Fig. 1

Flowchart of the study. First, a total of 420 features were extracted for each subject, and an intravoxel feature group (280 features), an intervoxel feature group (140 features) and their combination, an intra- and intervoxel feature group, were generated. After standardizing the features, the random forest algorithm and Spearman’s correlation were carried out to reduce the dimensionality of the dataset. Finally, decision tree, random forest, and extreme gradient boosting (XGBoost) were used to discriminate between PD-MCI and PD-NC subjects. SHapley Additive exPlanation (SHAP) analysis was performed to interpret the predictive model