| Summary: | 碩士 === 國立臺灣科技大學 === 材料科學與工程系 === 106 === The symptoms of the larynx are observed by endoscopes according to different standards. The results of the examination depend on the subjective judgment of the examiner's own experience, making the data and results of different studies almost incompatible. The purpose of this study was to develop a computer-aided diagnosis system for detectable laryngeal lesions based on objective criteria, including vocal cord polyps, vocal cord cysts, vocal cord leukoplakia, vocal cord tumors, and reflux pharyngitis, and to grade the severity of reflux pharyngitis. .
In this study, the changes in the throat caused by laryngeal lesions were quantified, and the automatic segmentation of the throat features in the endoscopic film was performed. Laryngeal lesions can be diagnosed by image processing techniques using hue, texture, and geometric analysis of areas including sacral cartilage, glottis, left vocal cords, and right vocal cords. This study also tested 36 features of hue and texture, using Fisher's linear discriminant screening to characterize the classification performance of reflux pharyngitis.
This study developed a total of 459 samples of laryngeal lesion detection system, including reflux pharyngitis, vocal cord polyps, cysts, leukoplakia and tumors. The classification accuracy of vocal cord lesions was 97.45%. The support vector machine (SVM) was applied to the detection of reflux pharyngitis. The results were evaluated by accuracy, sensitivity and false positive rate. The evaluation results were 97.16% accuracy, 98.11% sensitivity and 3.77% false positive rate. In this study, artificial neural network (ANN) was used as the classification severity of reflux pharyngitis, with an accuracy of 96.48%.
|
|---|
