Reconstruction of Bony Anatomy from Sparse Fluoroscopy Sampling Using Neural Radiance Fields

Computed tomography (CT) imaging is a cornerstone of modern orthopedic practice due to unmatched clarity in its volumetric reconstruction of internal bony anatomy. However, CT imaging carries the risk of elevated radiation exposure, is costly to acquire, and in certain situations is simply unavailable. While in the operating room, for instance, at a time when the information it can provide is arguably most needed, surgeons do not typically have access to CT. Instead, they must rely on 2D fluoroscopic imaging to visualize complex internal anatomy. Here, we present a method to accurately reconstruct bony anatomy from sparse fluoroscopy sampling (as few as 7 input views) in under 30 s by leveraging a recent advancement in computer vision/machine learning known as neural radiance fields, or NeRFs. Using two different surgical applications, we then demonstrate how the method performs given a variety of input fluoroscopy views, both in terms of the time it takes to reconstruct bony anatomy and the accuracy of the reconstruction. Based on these initial findings, we conclude that this method shows promise for use in the operating room as an adjunct to conventional intra-operative imaging capabilities.