Automatic Segmentation Method of Breast Tumor Ultrasonic Images Based on Attention-Enhancing Unet

Abstract

In order to improve the accuracy of the segmentation of the breast ultrasound image lesion, Attention-Unet was improved, and an Attention-enhancing Unet (AE-Unet) model is proposed. First, the network loss function was improved. Based on the output value of the traditional network end, output weights of all attention gate were integrated. Compared with the standard lesion template, it was used to obtain accurate network loss values; Secondly, the network training method was improved, and the strategy of combining thickness and fineness was adopted. The overall loss function was used to train the overall network to make the network basically stable; then the partial loss function was used to alternately train the backbone network and the attention gate module in turn. Fine-tuning was used to further improve the accuracy of network parameters. The combination of the two greatly improves the accuracy of segmentation of the breast ultrasound lesion area. The experimental results on the breast ultrasound data actually collected in the hospital show that the proposed AE-Unet model has an M-IOU of 81.24%, precision of 85.88%, F1 of 80.58%, Acc of 93.85% and specificity of 97.48%, PPV is up to 85.88%, which has achieved better segmentation results than existing advanced algorithms.