Lower-limb motion classification for hemiparetic patients through IMU and EMG signal processing

This study is to develop an automatic classification system for lower limb Brunnstrom stage of hemiparetic patients. The measurement system is 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 results of the experiment on seventeen men and seven women at mean age: 60.6 ± 12.4 years after stroke for more than 6 months show that SVM (95.2%) has the highest accuracy to classify Brunnstrom stage than k-NN (89.2%) and ANN (92.3%). 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.