Neural Radiance Fields for 3D Reconstruction in Monoscopic Laryngeal Endoscopy

We explore Neural Radiance Fields (NeRFs) for 3D reconstruction in monoscopic laryngeal endoscopy. Integrating 3D imaging technology is important in otorhinolaryngology where complex anatomical variations often demand precise and personalized surgical interventions. However, the intricate nature of the vocal tract, compounded by the limited field of view, variable illumination conditions, and the dynamic behavior of vocal folds imposes challenging barriers to accurate 3D reconstruction. Typically, monoscopic footage is available limiting 3D reconstruction. We propose the use of NeRFs to address these challenges and improve the accuracy of 3D reconstruction in laryngeal endoscopy. We show that pre-processing is crucial, including the estimation of depth maps and camera poses, as well as the implementation of a NeRF that yields the final 3D reconstruction. Our findings suggest that tailored NeRFs can allow 3D reconstruction in monoscopic laryngeal endoscopy potentially improving patient-specific diagnosis and therapy. To the best of our knowledge, this is the first work leveraging NeRFs for 3D reconstruction from non-stereo laryngeal endoscopy.