Using Different Classifiers to Validate the Detection of Myofascial Pain via SEMG

• Main Authors: Pao-Tieh Huang, 黃寶蝶
• Other Authors: Ching-Fen Jiang
• Format: Others
• Language: zh-TW
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/80433361402890227136

碩士 === 義守大學 === 生物醫學工程學系 === 102 === In E Generation, the popularity of mobiles causes the growing population of modern diseases. The number of patients with myofascial pain gradually increases. However, the traditional EMG characteristics cannot effectively differentiate the myofascial pain. Our laboratory has previously developed a multi-scale wavelet energy variation (MSWEV) model to detect the myofascial pain, and the accuracy is 69.6%. However, to classify the MSWEV feature through visual inspection is prone to misjudgment because of individual variation in the degree of myofascial pain. Therefore, this study used supervised and unsupervised classifiers to improve the objectivity and reproducibility of classification. This study has applied not only single type classifier, such as Back-propagation Neural Network (BPNN), K-means, and Support Vector Machine (SVM), but also mixed-type classifiers. We combined the BPNN with Principle Component Analysis (PCA) to reduce the number of the features, and to improve the performance of classification by using Adaboost. The results show no significant improvement in the accuracy using these methods. This study finally fine-tunes the false-negative cases in the BPNN classification and false-positive cases in K-means. The results show that the classification accuracy increases to 76.79%. Therefore, the classifier with combination of BPNN and K-means can provide a reliable reference for physicians to diagnose myofascial pain.