本研究之目的在於發展一套針對偏癱患者下肢運動之布氏動作階段(Brunnstrom recovery stage)自動分級系統。本研究藉由自製無線九軸慣性感測器，結合市售無線生理表面肌電圖擷取裝置所架構之量測系統，測量患者之下肢動作訊號；隨後，本研究對所量測之訊號進行分析，萃取出對於布氏動作階段分級有用之特徵，用以訓練本研究所提出基於決策規則之分類系統，分別比較k最佳鄰居法、人工類神經網路、以及支撐向量機等分類演算法之分類能力。交叉驗證則是使用更客觀的leave subjects out法取代過去常用之leave one out法，用以計算分類之成功率。實驗結果顯示使用支撐向量機演算法來分類布氏動作階段，相較於k最佳鄰居法、以及人工類神經網路能得到最高的正確率。本研究所提出之分類系統的穩健性則是透過訓練不同患者人數的資料進行驗證；並且根據其分類之結果，可以歸納出本研究所提出之自動分類演算法，具有準確預測偏癱患者下肢運動之布氏動作階段分類的潛力。;The purpose of this study is to develop an automatic system for classifying lower limb Brunnstrom stages of hemiparetic patients. In this study, the measurement system was employed both IMU and sEMG to acquire the lower limb motion signals from patients. Afterward, this study extracted some useful features for the proposed rule-based classification system and compared different classification algorithms such as k-nearest neighbor, artificial neural network and support vector machine. Instead of leave one out cross validation, the leave subjects out cross validation was used to calculate the successful rate of classification. The result of the experiment shows that SVM has the highest accuracy to classify Brunnstrom stage than k-NN and ANN. The robustness of this classification system was verified by training different number of subject data. According to the classification result, it can be concluded that the proposed classification system has the potential to predict the lower limb Brunnstrom stage for hemiparetic patients.
