The invention discloses a deep learning-based method for segmenting a breast cancer pathological image HE cancer nest, comprising: S1, inputting a piece of HE WSI, systematically segmenting a model, and under 1x, segmenting the contour area of the tissue in the slice; S2, segmenting 1x map the area under 40x, and extract the corresponding area; S3, crop the extracted area into a Patch with a size of 1024*1024 and overlap 128 pixels; S4, increase the magnification of all Patches to 80x; S5, increase the height of The resolution results are input into the semantic segmentation model, and the model outputs the segmentation Mask of each Patch; S6, merge each Mask according to the cropped coordinates to generate a complete binary Mask image; S7, perform morphological operations on the merged binary image , and extract contours hierarchically. The invention adopts the deep neural network for segmentation, which has stronger generalization ability and higher robustness, adopts the overlapping method, designs the processing mechanism of the boundary effect, and can effectively avoid the boundary effect.