A Feedback Learning Segmentation Algorithm for Medical Images of Breast Cancer Incorporating Attention Mechanism

The study of breast cancer image lesion segmentation algorithms is gradually playing an increasingly important role in the medical field to assist physicians in diagnosis. To further segment lesions precisely, a network model with UNet as a substrate and a fusion of attention mechanisms is proposed. The model is trained using breast cancer image data. An image chunking training strategy is used in the model training process. The maximum pooling is replaced by adaptive adjustment pooling in the downsampling process. The SE module is introduced at the bottom layer of the model. A five-in-one feature fusion strategy is added before convolution operation. A model feedback learning from deep to shallow layers is added to the bottom three layers of the model. The experimental results show that this algorithm improves the accuracy compared to the original UNet, residual ResUnet, and CBAM for the segmentation of breast cancer lesions.