Upper Limb Rehabilitation System for Stroke Survivors Based on Multi-Modal Sensors and Machine Learning

Nowadays, rehabilitation training for stroke survivors is mainly completed under the guidance of the physician. There are various treatment ways, however, most of them are affected by various factors such as experience of physician and training intensity. The treatment effect cannot be fed back in time, and objective evaluation data is lacking. In addition, the treatment method is complicated, costly, and highly dependent on physicians. Moreover, stroke survivors' compliance is poor, which leads to various limitations. This paper combines the Internet-of-Things, machine learning, and intelligence system technologies to design a smartphone-based intelligence system to help stroke survivors to improve upper limb rehabilitation. With the built-in multi-modal sensors of the smart phone, training action data of users can be obtained, and then transfer to the server through the Internet. This research presents a DTW-KNN joint algorithm to recognize accuracy of rehabilitation actions and classify to multiple training completion levels. The experimental results show that the DTW-KNN algorithm can evaluate the rehabilitation actions, the accuracy rates of the classification in excellent, good, and normal are 85.7%, 66.7%, and 80% respectively. The intelligence system presented in this paper can help stroke survivors to proceed rehabilitation training independently and remotely, which reduces medical costs and psychological burden.