Accurate detection of spatially calibrating laser points during transnasal, fiberoptic, laryngeal high-speed videoendoscopy using support vector machine

Performing calibrated spatial measurements of vocal fold structures (including pathology) and kinematics during phonation is essential for developing patient-specific voice production models and better quantifying treatment outcomes. Unfortunately, conventional videoendoscopic images cannot provide...

Full description

Saved in:
Bibliographic Details
Published in:The Journal of the Acoustical Society of America 2019-10, Vol.146 (4), p.3083-3083
Main Authors: Ghasemzadeh, Hamzeh, Deliyski, Dimitar, Hillman, Robert E., Mehta, Daryush D.
Format: Article
Language:English
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Performing calibrated spatial measurements of vocal fold structures (including pathology) and kinematics during phonation is essential for developing patient-specific voice production models and better quantifying treatment outcomes. Unfortunately, conventional videoendoscopic images cannot provide such measurements. However, adding laser-based fiducial markers to the field of view (FOV) can provide measurements in the coronal and transverse planes. Accurate detection of the laser points during in vivo recordings is a pre-requisite of such a measurement. This study presents a novel method based on machine learning for detection of the laser points. A custom-developed transnasal, fiberoptic endoscope projected a pattern of 7 × 7 green laser points on the FOV during sustained vowel tokens, which was recorded with a color high-speed video camera at 6000 frames per second. Analysis of the images showed that each laser point travels along a unique and deterministic trajectory. Based on this, the classification score of a support vector machine was employed for maximizing the posterior probability and finding the most probable laser point in each trajectory. Classification features were tailored to account for geometrical shape and gradient profile of laser points. Additionally, features were extracted from red, green, and blue color channels for accurate distinction between laser points and reflection light.
ISSN:0001-4966
1520-8524
DOI:10.1121/1.5137707