Registration Sanity Check for AR-guided Surgical Interventions: Experience From Head and Face Surgery

Abstract

Achieving and maintaining proper image registration accuracy is an open challenge of image-guided surgery. This work explores and assesses the efficacy of a registration sanity check method for augmented reality-guided navigation (AR-RSC), based on the visual inspection of virtual 3D models of landmarks. We analyze the AR-RSC sensitivity and specificity by recruiting 36 subjects to assess the registration accuracy of a set of 114 AR images generated from camera images acquired during an AR-guided orthognathic intervention. Translational or rotational errors of known magnitude up to ±1.5 mm/±15.5°, were artificially added to the image set in order to simulate different registration errors. This study analyses the performance of AR-RSC when varying (1) the virtual models selected for misalignment evaluation (e. g., the model of brackets, incisor teeth, and gingival margins in our experiment), (2) the type (translation/rotation) of registration error, and (3) the level of user experience in using AR technologies. Results show that: 1) the sensitivity and specificity of the AR-RSC depends on the virtual models (globally, a median true positive rate of up to 79.2% was reached with brackets, and a median true negative rate of up to 64.3% with incisor teeth), 2) there are error components that are more difficult to identify visually, 3) the level of user experience does not affect the method. In conclusion, the proposed AR-RSC, tested also in the operating room, could represent an efficient method to monitor and optimize the registration accuracy during the intervention, but special attention should be paid to the selection of the AR data chosen for the visual inspection of the registration accuracy.

Keywords

• Augmented reality
• computer-assisted surgery
• image-to-patient registration
• sanity check