A 2020 report shows that “There are over 13.7 million new strokes worldwide each year." Many stroke patients perform individually tailored exercises based on their own needs and symptoms. Activity monitors can help the patients to self-monitor and quantify their activity. The step counters mobile apps for healthy people are not applicable for stroke patients because of high variation in their movement patterns during walking, especially at a slow velocity. Furthermore, the variability of techniques and approaches individual stroke patients employ during rehabilitation is high. Thus, it is critical that the rehabilitation apps to be used by the stroke patients can address the high variability of movement patterns and velocities. We have developed a mobile app to analyse data collected from sensors placed on the lower limb to classify and quantify activities (e.g., counting steps or exercises) and to identify if movements were performed correctly. The key difference of our mobile app from the commercial step counters in the market is that our mobile app uses machine learning algorithms to learn and classify movement patterns of each individual stroke patient and can, therefore, address the variation in movement patterns in stroke patients. The core machine learning algorithms have been developed and have been tested by healthy people to learn, classify, and summarize different kinds of movements. In this project, we will pilot the mobile app with stroke patients to evaluate the accuracy of the mobile app to classify activities and exercises of stroke patients. We will also interview the stroke patients and therapists to assess the potential clinical benefit of the system and collect user experience. Based on the evaluation results, we will update the mobile app to make it more suitable to be used by the end-users, i.e., stroke patients and therapists, and publish the evaluation results of the mobile app.
Project leader: Jingyue Li
Institution: NORGES TEKNISK-NATURVITENSKAPELIGE UNIVERSITET NTNU