Machine learning-based optimization of dual subthalamic nucleus and substantia nigra targeting in deep brain stimulation.

Advances in deep brain stimulation lead technology have created new opportunities for multi-site network modulation, including applications for freezing of gait, but systematic strategies for trajectory planning are lacking. We evaluated trajectories targeting the subthalamic nucleus (STN) and the simultaneous engament of the substantia nigra (SN), specifically its pars reticulata (SNr), which is considered as a potential target in Parkinson's disease. By analyzing 612 electrode trajectories implanted with standard protocols, we found that 61% of trajectories engaged the SNr; simulating larger array spans or deeper implantation depth increased SNr engagement to 76%. We then trained Gaussian Process Classifiers to predict successful SNr engagement. Targeting a point ≥1.5 mm lateral to the medial STN border along Bejjani's line, with an AC-PC angle ≥55° was associated with a ≥ 95% probability of yielding an SNr trajectory. This framework demonstrates that machine learning-assisted data analysis can generate planning principles for precise dual-site stimulation approaches.