Abstract
BACKGROUND: White matter hyperintensities (WMHs) are closely associated with cognitive frailty (CF). This study aims to explore the potential diagnostic value of WMHs for CF based on radiomics approaches, thereby providing a novel methodology for the early diagnosis and timely intervention of CF. METHODS: The present study conducted a retrospective analysis on 147 patients (77 with CF, 70 in the control group). Following an 8:2 ratio, the patients were randomly divided into training and testing sets. Repeated 5-fold cross-validation was adopted for model training and evaluation. Optimal radiomic features were extracted and selected from T2-FLAIR images, and multiple logistic regression analysis was utilized to identify independent risk factors. Three machine learning algorithms-K-Nearest Neighbors (KNN), Logistic Regression (LR), and Support Vector Machine (SVM)-were used to construct radiomic models, clinical models, and combined models. The performance of each model in diagnosing CF was evaluated using metrics including the area under the curve (AUC), area under the net benefit curve (AUNBC), and Brier score. RESULTS: In the test set, the AUC values of KNN, LR, and SVM in the radiomics models were 0.860, 0.916, and 0.885, respectively; the AUC values of the clinical models were 0.868, 0.850, and 0.787, respectively; and the AUC values of the combined models were 0.906, 0.954, and 0.930, respectively. The decision curve analysis (DCA) demonstrated that the combined model was superior to the single models in terms of clinical decision-making efficacy. CONCLUSION: The radiomic model, clinical model, and combined model can effectively diagnose CF patients, with the combined model demonstrating the best diagnostic efficacy. CLINICAL TRIAL NUMBER: Not applicable.