Constructing fine-granularity functional brain network atlases via deep convolutional autoencoder

The proposed computational framework. (a) Sparse representation of fMRI data for generation of functional network maps in each individual. Then 10 RSNs (resting state networks) were manually labeled for these networks via sparse coding methods for each task/resting state fMRI scan of each subject. (b) Training 3D CAE (convolutional autoencoder) using all the labeled networks, and extracting features using the trained encoder. (c) Network clustering using the extracted features and generation of fine-granularity ICNs (intrinsic connectivity networks).178