Fig. 2.

The CNN comprised four convolutional layers with channels of 8, 16, 16, and 32. Because the distance between two elements within the connectome matrix does not represent spatial distance as in general images, we customized the structure of the CNN: a a filter size of 1 was used for all convolutional layers and b max-pooling layers were omitted from the CNN. The processed data were further processed at three fully connected layers with the number of units of 4096, 4096, and 2. Conv, convolutional layer; FC, fully connected layer; ker, numbers of kernels of the convolutional layer; out, numbers of output values of the fully connected layer