AOTUMATIC RESTING STATE NETWORKS LABELLING

Abstract

A 3-D convolutional neural network (3-D CNN) is proposed for the classi cation of functional brain networks. fMRI technique is widely used to image the neuronal activity (which results in formation of functional networks) while lying static inside a MRI machine. The idea is that the body at rest can simulate the neuronal activity of that when engaged in extrinsic tasks. The study of neural activity is signi cant for understanding the working of brain. It is believed that low-frequency uctuations observed in the BOLD signals re ect the spontaneous neural activity and that the synchronized

uctuations in distinct brain regions, therefore, point to functional connections between them. Di erent functional connectivity networks have been found, and these networks change in patients with multiple pathologies (neurological, psychiatric). Determination of networks correctly is essential for getting in depth understanding of brain functioning or understanding the di erences between brain region connectivity of a normal being and a patient for diagnostic and clinical purposes. Machine learning techniques became very popular in the eld of resting state fMRI network based classi cation. However, the application of convolutional neural networks has been proposed only very recently and has remained largely unexplored. A 3-D CNN is designed to classify the functional networks with more speed and accuracy.