Application Of Upper Limb Rehabilitation Movement Evaluation For Stroke Patients Based On Inertial Sensor Data And Machine Learnin

Nowadays,with the rise of computer-assisted technologies such as the Internet and AI,digital health-related research is rapidly increasing,and more and more new technologies are widespread to apply in the area of rehabilitation.At the moment,there are a large number of stroke patients in our country and the resources for stroke rehabilitation are far from meeting the long-term rehabilitation needs of patients.Furthermore,it is difficult for them to regularly return to the hospital for examinations affected by the new crown pneumonia epidemic,and the high treatment costs continue to burden the patient’s family.Stroke patients need long-term systematic upper limb rehabilitation training to restore upper limb motor function after discharge.However,due to psychological barriers and lack of subjective enthusiasm,only a few patients can maintain a high rehabilitation enthusiasm after discharge and follow the rehabilitation physician’s instructions then complete the rehabilitation training program on time.Meantime,rehabilitation physicians cannot track and grasp the completion of daily rehabilitation and related rehabilitation training data of patients in a timely manner,and it is difficult to adjust the rehabilitation plan of each patient in a targeted and timely manner to improve the efficiency of rehabilitation.In response to the above problems,this paper has mainly done the following two works:(1)Based on technologies such as the Internet,machine learning and software development,this paper devises upper limb functional rehabilitation system based on mobile devices.Patients carry out upper limb rehabilitation training on the affected upper limb by holding or binding a smartphone.During the process,the rehabilitation system become aware of the real-time acquisition of the inertial data of the upper limb rehabilitation training action through various sensors built in the smartphone,and then sends the upper limb rehabilitation training action data to the remote server through the Internet,and uses the dynamic time warping-K nearest neighbor(DTW-KNN)to calculate and analyze the multi-dimensional inertial sensor data,so as to realize the classification and evaluation of the completion of the upper limb rehabilitation action.The experimental results show that DTW-KNN can better classify and judge the elbow flexion and upper limb rehabilitation movements with different completion conditions.This upper limb function rehabilitation system is not limited by time and space,and patients can perform effective upper limb rehabilitation training independently at home,rehabilitation physicians can adjust the rehabilitation plan in time.To a large extent,it makes up for the shortcomings of traditional upper limb rehabilitation methods,reduces the medical cost of stroke and upper limb rehabilitation,and effectively alleviates the shortage of rehabilitation physicians.(2)The inertial sensing data of a series of upper limb rehabilitation actions collected by the designed upper limb functional rehabilitation system in this paper have obvious time series.Therefore,this paper adopts two recurrent neural network(RNN):long short-term memory neural network!LSTM "and gated recurrent unit neural network!GRU",and constructs the classification model of recurrent neural network based on the characteristics of multidimensional inertial sensing data collected by the upper limb functional rehabilitation system,so as to realize the classification and evaluation of various upper limb rehabilitation actions under different completion conditions at last.The experimental results show that the overall classification accuracy of DTW-KNN for elbow flexion,elbow flexion & forearm abduction and shoulder flexion is 72.2%,47.2% and 69.4%,respectively,and the classification accuracy corresponding to LSTM is accurate.The accuracy rates are 97.2%,94.4% and 91.7%,respectively,and the corresponding accuracy rates of GRU are 97.2%,83.3% and94.4%.Under the above three upper limb rehabilitation actions,the classification accuracy of the two types of neural network models is higher than that of DTW-KNN.25%,47.2%,22.3% and 25%,36.1% and 25%.In addition,the neural network model takes much less time for classification than DTW-KNN,and GRU is better than LSTM in classification time with the same high classification accuracy as LSTM.