#317: Ensemble of deep cascades for detection of laryngeal adductor reflex events in Endoscopy Videos

A. S. Hamad, Y. Y. Wang, T. E. Lever, and F. Bunyak

IEEE International Conference on Image Processing (ICIP), pgs. 300-304, 2020

deep learning, laryngeal adductor reflex, automated detection, endoscopy video analysis

PlainText, Bibtex, DOI, Google Scholar


Vocal fold motion plays a critical role in life-sustaining functions of breathing and swallowing. Multiple muscles, nerves, and other anatomical structures are involved in coordinated motion of the vocal folds. Vocal fold function is affected by a wide range of disorders. Laryngeal endoscopy is used in clinical practice to inspect the larynx and to assess vocal fold function. In this paper we propose a deep learning based system for analysis of laryngeal endoscopy videos and for automated detection of laryngeal adductor reflex (LAR) events which are an airway protective mechanism causing brief closure of the vocal folds to prevent aspiration. The proposed system consists of a two-stage solution that combines a vocal fold and glottal region segmentation stage, with a segmentation aided LAR event detection stage. Experimental results show promising LAR detection results robust to many challenges caused by imaging, anatomical, and behavioral variations.