Inside-Out Accurate Head Tracking with Head-Mounted Augmented Reality Device

Abstract

Objective: External Ventricular Drainage (EVD) is a widely used procedure in neurosurgery that is restricted in accuracy and reproducibility due to free-hand operation. Augmented reality (AR) improves punctuation success rate by superimposing virtual paths on the operation area. However, the effectiveness of surgical guidance is affected by tracking accuracy. This paper aims to achieve accurate and stable head tracking during EVD surgery. Methods: We propose a dynamic inside-out tracking method combining retro-reflective markers and point clouds. First, built-in infrared depth sensor of HoloLens 2 is used to identify markers pasted on patient's head for coarse registration of preoperative images and intraoperative patient. Real-time 3D point clouds and point-to-plane ICP registration are then used to further improve tracking accuracy and stability. Meanwhile, we calibrate and correct the depth distortion of the HoloLens 2 depth sensor on different materials, improving the accuracy of point cloud-based tracking methods. Results: The root mean square error (RMSE) of preoperative registration is less than $1.6mm$; average RMSE of intraoperative head tracking is less than $1.28mm$. Meanwhile, average angular tracking jitter is reduced by more than $40\%$ when integrating point clouds. The proposed method can achieve 37.7fps tracking. Conclusion: The retro-reflective marker and point cloud hybrid tracking method in this paper can achieve high-precision real-time head tracking, providing the potential for accurate visual guidance in EVD surgery.

Haowei Li and Wenqing Yan have contributed equally to this work.