Prediction of Excitatory and Inhibitory Connections in the Human Brain Based on SDK-GAN

Research shows that the structural connectivity (SC) and functional connectivity (FC) of the human brain are closely related. Exploring how FC is generated by its SC is an important scientific issue in neuroscience. However, the SC obtained by diffusion tensor imaging (DTI) cannot currently describe the excitatory and inhibitory (EI) connections between brain regions, while FC can characterize the positive and negative correlations between brain regions. In addition, studies in neuroscience suggest that the EI connections between brain regions play a crucial role in shaping cortical activity. Therefore, this article aims to develop a model that establishes the EI relationships among brain regions based on SC. Specifically, we propose to combine the single diffusion kernel (SDK) with the generative adversarial network model (SDK-GAN) to predict EI connections, which not only enhances the capture of structural information but also maintains the universality, scalability, and tractability of the deep learning framework. We applied the model to two empirical datasets, the 246 regions of interest (ROIs) and 360 ROIs datasets respectively, and demonstrated its effectiveness in predicting FC as well as the EI connections. We achieved EI prediction accuracies of 0.7072 and 0.7047, with the two datasets respectively, both outperforming existing methods.