Structured-light three-dimensional surface scanning with deep learning-assisted N-bar localizer detection for stereotactic registration: a technical feasibility study in deep brain stimulation

Intraoperative computed tomography (CT) is the clinical standard for stereotactic registration in deep brain stimulation (DBS) surgery, but it adds ionizing radiation, operative time, and workflow complexity. Whether an optical surface-scanning approach could ultimately support CT-independent stereotactic registration remains unknown. In this technical feasibility study, we developed a structured-light three-dimensional surface scanning (3DSS) workflow with deep learning (DL)-assisted N-bar localizer detection and quantified its agreement with conventional CT-based registration.

Ten patients undergoing frame-based DBS for Parkinson’s disease or essential tremor underwent handheld craniofacial and N-bar scanning, completed in 1–3 min without interrupting surgery. CT-derived craniofacial surface models were co-registered with the surface scans using coarse-to-fine point-cloud registration with iterative closest point (ICP) refinement. A supervised DL object-detection model was used specifically to identify N-bar rod cross-sections from voxelized surface scans; and the detected localizer geometry was registered to the known N-bar model using deterministic point-to-line ICP to derive the 3DSS-based stereotactic transformation.

Across the 10 independent patient registrations, CT–3DSS surface alignment reached a mean absolute error of 0.85 ± 0.10 mm, and the resulting 20 hemisphere-level comparisons showed a mean CT–3DSS target-coordinate deviation of 1.91 ± 0.87 mm and a mean trajectory angular deviation of 0.81° ± 0.60°. Despite these average levels of agreement, 10 of 20 target deviations exceeded 2 mm (maximum, 3.48 mm), underscoring the variability of the current approach. In addition, the present implementation remains CT-dependent because a CT-derived craniofacial surface is required to map the 3DSS data into the CT/DICOM world coordinate system, while CT-based stereotactic registration served as the comparison reference.

Taken together, these preliminary single-center findings demonstrate technical feasibility while quantifying the performance gap that must be addressed through hardware and algorithmic refinements before radiation-free, CT-independent stereotactic registration can be considered for clinical use. Digital imaging and communications in medicine Interventional magnetic resonance imaging The authors thank Judi Mills, Kathryn Morsch, and Alex Huwaish for their assistance in the operating room, and are especially grateful to the patients who volunteered for this study. This research was supported by the Mayo Foundation and the Charles B. and Ann L.

Johnson Professorship of Neurosurgery awarded to KHL. Koller and Hojin Shin have contributed equally to this work. Neural Engineering and Precision Surgery Laboratories (NEPS), Mayo Clinic, Rochester, MN, USA Adam L.

Koller, Hojin Shin, Jennifer Tang-Cabrera, Adriano Alfonsi, Elliott Lee, Paul J. Hawkes, Yoonbae Oh, Kendall H. Lee & Jaeyun Sung Division of Computational Biology, Department of Quantitative Health Sciences, Mayo Clinic, Rochester, MN, USA Department of Neurologic Surgery, Mayo Clinic, Rochester, MN, USA Hojin Shin, Jennifer Tang-Cabrera, Paul J.