14 - Precision of an Augmented Reality and Artificial Intelligence Navigation System for Pedicle Screw Insertion

Introduction: Spinal navigation is intended to address some of the challenges associated with posterior spinal instrumentation, including minimizing pedicle screw misplacement, decreasing radiation exposure, and improving procedural efficiency. Computer-based surgical navigation systems are increasingly used to aid pedicle screw placement in open and percutaneous spinal fixation. These systems involve intraoperative 3D computer tomography (CT) and/or 2D C-arm fluoroscopy scanners to capture anatomical structures and infrared cameras to track surgical instruments with respect to the scanned and registered patient spine. Augmented reality (AR) and artificial intelligence (AI)-enabled software with convolutional neural networks to segment and register patient anatomy could provide additional benefits.

Aims and

Objectives: The purpose of this cadaveric study was to assess the 3D positional and angular errors of navigated lumbosacral pedicle screw placements using a novel spine guidance system. This navigation system combines AR with AI for autonomously segmenting and displaying all spinal components (e.g., pedicles, vertebral bodies, lamina, and spinous, transverse, and articular processes) directly over the surgical field. The system also autonomously registers the patient anatomy for determining the spatial relationship between the spine and surgical instruments. The results of this study were compared to published pedicle screw placement errors of commercially-available navigation systems for spine fixation.

Methods: Eighty lumbosacral pedicle screws were placed by four spine surgeons in eight cadavers using the AR/AI navigation systems. Both pre- and post-operative CT scans of the cadaveric spines were obtained before and after the insertion of the pedicle screws and then co-registered to perform the accuracy analysis in the same coordinate system. The performance validation procedure consisted of three steps: (1) automatic image-based registration and 3D coordinate transformation, where the anatomy in the pre-operative scan volume was aligned to the anatomy in the post-operative scan volume to determine the virtual screw 3D positions and orientations in the post-operative scan coordinate system; (2) manual analysis of the post-operative scan to determine the 3D positions and orientations of the real implants as they appeared in the CT scan volume; and (3) automatic comparison between the 3D positions and orientations of the virtual and real screws. These errors were computed by comparing the 3D Euclidean distances and 3D angles of the virtual screws with the corresponding real screws.

Results: The mean positional error was 2.31 mm with a standard deviation of 1.01 mm and a 99% confidence interval (CI) between 2.02 and 2.60 mm. The mean angular error was 1.45 degrees with a standard deviation of 0.72 degrees and a 99% CI between 1.24 and 1.66 degrees. Fig. 1 shows comparison results of the positional and angular errors of the novel AR/AI system with respect to FDA-cleared navigation systems. Publicly reported upper bound limits of the 99% confidence interval for each navigation system were considered for this comparison.

Conclusions: This study shows that the novel AR/AI surgical navigation system provides a mean 3D positional error lower than 3.0 mm and a mean 3D angular error lower than 3.0 degrees, demonstrating that it is equivalent to other navigation systems on the market.