Supplementary Materials of ICAM: Interpretable Classification via Disentangled Representations and Feature Attribution Mapping

A.1 Network Architecture Our encoder-decoder architecture for a 3D input is shown in Fig. A.1. The architecture for a 2D input is the same, only using 2D convolutions and a 2D attribute space. The key components of the attribute encoder include using down ResNet blocks (with average pooling, and leaky ReLU activation) for encoding the input image into a relatively large 3D latent space of size 8 10 8 (in the 3D case), as opposed to a 1D vector, which is commonly seen in Variational Autoencoders (VAEs). We also added a fully connected layer to the attribute latent space to enable classification. In early development, we found that using a 1D vector in the latent space was insufficient for encoding the required class information for brain imaging, and observed that some class information was instead encoded in the content encoder, which is meant to be class invariant.