Predicting mild cognitive impairment in patients with Parkinson's disease by integrating striatal MRI radiomics with...

Mild cognitive impairment (MCI), a common and impactful non-motor complication in Parkinson's disease (PD) that often precedes dementia, underscores the urgent need for early predictive tools applicable to routine clinical practice. This study aims to address this issue by investigating whether integrating striatal radiomics features from structural magnetic resonance imaging (MRI) with clinical data can predict MCI in PD patients.

Baseline T1-weighted MRI images and clinical data of 254 PD patients from the Parkinson's Progression Markers Initiative (PPMI) database were retrospectively analyzed. Cognitive function was assessed using the Montreal Cognitive Assessment (MoCA), with PD patients classified as PD-MCI or cognitively normal (PD-CN). A total of 1,316 radiomics features were extracted from the bilateral caudate nucleus (CN) and putamen (PU). After dimension reduction and feature selection, a radiomics model was constructed. Independent clinical risk factors were identified via univariate and multivariate logistic regression, and further integrated with radiomics features to develop a clinical-radiomics combined model for PD-MCI prediction. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, confusion matrix, F1 score, and decision curve analysis (DCA). Correlations between key radiomics features and MoCA scores were also evaluated.

Age and years of education (YOE) were identified as independent clinical risk factors for PD-MCI. The clinical-radiomics combined model outperformed the radiomics-only model in both the training and test sets, with the model incorporating the right PU (PUR) radiomics features achieving the highest AUC: 0.852 (95% CI: 0.787-0.918) in the training set and 0.790 (95% CI: 0.657-0.923) in the test set. The corresponding F1 scores were 0.704 and 0.667, respectively. Additionally, specific radiomics features showed weak but significant correlations with MoCA scores (P < 0.05).

Integration of striatal radiomics features derived from structural MRI images with routine clinical factors demonstrates promising predictive performance for PD-MCI. The proposed clinical-radiomics combined model leverages clinically accessible resources, and its predictive value for PD-MCI establishes a preliminary foundation for subsequent related explorations. However, the model's generalizability remains unconfirmed, further validation on independent datasets is required before any consideration of its clinical application.