Structure From Motion Reconstruction of the Pediatric Larynx: A Clinical Case Series.

Abstract

<h4>Objective</h4>Endoscopy is the gold-standard diagnostic modality for many pediatric airway disorders but does not enable quantitative assessment of airway morphology. In a series of pediatric patients, we aim to demonstrate that structure from motion (SfM) photogrammetry, a computer-vision algorithm, can accurately reconstruct three-dimensional (3D) models of laryngeal anatomy directly from bronchoscopy videos.<h4>Study design</h4>Prospective case series.<h4>Setting</h4>Tertiary academic pediatric hospital.<h4>Methods</h4>Pediatric patients undergoing direct laryngoscopy/bronchoscopy and same-day neck and chest computed tomography (CT) as a part of clinical care were recruited. Optical calibration of the imaging system occurred after each bronchoscopy. SfM was used to reconstruct 3D surfaces from bronchoscopy videos. SfM-reconstructed surfaces were compared against CT-derived geometries to determine reconstruction accuracy at glottic and subglottic levels.<h4>Results</h4>SfM reconstruction was successfully completed for all four patients. Circular equivalent diameter (D_CE) of cross-sectional area slices measured from SfM-reconstructed airways was highly accurate compared to D_CE measured from the gold-standard CT scans, with root mean squared error of 0.1, 0.375, and 0.225 mm at the glottis, proximal subglottis, and distal subglottis subsites, respectively. This represents average percent errors of 2%, 4%, and 4%, for each site, respectively.<h4>Conclusion</h4>We demonstrate the feasibility of reconstructing laryngeal anatomy with high fidelity from SfM photogrammetry in a clinical series of pediatric patients. This reproducible method provides highly accurate anatomic models directly from endoscopy, enabling quantitative assessment without CT scanning and radiation exposure. This approach has the potential to impact many areas of otolaryngology, including virtual surgical planning, simulation, and family counseling.