Knowledge Enhanced Graph Neural Network for Advanced Classification of Depression in fMRI Imaging

fMRI data demonstrates considerable efficacy in diagnosing depression. However, it is challenged by high dimensionality and limited sample sizes, hindering the effectiveness of traditional data processing methods. In response, we introduce a data and knowledge dual-driven deep learning model capable of adeptly processing unstructured fMRI data, even in small-sample scenarios. Our work features the development of an innovative graph convolutional network that dynamically updates node and edge information while integrating prior medical statistical knowledge. This synergy of deep learning with medical insights enables a more precise differentiation between individuals with depression and healthy subjects, offering fresh perspectives in the search for depression biomarkers. This approach promises to enhance the tools and strategies available for early depression diagnosis and treatment, promoting a substantial stride in mental health research.

1. INTRODUCTION

Over the past few decades, functional magnetic resonance imaging (fMRI) has been widely applied in the field of neuroscience research. By detecting functional connectivity between different regions of the brain, it can be used to assess patients’ cognitive functions, emotional states, and pain perception [1]. The diagnosis and treatment of depression is currently a hot research topic in fMRI. Zhang et al. [2] used resting-state fMRI (rs-fMRI) to scan brain regions of patients with depression and healthy individuals, finding changes in spontaneous neural activity in patients’ right anterior cingulate, superior frontal gyrus, and hippocampus compared to healthy individuals. Yan et al. [3] explored brain region changes in patients with varying degrees of depression using rs-fMRI, discovering that the regional homogeneity (ReHo) values in certain brain areas of depressed patients increased with the severity of the condition. These studies provide a basis for understanding the pathophysiological mechanisms of depression.