International Journal of Imaging Systems and Technology, Journal Year: 2024, Volume and Issue: 34(6)

Published: Nov. 1, 2024

ABSTRACT Breast cancer remains one of the most significant health threats to women, making precise segmentation target tumors critical for early clinical intervention and postoperative monitoring. While numerous convolutional neural networks (CNNs) vision transformers have been developed segment breast from ultrasound images, both architectures encounter difficulties in effectively modeling long‐range dependencies, which are essential accurate segmentation. Drawing inspiration Mamba architecture, we introduce Vision Mamba‐CNN U‐Net (VMC‐UNet) tumor This innovative hybrid framework merges dependency capabilities with detailed local representation power CNNs. A key feature our approach is implementation a residual connection method within utilizing visual state space (VSS) module extract features maps effectively. Additionally, better integrate texture structural features, designed bilinear multi‐scale attention (BMSA), significantly enhances network's ability capture utilize intricate details across multiple scales. Extensive experiments conducted on three public datasets demonstrate that proposed VMC‐UNet surpasses other state‐of‐the‐art methods segmentation, achieving Dice coefficients 81.52% BUSI, 88.00% BUS, 88.96% STU. The source code accessible at https://github.com/windywindyw/VMC‐UNet .

Language: Английский