FS-Mamba: Lightweight Mamba UNet Achieves Expert-Level Multi-Organ Segmentation for Liver Cancer Radiotherapy

On an in-house liver cancer radiotherapy dataset, FS-Mamba achieved 92.68% DSC and a clinician score of 4.69/5, while being 4.5× smaller and 6.5× faster than TotalSegmentator (20.6M params).

• Retrospective validation on 50-patient in-house liver cancer dataset plus two public benchmarks (140 and 30 cases).
• Highest subjective score (4.69/5) from 10 blinded physicists, approaching ground truth (4.94/5).
• Main limitation: single-center dataset without external clinical validation; small sample size.

RadPigeon summaries are original and for information only. They are not clinical advice.