PURPOSE: Accurate image registration is essential for aligning preoperative and intraoperative scans in image-guided neuro interventions, where even small misalignments can compromise navigation. This work introduces a model-based framework for rigid and nonrigid volumetric registration using shape-constrained deformable brain segmentation to establish anatomical point-based correspondence. METHODS: Unlike traditional methods, the registration relies solely on anatomical geometry. Rigid registration aligns centroids and estimates transformations between segmented meshes. Nonrigid registration fits B-spline surfaces to corresponding mesh vertices, generating smooth deformation fields that capture local anatomical changes. RESULTS: Quantitative validation was performed using synthetically transformed MR scans. Additional structure-specific analysis confirmed accuracy across multiple brain regions. The rigid registration performed noninferiorly to an FDA-cleared approach, while the nonrigid registration effectively captured realistic deformations such as those resulting from brain shift. Both methods offer high accuracy and computational efficiency. CONCLUSION: The proposed model-based registration framework offers a robust, anatomically driven alternative to conventional image-based registration methods that eliminates reliance on image intensity. It results in improved registration accuracy and demonstrates strong clinical potential for integration into interventional workflows, with the promise of enhancing procedural precision, safety, and patient outcomes.
Zagorchev et al. (Tue,) studied this question.