FIGURE 9.

U‐Net++ structure and its application in MRI images. Up‐sampling and down‐sampling are used to adjust the size and resolution of images or feature maps. Skip connections are used to mitigate the vanishing gradient problem in deep neural networks by connecting the output of one layer directly to the input of another layer several layers away. Deep supervision is a technique that adds additional supervision signals at intermediate layers to improve the training of deep neural networks. The specific implementation of deep supervision in U‐Net++ is to add a 1 × 1 convolution kernel after X^0,1, X^0,2, X^0,3, X^0,4, which is equivalent to supervising the output of U‐Net at each level or each branch. MRI, magnetic resonance imaging.