Learning Connectivity Patterns via Graph Kernels for fMRI-Based Depression Diagnostics

Maksim Sharaev, Alexey Artemov, Ekaterina Kondrateva, Sergei Ivanov, Svetlana Sushchinskaya, Alexander Bernstein, Andrzej Cichocki, Evgeny Burnaev

Abstract: It has long been known that patients with depression exhibit abnormal brain functional connectivity patterns, that are often studied from a graph-theoretic perspective. However, while certain simpler graph features have been examined, little has been done in the direction of advanced feature learning methodologies such as network embeddings. Our work aims to extend the understanding of importance of graph-based features for medical applications by evaluating the recently proposed anonymous walk embeddings (AWE) in difficult depression classification problems. For two challenging datasets, we obtain performance gains and investigate the learned vector representations. Our results indicate that using AWE-based features is a promising new direction for medical applications.