Analysis of Laryngeal High-Speed Videoendoscopy recordings – ROI detection

Abstract

Accurate detection of the glottis in video sequences obtained during the vocal cords examination using Laryngeal High-Speed Videoendoscopy (LHSV) is a prerequisite for the calculation of parameters describing vocal cord kinematics. This work presents the knowledge and methods related to the determination of the region of interest (ROI), which is one of the important steps in the processing of LHSV video sequences with the aim of automatic glottis detection. ROI is defined as the area between the vocal folds and anterior and posterior commissures.A number of methods have been published on this topic, which are used mainly in experimental LHSV video processing systems. To determine the ROI, we decided to use a method based on frequency analysis of oscillations of the vocal cord anatomical structures, which, as we know, has not been used in this context yet. The oscillation is represented by the change of brightness of the corresponding pixels in the LHSV images. The ROI is then successfully detected even in the relatively heterogeneous structure of tissues and fluids and for videos of various qualities, including luminance reflections, where the movement of the vocal cords can be detected.These methods extend the currently used system using a thresholding method for ROI detection and improve the success rate from 69% to 89%. These methods were tested on the LHSV video corpus, which contains 412 video sequences with different recording quality, diagnoses, and age groups of patients, obtained from ENT clinical practice.