1 Department of Neurology, Philipps-University Marburg, Baldingerstraße, 35043, Marburg, Germany. alexander.calvano@staff.uni-marburg.de.
2 Department of Computer Science and Software Engineering, Concordia University, Montreal, Québec, Canada.
3 Department of Neurology, Philipps-University Marburg, Baldingerstraße, 35043, Marburg, Germany.
4 Donders Institute for Brain, Cognition and Behaviour, Centre of Expertise for Parkinson & Movement Disorders, Radboud University Medical Centre, 6525 EN, Nijmegen, The Netherlands.
5 Centre of Mind, Brain and Behaviour, Philipps-University Marburg, Marburg, Germany.
6 Department of Neurosurgery, Philipps-University Marburg, Marburg, Germany.

Purpose: Accurate segmentation of the subthalamic nucleus (STN) is paramount for optimising outcomes under deep brain stimulation (DBS) in Parkinson's disease (PD). Clinically available tools like Brainlab Elements (BL-E) enable automated segmentations for surgical planning, yet their spatial relationship with postoperative volumes of tissue activated (VTAs) remains insufficiently characterised. Using multi-atlas segmentation (MAS) as an external anatomical reference, we compared the spatial correspondence of STN segmentations derived from BL-E with effective VTAs following monopolar contact review.

Methods: We analysed imaging data from 40 PD patients with chronic STN-DBS. Segmentations were obtained using BL-E based on T1w and T2w scans and MAS derived from a library of 20 manually segmented midbrain nuclei atlases. Spatial correspondence was assessed using Dice Coefficients, Jaccard Indices, and Euclidean centroid distances. Distances between VTA centroids and clinically established settings for STN-DBS were calculated to evaluate targeting consistency. Statistical differences between metrics were assessed using Wilcoxon signed-rank tests.

Results: BL-E segmentations demonstrated superior spatial correspondence with VTAs compared to MAS, with smaller Euclidean distances between centroids (p < 0.001). Dice Coefficients and Jaccard Indices showed no significant differences (p = 0.18). VTA centroid distances to the most efficient stimulation location were consistent across hemispheres (left: 2.54 mm [1.92-3.25]; right: 2.87 mm [1.85-3.82]) MAS targets were positioned more inferiorly and anteriorly compared to BL-E targets.

Conclusion: Clinically applied VTAs showed good spatial correspondence with planning segmentations, suggesting within-workflow reproducibility but not superior correspondence to anatomical ground truth per se. Future studies should incorporate connectomic information to more accurately reflect the functional relevance of stimulation and its therapeutic effects.