Segmentation of Breast Ultrasound Images Based on CBAM-Extended Residual U-Net

Breast cancer has a high malignancy rate and poses a serious threat to women’s health. In China, ultrasound imaging is the primary method for breast cancer screening, and precise lesion segmentation is crucial for diagnosing benign and malignant cases. The traditional U-Net model concatenates feature maps from the encoder and decoder, which introduces redundant information, and its skip connections struggle to transmit longrange contextual information. This study proposes a U-Net model based on attention mechanisms and extended residual convolution to address these issues. The extended residual convolution replaces traditional convolution modules to expand the receptive field and capture multi-scale features. An improved CBAM attention module is added to the skip connections to enhance the combination of shallow and deep features. Additionally, the DiceIoULoss function is used to balance the Dice and IoU coefficients. Experimental results show that the proposed method outperforms the original U-Net model, with improvements of 8.6%, $23.7 \%, 5.6 \%$, and $5.1 \%$ in IoU, Dice, Recall, and F1-score, respectively. These results demonstrate the model’s effectiveness in breast nodule lesion segmentation.