Predicting cortical-thalamic connectivity using functional near-infrared spectroscopy and graph convolutional networks

Abstract

Functional near-infrared spectroscopy (fNIRS) measures cortical changes in hemoglobin concentrations, yet cannot collected this information from the subcortices. To address this drawback, we propose a machine-learning-based approach to predict cortical-thalamic functional connectivity using cortical fNIRS data. We applied graph convolutional networks (GCN) on two datasets obtained from adults and neonates, respectively. Each dataset contained fNIRS data as input to the predictive models and connectivities of functional magnetic resonance imaging (fMRI) as training targets. GCN models performed better compared to conventional methods, on both identifying the connections, and regressing the quantified strengths of connections. We also propose the addition of inter-subject connections into the GCN kernels could improve performance. Our results show it is feasible to indicate subcortical activity from cortical fNIRS recordings.