Interpretable machine learning for cognitive impairment assessment: integration of clinical and radiomic white matter hyperintensities features.

backgroundWhite matter hyperintensities (WMH) are recognized as important imaging biomarkers associated with cognitive impairment. This study aimed to develop a machine learning model that combines clinical data and WMH radiomic features for improving the assessment of cognitive impairment

methodsA retrospective study was conducted on 303 patients with WMH. Clinical data and magnetic resonance imaging scans were collected, and cognitive function was evaluated using the Montreal Cognitive Assessment (MoCA). WMH lesions were segmented on T2 Fluid-Attenuated Inversion Recovery images using the SAM2UNET model, followed by the extraction of radiomic features from the segmented WMH regions. After feature selection with LASSO and recursive feature elimination (RFE), six machine learning models were developed, and the optimal model was identified. SHapley Additive exPlanations (SHAP) were applied to enhance the interpretability of the model. Model performance was evaluated using metrics such as the area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, and F1-score.

resultsThe integrated TabPFN model, utilizing 10 clinical and 5 radiomic features, achieved the superior overall predictive performance. The model yielded an AUROC of 0.842, an F1-score of 0.737, an accuracy of 0.754, a recall of 0.750, a precision of 0.724, and a specificity of 0.758, respectively. Calibration and decision curve analyses indicated good agreement and favorable clinical utility of the model in assessing cognitive impairment.

conclusionThis study established a reliable and interpretable TabPFN model integrating routine clinical and radiological data, offering a promising tool for the early detection and personalized management of cognitive impairment in community populations.