BackgroundMagnetic resonance imaging (MRI) findings for neural nuclei are an important reference for the diagnosis of Parkinson's disease (PD) and target localization in deep brain stimulation (DBS). The MRI characteristics of the subthalamic nucleus (STN) in PD patients are heterogeneous and may be indicative of differing levels of motor dysfunction in these individuals.ObjectiveTo investigate whether the radiological characteristics of the STN on preoperative T2-MRI can assist in predicting motor function improvement in PD patients following STN-DBS through radiomics.Methods137 patients with good improvement (Good) and 72 patients with poor improvement (Poor) were enrolled. T2-MRI images of the STN were used to extract radiomics features. Three machine learning models were used to classify the patients according to their radiomics features. Finally, the performance and clinical benefits of the models (radiomics model, clinical model, and clinical-radiomics model) were evaluated by calibration curve, receiver operating characteristic (ROC) curve, and decision curve analysis (DCA).ResultsThe logistic regression and support vector machine models optimally distinguished Good and Poor, with areas under the curve (AUCs) of 0.844 and 0.853, respectively. The ROC curve, calibration curves, and DCA demonstrated that the integrated clinical-radiomics model had the highest clinical benefit among all models tested, in the test set (accuracy 0.876 and AUC 0.937).ConclusionsThe combined model incorporating the radiomics features of the STN and clinical features predicted motor function improvement following STN-DBS for PD well and may provide a noninvasive and effective approach for evaluating surgical indications.