Using Algorithm to Search for the Dynamic Images of Strobe-Laryngoscope in Automated Analysis Identification System for Glottis Images

碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 99 === Human glottis opens to the largest extent when individuals breathe with maximal effort and closes to the smallest extent when individuals make voice. Therefore, the images of largest extent and the smallest extent of glottis are captured during video laryngos...

Full description

Bibliographic Details
Main Authors: Wen-lin Chu, 朱玟霖
Other Authors: Chung-feng jeffrey Kuo
Format: Others
Language:zh-TW
Published: 2011
Online Access:http://ndltd.ncl.edu.tw/handle/936389
Description
Summary:碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 99 === Human glottis opens to the largest extent when individuals breathe with maximal effort and closes to the smallest extent when individuals make voice. Therefore, the images of largest extent and the smallest extent of glottis are captured during video laryngoscope as the basis of identification of possible vocal fold lesion when the glottis is being examined. At present, doctors still select the static images of glottis opening to the largest extent and closing to the smallest extent in video clip recorded by strobe-laryngoscope manually, so the automatic recognition of dynamic image aided medical system is an important demand. This study analyzed 220 groups of dynamic images photographed by strobe-laryngoscope and 30 groups of dynamic images photographed with additional laser marking module. The static images of glottis opening to the largest extent and closing to the smallest extent were filtered out automatically by using color space conversion and image preprocessing and the glottal area was quantized. As the tongue base movement affected the position of laryngoscope and saliva would result in unclear images, this study puts forward setting the threshold by using gray level adaptive entropy value, setting up an elimination system to improve the effect of automatically captured images of glottis opening to the largest extent and closing to the smallest extent. The accuracy rate can be 96%. In addition, the glottal area and region segmentation threshold were calculated effectively by using various image preprocessing. The vocal area segmentation was corrected, and the glottis area waveform (GAW) pattern was drawn automatically to assist in vocal fold health. In the aspect of intelligent recognition system for vocal folds disorders, this study aims at four kinds of vocal folds such as normal vocal fold, vocal fold paralysis, vocal fold polyp and vocal fold cyst. By analyzing the characteristic eigenvalues of these four vocal folds patterns and the support vector machine (SVM) classifier to identify vocal folds disorders, the identification accuracy rate can be 98.75%. The results can be provided as assistance for doctors to diagnose the patient's conditions.