Artificial Intelligence for STN-DBS Surgical Planning in Parkinson’s Disease: A Multicenter Study Comparing Conventional Targeting Versus Supervised Statistical Machine Learning

Objective: Deep Brain Stimulation (DBS) has been consolidated as a valid therapeutic option for advanced Parkinson’s disease (PD). The identification of specific targets can be achieved through different methods, including conventional direct and indirect methods. The aim of our multicentric study is to provide a comparison between these traditional methods and artificial intelligence (AI) in the ascertainment of the ideal targets. Materials and Methods: A total of eight patients, who received bilateral STN (subthalamic nucleus) DBS implantation between 2022 and 2023 were analyzed. Target coordinates were calculated based on the Schaltenbrand and Wahren atlases and the AI using the RebrAIn system during the planning phase; intraoperatively, the targets were either confirmed or adjusted according to microelectrode recordings (MERs). The differences in the three Cartesian axes of stereotactic coordinates (X, Y, and Z) according to these methods were evaluated and compared through non-parametric ANOVA Friedman test. Results: The results revealed significant agreement in the lateral–lateral coordinates (X, X′, X″), indicating stability in target determination along this axis across the methods. However, more substantial discrepancies were observed in the antero-posterior and cranio-caudal coordinates, suggesting lower consistency between the examined methodologies. Conclusions: Our preliminary study results suggest that, despite the challenges posed by interindividual anatomical variability and the limitations of imaging techniques, artificial intelligence has shown comparable values on the lateral–lateral X coordinates. The accuracy of predictive targeting using machine learning models needs to be validated by further studies, but the preliminary results appear to indicate a potential promising role for artificial intelligence in integrating the preoperative workflow.