Answers are grounded in this paper and cite their sources.
Computer vision can effectively assist doctors in the early detection and analysis of breast cancer; however, existing methods still have significant limitations. CNNs struggle with long-range dependencies, Transformers incur high computational costs, and Mamba’s autoregressive approach fails to effectively leverage the strong correlations in local regions when processing images. Additionally, issues in cross-scale feature fusion and loss function design hinder the simultaneous optimization of small-lesion detection, boundary accuracy, and segmentation integrity. We propose CMAMUNet, where the HSMam module integrates CNN-based local detail extraction with Mamba-based long-range dependency modeling. The Spatial Attention Bridge (SAB) and Channel Attention Bridge (CAB) enable precise cross-scale feature fusion, while the multidimensional loss function effectively enhances small-lesion detection, boundary precision, and global completeness. Experimental results on three public datasets—BUSI, Dataset-B, and CBIS-DDSM—show that CMAMUNet outperforms 14 state-of-the-art models in key metrics. Additionally, with a parameter count of 19.1 M and FLOPs of 35.2G, the model is well-suited for clinical deployment.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: