An Artificial Intelligence Approach for Outcomes Assessment in Laryngeal Reanimation
Nat Adamian, HBSc (C), Matthew Naunheim, MD, MBA and Nate Jowett, MD, Harvard Medical School / Massachusetts Eye and Ear, Boston, MA
Introduction: Vocal fold paralysis secondary to recurrent laryngeal nerve insult is amenable to microsurgical nerve repair or transfer, however there exists no rigorous means for tracking outcomes. Dynamic quantification of the angle formed by the vocal cords at the anterior commissure (the anterior glottic angle) represents a logical biometric for assessment of laryngeal reanimation techniques. Herein, we present a novel computer vision tool for automated tracking of the anterior glottic angle from videolaryngoscopy using a deep learning approach.
Materials & Methods: Videolaryngoscopy data from adult healthy controls (N=20) and patients with unilateral vocal fold palsy (N = 20) was employed for training a novel algorithm using deep neural networks in vocal fold localization and automated anterior glottic angle estimation. The algorithm was packaged into a user friendly software package for rapid dissemination among clinicians and researchers in the field.
Results: Algorithm sensitivity, specificity, positive predictive value, and negative predictive value in vocal fold identification and accuracy of vocal fold localization and anterior glottic angle calculation in comparison to manual expert marking is characterized.
Conclusions: An open-source software package for fully-automated and high-throughput framewise quantification of the anterior glottic angle from clinical videolaryngoscopy is presented. This tool may prove useful for outcomes tracking in laryngeal reanimation.
Back to 2020 ePosters