Graduate Institute of Automation and Control, National Taiwan University of Science and Technology, Taipei 106, Taiwan.
Comput Methods Programs Biomed. 2013 Oct;112(1):228-36. doi: 10.1016/j.cmpb.2013.06.021. Epub 2013 Jul 31.
This study used the actual laryngeal video stroboscope videos taken by physicians in clinical practice as the samples for experimental analysis. The samples were dynamic vocal fold videos. Image processing technology was used to automatically capture the image of the largest glottal area from the video to obtain the physiological data of the vocal folds. In this study, an automatic vocal fold disease identification system was designed, which can obtain the physiological parameters for normal vocal folds, vocal paralysis and vocal nodules from image processing according to the pathological features. The decision tree algorithm was used as the classifier of the vocal fold diseases. The identification rate was 92.6%, and the identification rate with an image recognition improvement processing procedure after classification can be improved to 98.7%. Hence, the proposed system has value in clinical practices.
本研究使用医师在临床实践中实际拍摄的频闪喉镜视频作为实验分析的样本。这些样本是动态声带视频。本研究使用图像处理技术自动从视频中捕捉最大声门区的图像，以获取声带的生理数据。本研究设计了一种自动声带疾病识别系统，可以根据病变特征从图像处理中获得正常声带、声带麻痹和声带小结的生理参数。决策树算法作为声带疾病的分类器，识别率为 92.6%，经过分类后的图像识别改进处理程序，识别率可提高至 98.7%。因此，该系统在临床实践中具有应用价值。
