Graph Neural Networks for Analysis of rs-fMRI Differences in Open vs Closed Conditions

Functional Magnetic Resonance Imaging (fMRI) is a noninvasive neuroimaging technique widely used for research purposes. Appliation of fMRI for medical purposes is still very limited inspite of considerable potential for offering valuable prognostic and differential diagnostic information. One of the problems limiting the use of fMRI in medical settings is that fMRI data is represented as a four-dimensional array of information, and diagnostics relies on the methods employed for data processing only while visual analysis of raw data is impossible. Thus further development of the use of fMRI in clinical practice directly depends on the effectiveness and reliability of the data processing methods used. Resting-state is the main way of scanning in clinical neuroimaging. Resting-state fMRI (RS-fMRI) data can be collected under three conditions: eyes closed (EC), eyes open (EO), and eyes fixated on a target (EO-F), each presenting distinct neuronal activity patterns. It is widely acknowledged that significant differences exist between these three states, making the classification of eye open/closed states a robust basis for verifying models that can be used for diagnostic purposes. We have studied the performance of graph neural networks (GNNs) in identifying dissimilarities between eyes closed and fixated conditions. Additionally, we employ interpretation algorithms to gain insights into the crucial edges influencing the GNN model’s classification. Our proposed GNN model achieves an accuracy of up to 81% in distinguishing between these conditions, with notable brain regions, including visual networks, the default mode network, and the frontoparietal cognitive control network, playing a vital role in accurate classification, consistent with findings from existing literature. Our research highlights the potential of GNNs as a promising approach for exploring functional connectivity differences in RS-fMRI data.