Figure 1.

A case for the segmentation process. The Unet model was used for the segmentation of the breast tumors. The down-sample layer is the blue module in the figure, which is realized by max-pooling. The up-sample layer is the red module in the figure, which is realized by deconvolution. Moreover, skip connection is represented by a gray line, which combines low-level features and high-level semantic features to realize up-sampling step by step. Finally, the feature map is converted into the probability graph through softmax operation.