1 读取dicom图像中起点位置为1010大小为100100的窗口；2 将窗口图像输入卷积神经网络第一层包括两组卷积核两组卷积核参数分别为Sobel滤波的水平滤波器以及垂直滤波Same填充输入与输出大小一致步长为1；将输出的结果经过ReLu非线性映射；再将输出结果输入第二层卷积第二层卷积包括两组卷积核第一组卷积对应各通道滤波器参数均为Sobel水平滤波器第二组卷积对应各通道滤波器参数均为Sobel

Read the window with starting position (10,10) and size (100,100) from the DICOM image.
Input the window image to a convolutional neural network. The first layer includes two sets of convolutional filters, with one set using the Sobel horizontal filter and the other set using the Sobel vertical filter. Same padding is used to keep the input and output sizes equal, and the stride is set to 1. The output is passed through the ReLU activation function. The output of the first layer is then input to the second layer, which includes two sets of convolutional filters. The first set applies the Sobel horizontal filter to all input channels, while the second set applies the Sobel vertical filter to all input channels. The bias values for all filters are set to 0. The output is again passed through the ReLU activation function. The resulting feature maps are displayed. The output feature maps are then input to a max pooling layer with a filter window size of 2x2, no padding, and a stride of 2. The resulting feature maps are again displayed. Finally, the output is passed through a fully connected layer with a single output node, using randomly assigned weights. The output is then passed through the sigmoid activation function to obtain the final mapping result.