Multimodal Neuroimaging-Based Machine Learning Models Leveraging Cerebral Morphometry and Glymphatic Parameters Predict Short-Term Post-Programming STN-DBS Motor Response in Parkinson's Disease.
Authors
Affiliations (8)
Affiliations (8)
- Department of Neurosurgery of Huashan Hospital, State Key Laboratory of Medical Neurobiology, MOE Frontiers Center for Brain Science and Institutes of Brain Science, Fudan University, Shanghai, China.
- Shanghai Key Laboratory of Brain Function and Restoration and Neural Regeneration, Shanghai, China.
- Shanghai Clinical Medical Center of Neurosurgery, Shanghai, China.
- National Center for Neurological Disorders, Shanghai, China.
- Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China.
- Department of Nursing, Huashan Hospital, Fudan University, Shanghai, China.
- Department of Neurosurgery, Third Division General Hospital, Xinjiang Production and Construction Corps, Tumushuke, Xinjiang, China.
- Tianqiao and Chrissy Chen Institute Clinical Translational Research Center, Shanghai, China.
Abstract
Deep brain stimulation of the subthalamic nucleus (STN-DBS) is effective for medication-refractory Parkinson's disease (PD) motor symptoms, but clinical response varies across symptom domains, particularly tremor and gait. Accurate preoperative stratification is clinically important, especially for early post-programming outcomes. We retrospectively enrolled 155 patients with PD undergoing bilateral STN-DBS and 43 healthy controls. Preoperative structural magnetic resonance imaging and diffusion-weighted imaging were used to quantify brain morphometry and glymphatic markers, including diffusion tensor imaging along the perivascular space (DTI-ALPS) and choroid plexus volume (CPV). Total motor response was evaluated in all 155 patients, tremor response in 133 patients with complete tremor subscores, and an exploratory data-driven gait-improvement phenotype in 66 patients with paired instrumented gait assessments. Machine-learning models were developed using fold-wise feature selection and hyperparameter tuning and were evaluated by fivefold cross-validation. Best-performing trimodal models yielded AUCs of 0.850 ± 0.045 (95% CI, 0.794-0.906) for total motor response, 0.861 ± 0.047 (95% CI, 0.803-0.919) for tremor response, and 0.970 ± 0.019 (95% CI, 0.946-0.994) for the exploratory gait-improvement phenotype. Morphometric-only models retained substantial predictive performance, with maximum AUCs of 0.830, 0.849, and 0.955 for the motor, tremor, and gait-related endpoints, respectively. For the exploratory gait phenotype, clinical-plus-morphometric and trimodal models performed similarly, suggesting limited incremental value of glymphatic variables in this subgroup. Preoperative cerebral morphometry, complemented by selected glymphatic markers and baseline clinical variables, may help stratify short-term post-programming STN-DBS response in PD. These findings support further development of imaging-informed DBS outcome prediction, while external validation and longer-term follow-up remain necessary before clinical implementation.