Metal Artifact Avoidance: Improved CBCT Image Quality through Tilted C-Arm 3D Scans

Metal artifacts significantly degrade the image quality of cone-beam computed tomography (CBCT), particularly in spine surgeries involving pedicle screws, complicating the assessment of implant positioning and surrounding anatomy. This study introduces a novel Metal Artifact Avoidance (MAA) workflow that leverages deep learning for trajectory optimization to reduce artifacts during CBCT acquisition. The automated approach incorporates real-time user verification, enabling tailored C-arm adjustments based on a predictive artifact model. The MAA workflow begins with scout views to detect metallic objects using a pretrained FasterRCNN model, followed by triangulation of their 3D positions. A physics-based artifact metric predicts the impact of various tilt angles, with the optimal trajectory suggested to the user through an intuitive visualization interface. The method is demonstrated on cadaveric data with pedicle screws in the lumbar spine, comparing standard and MAA-guided tilted scans. Results show that MAA-guided scans visibly reduced artifacts and enhanced visualization of critical anatomical structures, such as cortical surfaces around screws, compared to standard scans. The improvements achieved were consistent, even in cases where post-processing techniques like fsMAR failed to effectively mitigate artifacts. This study demonstrates that combining automated artifact prediction with user-verified trajectory adjustments provides a practical and reliable solution for artifact reduction. Future work will focus on validating the method on larger datasets and optimizing its integration into clinical workflows for broader adoption in spine surgery.