Figure 1.

Objective quantitative depression diagnosis system

(A) Workflow of the diagnosis system.

(B) Framework of LGMF-GNN. (i) Feature extraction and graph initialization. Features were extracted from the data of the three modalities and further used to initialize the functional graph, structural graph, and demographic graph. (ii) The hierarchical LGMF-GNN structure. Local ROI GNN generates the embedding for each subject based on the ROI graph and ROI BOLD signals of each individual. Global subject GNN aggregates multiple information from the subject graphs of different modalities to obtain the final prediction result. (iii) Detailed network structure of local ROI GNN. (iv) Detailed network structure of global subject GNN.