BMSNet: a boundary-guided multi-scale polyp segmentation network

Clinical Snapshot

Key Findings

• P Value: Not reported

• Effect Size: Dice scores: ClinicDB 0.943, Kvasir 0.920, ColonDB 0.828, ETIS 0.825, CVC-300 0.913

• Primary Outcome: Dice similarity coefficient for polyp segmentation across five benchmark datasets

• Nnt Or Sensitivity: Sensitivity and specificity not reported; Dice score is a segmentation overlap metric, not a clinical diagnostic accuracy measure. No NNT, likelihood ratios, or AUC-ROC values provided.

• Confidence Interval: Not reported

Clinical Application

Computational complexity is described as 'moderate' but no specific inference time, GPU requirements, or real-time processing capability data are provided in the abstract. Integration into live colonoscopy workflow would require real-time performance (typically <50ms per frame), which is not assessed. Clinical deployment would require TGA regulatory approval as a Class IIb or Class III medical device software (SaMD) under the Australian Therapeutic Goods Act. Australia has a National Bowel Cancer Screening Program (NBCSP) with high colonoscopy volumes. Computer-aided detection (CADe) and computer-aided characterisation (CADx) tools are an active area of clinical interest, with some systems (e.g., GI Genius, EndoBRAIN) having received international regulatory clearance. BMSNet has not been evaluated in Australian colonoscopy datasets, has not received TGA approval, and has not been assessed against RACGP or GESA (Gastroenterological Society of Australia) colonoscopy quality benchmarks. PBS reimbursement for AI-assisted colonoscopy does not currently exist in Australia. This study does not provide sufficient clinical evidence to support adoption in Australian practice. In principle, adult patients undergoing colonoscopy for colorectal cancer screening or surveillance in whom computer-aided polyp detection/delineation tools are deployed. However, the current evidence base supports application only to benchmark image datasets, not to prospective clinical populations.

Abstract

Accurate polyp segmentation in colonoscopy images is essential for computer-aided diagnosis and early colorectal cancer screening. However, this task remains challenging due to large variations in polyp size, shape, and texture, as well as blurred boundaries and low contrast between polyps and surrounding mucosa. To address these challenges, we propose boundary-guided multi-scale polyp segmentation network (BMSNet), a boundary-guided multi-scale network for accurate polyp segmentation. First, an Scale-Aware Modulation Meets Transformer-T backbone initialized with officially released pretrained weights is adopted to extract multi-level features efficiently. Then, a boundary prediction module is introduced to generate preliminary boundary priors through bidirectional feature interaction and differentiable Canny-based auxiliary supervision. Based on the generated boundary information, we design a boundary-guided feature enhancement module to refine feature representations. This module combines frequency-domain enhancement, multi-scale feature interaction, and boundary-guided modulation. Finally, a feature fusion unit is introduced to progressively integrate multi-scale features and generate the final segmentation prediction. Extensive experiments on five benchmark datasets demonstrate the effectiveness of the proposed method. On ClinicDB, Kvasir, ColonDB, ETIS, and CVC-300, BMSNet achieves Dice scores of 0.943, 0.920, 0.828, 0.825, and 0.913, respectively. These results indicate that BMSNet achieves competitive segmentation performance with moderate computational complexity under the evaluated experimental setting.

References

1. 1.Gao, T., Song, Z., Yang, Z., Chen, Y., & Zhang, Y. (2026). BMSNet: a boundary-guided multi-scale polyp segmentation network. Biomedical Physics & Engineering Express. https://doi.org/10.1088/2057-1976/ae7df5