Despite emerging research on clinical predictors of cognitive frailty (CF), no studies have yet leveraged radiomic methods to characterize this high-risk geriatric syndrome, creating a critical gap in neurostructural biomarker integration for early detection. This study aims to develop and validate machine learning models for predicting cognitive frailty in the elderly using MRI radiomics and clinical risk factors. This retrospective study analyzed 173 patients undergoing brain MRI at our institution (April 2016–December 2021), categorized into cognitive frailty (CF, n = 83) and cognitively normal (CN, n = 90) groups based on IANA/IAGG criteria. Participants were randomly divided into training (n = 121) and test sets (n = 52; 7: 3 ratio). Independent clinical predictors were identified through univariate analysis and binary logistic regression. Temporal lobe regions of interest (ROIs) were manually segmented on T2-FLAIR sequences, from which 1, 130 radiomics features were extracted. Six machine learning models—logistic regression (LR), support vector machine (SVM), random forest (RF), k-nearest neighbors (KNN), gradient boosting decision tree (GBDT), and extreme gradient boosting (XGBoost) —were developed using two input modalities: (1) optimal radiomics features selected via t-tests and LASSO regression, and (2) combined radiomics-clinical features. Models were trained with 10-fold cross-validation and evaluated using ROC curves, AUC, accuracy, sensitivity, specificity, F1-score, Brier score, DeLong test (AUC comparison), and decision curve analysis (DCA). Model interpretability was quantified via SHAP (Shapley Additive Explanation) values. The machine learning models integrating radiomics and clinical features demonstrated superior predictive performance for cognitive frailty (CF). Among the six algorithms evaluated, the SVM model achieved superior predictive accuracy with an accuracy of 0. 865, specificity of 0. 963, sensitivity of 0. 760, F1 score of 0. 844, AUC-ROC of 0. 914 (95% CI = 0. 818–0. 990) on the testing set. Integrated radiomics-clinical models significantly outperformed radiomics-only models (DeLong test: P < 0. 05; DCA showing superior net benefit). SHAP analysis identified grip strength, age, and four radiomics features (wavelet-HLLglcmImc1, exponentialglcmIdmn, originalₛhapeMajorAxisLength, gradientfirstorderSkewness) as top predictors. Machine learning models integrating MRI radiomics and clinical features effectively predict cognitive frailty, with SVM demonstrating the highest diagnostic accuracy. This interpretable multimodal approach offers a promising decision-support tool for early CF detection in elderly populations. Not applicable.
Hu et al. (Fri,) studied this question.