Improved Depth-Based Tracking in Augmented Reality-Assisted Neurosurgery

Neurosurgical navigation is hindered by brain shift and occlusion. We propose an enhanced depth-based tracking algorithm that integrates adaptive bilateral filtering with finite element modeling (FEM) of tissue deformation. A deformable ICP aligns intraoperative point clouds with preoperative MRI surfaces, updated by FEM-predicted displacements. In 10 phantom experiments simulating brain shifts up to 15 mm, average error decreased from 3.2 mm (baseline ICP) to 1.2 mm, while maintaining 25 fps. Compared with conventional methods, alignment stability improved by 38%, supporting safer tumor resections.