Figure 1.

Flowcharts of the proposed disease diagnosis algorithm by using Riemann Kernel PCA for feature extraction. Part (a) is the main flowchart. Connectivity matrices were first normalized as their graph Laplacians. Using proposed Riemann Kernel, we could generate the kernel matrix which describing the geodesic distances among all subjects. By selecting the eigenvalues of kernel matrix, PCA was performed on FC matrices. The extracted features were sent to XGBoost for disease diagnosis. Part (b) presents a schematic representation of Riemann kernel PCA. Using kernel trick, nonlinearly distributed data points could be projected onto its principal components. Part (c) illustrates that we could retrieve decisive features from XGBoost classifier by using pre‐image based kernel PCA reconstruction algorithm