PMC

3D feature representations for PD and normal classes according to feature selection cases: (a) the feature representation for PD class and no feature selection; (b) the feature representation for Normal class and no feature selection; (c) the feature representation for PD class and CLS feature selection; (d) the feature representation for normal class and CLS feature selection.

(A)Accuracy of the feature selection process. (B)The number of feature subsets of the feature selection process.

Illustrates model performance metrics with single- and two-tier feature selection, including feature selection using only XGBoost and Boruta, along with receiver operating characteristic curves (ROC), precision-recall curves (PRC) predicted by models with two-tier feature selection. These evaluations are conducted on both the training set (MIMIC-III) and the validation sets (MIMIC-IV, eICU-CRD), with the receiver operating characteristic curves presented alongside a shaded 95% confidence interval (ROC (95% CI)). (A–C) denote the ROC, PRC, and ROC (95% CI) predicted by the model using only XGBoost for feature selection, respectively. (A′–C′) denote the ROC, PRC, and ROC (95% CI) predicted by the model using only Boruta for feature selection. (E–G) denote the ROC, PRC, and ROC (95% CI) predicted by the model using two-tier feature selection on the training set. (E′–G′) denotes the AUROC, PRC, and ROC (95% CI) predicted by the model using two-tier feature selection on Validation 1 (MIMIC-IV). (E″–G″) denotes AUROC, PRC, and ROC (95% CI) predicted by the model using two-tier feature selection on Validation2 (eICU-CRD). CI confidence interval.

Overview of feature selection strategy, including four main steps: ① rs-fMRI acquisition, ② D-FCN construction, ③ feature extraction, and ④ feature selection. (M) Feature extraction. What’s more, we display three feature selection strategies: (A) single-view based feature selection (SVFS) strategy, (B) concatenation based feature selection (CFS) strategy; and (C) unit-based personalized fingerprint feature selection (UPFFS) strategy.