Figure 2.

The flow of the classification and visualization in the 3D SE-VGG-11BN CNN model. This model consists of a modified 3D VGG-11 network with squeeze-and-excitation (SE) block and batch-normalization (BN) using T1W MRI as the model input. The class of one given T1W scan is predicted by two steps in the model: (1) extracting hierarchical features, and (2) classifying these features. In the feature extractor portion, the data is firstly under-sampled × 2 and goes through several convolution blocks consisting of 3D convolution, 3D batch normalization, 3D max pooling, and 3D SE operation. The classifier consisting of three dense layers with dropout regularization yields the final prediction result. The classifier consisting of three dense layers with dropout regularization yields the final prediction result. In the feature extractor part, feature maps generated by filters at the last convolution layer are shown. These feature maps are used for visualization through the generation of the class activation map by weighting them with channel-wise average gradients.