Multimodal Image Guidance in Subthalamic Deep Brain Stimulation for Parkinson's Disease.
The authors developed and validated a multimodal, imaging-informed model that integrates contact coordinates, electric fields, tract activations, and network data to predict motor improvement and reliably identify optimal or adjacent STN-DBS contacts across multiple cohorts.
What the AI sees
The authors developed and validated a multimodal, imaging-informed model that integrates contact coordinates, electric fields, tract activations, and network data to predict motor improvement and reliably identify optimal or adjacent STN-DBS contacts across multiple cohorts.
Research significance
While not a molecular therapeutic, this clinically validated, multimodal imaging approach has clear translational value for personalizing and streamlining DBS programming, improving patient outcomes and efficiency in Parkinson's neuromodulation care and trials.
Source abstract
OBJECTIVE: Accurate electrode placement and individual stimulation parameters influence the outcomes of subthalamic deep brain stimulation in Parkinson's disease. Neuroimaging-based models can help evaluate how electrode placement impacts improvement, aiming to reduce the burden of programming. However, most existing models have been developed to explain differences between patients rather than differences between contacts within the same patient, leaving the clinical relevance of image-guided programming unclear. METHODS: We analyzed data from patients with Parkinson's disease treated with subthalamic deep brain stimulation to develop and validate a neuroimaging-informed model of motor improvement measured by the Unified Parkinson's Disease Scale. Five approaches were tested: active contact coordinates, electric fields, tract activations, as well as structural and functional networks. All approaches were integrated into a combined ridge regression model and validated using 2 hold-out datasets. RESULTS: The sample included 236 patients (604 stimulation sites), divided into a training cohort (N = 129), a retrospective validation cohort (N = 89), and a prospectively acquired validation cohort (N = 21 electrodes). Consistent with expectations, our model explained approximately 12% of the variance in unseen group-level data (R2 = 0.12, p = 0.001). At the individual level, the model identified the optimal clinical contact or its neighboring contact in all but one case (mixed-effects R2 = 0.31, p = 3.67 × 10-10). INTERPRETATION: An imaging-informed model explained the expected variance at the group level and demonstrated potential for guiding stimulation programming, suggesting that image-guided approaches may improve clinical decision making while reducing the need for lengthy postoperative testing. ANN NEUROL 2026.