Fig. 2

The DA model used in this study to extract deep radiomic features. There are two convolutional layers and two max-pooling layers in the encode phase, two convolutional layers, and two upsampling layers in the decode phase. For the (None, n, n, m), “None” is the batch size which we used to load the samples into the model, (n, n) in the middle indicates the number of features in each kernel. m is the number of kernels