High-order functional connectivity based Major depressive disorder classification using self-attention based CNN

Major depressive disorder (MDD) is a common and socially significant psychiatric disorder with extremely complex pathologic mechanisms. In recent years, resting-state functional magnetic resonance imaging (rs-fMRI) has been widely used for the classification and diagnosis of MDD. Most current studies reveal the relationship between MDD and brain function by extracting static low-order functional connectivity (SFC) between brain regions, often neglecting potential high-order interactions that may affect classification performance. Therefore, to fully exploit the more complex biomarker features between brain regions and enhance classification performance, this project extracts two types of dynamic high-order functional connectivity (dHOFC): dynamic topographic high-order functional connectivity (dtHOFC) and dynamic associative high-order functional connectivity (daHOFC). It fuses the features of these two types of dynamic high-order functional connectivity and proposes a self-attention-based CNN method for MDD classification validation. The results demonstrate that the proposed deep learning algorithm achieves high classification performance, with the fusion of dtHOFC and daHOFC features providing higher classification accuracy compared to using individual high-order functional connectivity features.