Abstract
Augmented Reality (AR) has emerged as a promising visualization tool in computer-assisted surgery, but clinical adoption is limited by projection inaccuracies caused by parallax-effect, particularly at short goggle-to-target distances. The goal is to develop a method that dynamically corrects the perceived AR projection errors through a novel user-specific calibration technique. First, a computationally decentralized architecture was designed to offload spatial computations from the headset to external hardware while enabling goggle-navigation registration. Second, the Active Parallax-effect Compensation(APC) framework was developed to account for parallax-induced distortions in real-time. The APC framework was validated using both a standard ASTM F2554-22 model and a patient-specific spine model. Projection accuracy was evaluated under APC ON/OFF conditions, with errors measured at predefined target points. With APC OFF, projection errors measured 8.50 $\pm$± 2.98 mm (close) and 6.26 $\pm$± 2.54 mm (far). With APC ON, errors were significantly reduced to 2.10 $\pm$± 1.35 mm (close) and 1.72 $\pm$± 1.27 mm (far), representing reductions of 75.2% and 72.5%, respectively. Statistical equivalency testing confirmed margins of 0.58 mm for the ASTM model and 0.68 mm for the spine model (p $< $< 0.05). The proposed APC framework substantially reduces AR projection error by compensating for the parallax-induced perceptual distortions, thereby improving projection accuracy for AR-assisted surgery. This study demonstrates a scalable approach for enhancing AR projection reliability in surgical navigation, supporting more accurate and clinically viable optical see-through AR guidance systems.