature-wise scanning Mamba UNet for automatic image segmentation in liver tumor radiotherapy
Peijun Yin1, Xueren Zhang2, Xin Liu3
1Department of Radiation Physics, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China.
Journal of Applied Clinical Medical Physics
|July 8, 2026
Summary
FS-Mamba offers accurate and efficient medical image segmentation for liver cancer radiotherapy, significantly improving speed and reducing model size compared to existing methods.
Area of Science:
- Medical imaging and computational anatomy
- Radiotherapy planning and delivery
- Artificial intelligence in healthcare
Background:
- Accurate medical image segmentation is crucial for radiotherapy, but challenges persist due to anatomical variations and low contrast in CT scans.
- Delineating organs at risk and tumor targets requires high precision for effective radiotherapy planning.
Purpose of the Study:
- To develop a lightweight, high-precision automatic segmentation network for online adaptive liver cancer radiotherapy.
- The network aims to meet clinical demands for both accuracy and computational efficiency.
Main Methods:
- Proposed FS-Mamba, a U-Net architecture incorporating a novel Frequency-Long SSM Block with flip-selective scanning, frequency-domain modeling (FFT), and a long-memory state space model (LSSM).
- Validated on CT-ORG, Synapse, and an in-house liver cancer dataset, comparing against UNet, nnUNet, TransUNet, Swin-UMamba, and TotalSegmentator using DSC, HD95, and subjective scoring.
Main Results:
- FS-Mamba achieved superior segmentation performance (e.g., 92.81% DSC on CT-ORG) across all datasets, outperforming existing methods (p < 0.05).
- Received high subjective scores from medical physicists (4.69/5.0), nearing ground-truth accuracy.
- Demonstrated significant efficiency with the smallest model size (20.57M parameters) and fastest inference (20.16ms/case), offering 4.5x parameter reduction and 6.5x speedup over TotalSegmentator.
Conclusions:
- FS-Mamba presents a clinically viable solution for automated multi-organ and tumor segmentation in radiotherapy.
- The model meets stringent time constraints for online adaptive radiotherapy while delivering expert-level accuracy.
