DESIGN AND EVALUATION OF A SYSTEM FOR CT-FREE VOLUME RECONSTRUCTION FROM INTRA-OPERATIVE FLUOROSCOPY FOR NAVIGATION IN ORTHOPEDIC SURGERY

Abstract

A novel biomechanical guidance system (BGS) for markerless intra-operative bone tracking that seamlessly integrates into the surgical setting was recently developed [1]. It utilizes 3D models from pre-operative CT, via 3D-to-2D registration methods, to update object poses based on 2D fluoroscopic images. However, on occasion pre-operative CT images may be unavailable. We developed a CT-free method to enable BGS use that leverages neural radiance fields (NeRF) to generate a continuous volumetric scene [2,3] from intra-operative fluoroscopy. For proof of concept, digitally reconstructed radiographs (DRRs) were created from pelvic CT data as a stand-in for intra-operative fluoroscopy. DRRs and virtual C-arm positions were fed into an existing NeRF scene reconstruction system,[2,3] and bones were segmented from the reconstruction. The accuracy of NeRF-derived segmentations was evaluated by comparison to gold-standard CT segmentation. The number of input DRRs was varied to study how this parameter influences reconstructions. Volume reconstructions were readily obtained using the NeRF scene reconstruction system. RMS errors of the NeRF-derived segmentations ranged from 1.23 mm when using 36 input DRR images to 1.98 mm when using only 8 input DRR images. Based on this performance, we conclude that CT-free NeRF volume reconstruction from intra-operative fluoroscopy holds great potential for use in surgical navigation applications involving bony procedures.