Research Article

Segmentation of Breast Tubules in H&E Images Based on a DKS-DoubleU-Net Model

Figure 3

The illustrative tubule segmentation results of different models on the test set of the private clinical dataset. The first row lists the original images; images a(1)-f(1) in the second row are the corresponding tubule annotation masks; images a(2)-f(2) in the third row are the tubule segmentation results performed by DKS-DoubleU-Net; images a(3)-f(3) in the fourth row are the results output by the DoubleU-Net which added DenseNet; images a(4)-f(4) in the fifth row are the results obtained by the DoubleU-Net which added Kernel Selecting Module (KSM); images a(5)-f(5) in the sixth row are the results output by the DoubleU-Net; images a(6)-f(6) in the seventh row are the results output by Unet; images a(7)-f(7) correspond to the results of ResUnet++; images a(8)-f(8) are the results obtained by DeepLabV3+, and images a(9)-f(9) show the tubule segmentation results of HRNet.