Mental Disease Feature Extraction with MRI by 3D Convolutional Neural Network with Multi-channel Input

Magnetic resonance imaging (MRI) plays an important role in early diagnosis, which can accurately capture the disease variations of the anatomical brain structure. We propose a novel method for improving feature extraction performance from magnetic resonance images (MRI). This study presents a combination of multi-channel input and 3D convolutional neural network architecture which can reduce the feature dimensionality. Multi-channel input scheme is devised to apply prior knowledge on MRI original inputs in order to overcame possibly uncertainty and unsteadiness on the final features. While, the 3D-CNN model can simultaneously extract features from spatial and temporal dimensions for purpose of capturing the variations of constructive information.