Classification of major depressive disorder using functional connectome data

Abstract

The brain consists of billions of neurons, communicating with each other to give humans cognitive, sensing and reasoning abilities. Major Depressive Disorder (MDD) has affects around 4% of the world population. MDD has debilitating effects on both physical and emotional health of an individual. It is important to study the brain biomarkers of the disease and develop methods to improve its diagnosis. In recent years, there has been a rise in research related to brain state classification using data from different neuroimaging modalities. In this work, we use features obtained from fMRI scans to distinguish patients suffering from Major Depressive Disorder (MDD) and Normal control (NC).

We used both supervised and unsupervised methods to classify/cluster patients from healthy controls. For supervised techniques, we used Support Vector Machines (SVM), Feedforward Neural Network (FNN) and Convolutional Neural Networks (CNN), whereas for unsupervised technique we used clustering on features derived from the functional connectome of each subject. We propose an extension to existing methods for extracting weighted Graphlet Degree Vectors (w-GDV) and use the derived features for clustering. We achieved an accuracy of 78.3% with a deep feed-forward neural network, while with unsupervised clustering using euclidean distance, we achieved a cluster purity of 52%