Physiotherapy is a key part of treatment for neurological and musculoskeletal disorders, which affect millions in the U.S. each year. Physical therapy treatments typically consist of an initial diagnostic session during which patients’ impairments are assessed and exercises are prescribed to improve the impaired functions. As part of the treatment program, exercises are often assigned to be performed at home daily. Patients return to the clinic weekly or biweekly for check-up visits during which the physical therapist reassesses their condition and makes further treatment decisions, including readjusting the exercise prescriptions. Most physical therapists work in clinics or hospitals. When patients perform their exercises at home, physical therapists cannot supervise them and lack quantitative exercise data reflecting the patients’ exercise compliance and performance.
Without this information, it is difficult for physical therapists to make informed decisions or treatment adjustments. To make informed decisions, physical therapists need to know how often patients exercise, the duration and/or repetitions of each session, exercise metrics such as the average velocities and ranges of motion for each exercise, patients’ symptom levels (e.g. pain or dizziness) before and after exercise, and what mistakes patients make.
In this thesis, I evaluate and work towards a solution to this problem. The growing ubiquity of mobile and wearable technology makes possible the development of “virtual rehabilitation assistants.” Using motion sensors such as accelerometers and gyroscopes that are embedded in a wearable device, the “assistant” can mediate between patients at home and physical therapists in the clinic. Their functions are to:
• use motion sensors to record home exercise metrics for compliance and performance and report these metrics to physical therapists in real-time or periodically;
• offer real-time mistake recognition and feedback to the patients during exercises;
• allow physical therapists and patients to quantify and see progress on a fine-grain level, and
• record symptom levels to further help physical therapists gauge the effectiveness of exercise prescriptions.
One contribution of this thesis is an evaluation of the feasibility of this idea in real home settings. Because there has been little research on wearable virtual assistants in patient homes, there are many unanswered questions regarding their use and usefulness:
Q1. What patient in-home data could wearable virtual assistants gather to support physical therapy treatments?
Q2. How is this wearable in-home technology received by patients?
Q3. Can patient data gathered by virtual assistants be useful to physical therapists? I sought to answer these questions by implementing and deploying a prototype called “SenseCap.” SenseCap is a small mobile device worn on a ball cap that monitors patients’ exercise movements and queries them about their symptoms. The feasibility study showed that this technology would be feasible and acceptable for in-home use by patients, and that it could gather important compliance, performance, and symptom data to assist physical therapists’ decision-making. Another contribution of this thesis is the development of a tool to make virtual assistants usable by therapists. With current technology, virtual assistants of the kind tested in the feasibility research require engineering and programming efforts to design, implement, configure and deploy them. Because most physical therapists do not have access to an engineering team, they and their patients would be unable to benefit from this technology. With the goal of making virtual assistants accessible to any physical therapist, I will explore the following research questions:
Q4. Would a user-friendly rule-specification interface make it easy for physical therapists to specify correct and incorrect exercise movements directly to a computer? What are the limitations of this method of specifying exercise rules?
Q5. How might physical therapists’ specification of exercise rules to patients differ from their specification of the rules to a computer, and how can we use this knowledge to improve the interaction design of the virtual assistant authoring tool?
Q6. Is it possible to create a CAD-type authoring tool, based on the above interface, that physical therapists can use to create their own customized virtual assistant for monitoring and coaching patients?
Q7. What is the recognition accuracy of a virtual rehabilitation assistant created by this tool?
This dissertation research will improve our understanding of the barriers to rehabilitation that occur because of the invisibility of home exercise behavior, will lower these barriers by making it possible for patients to use a widely-available and easily-used wearable device that coaches and monitors them while they perform their exercises, and will improve the ability of physical therapists to create an exercise regime for their patients and to learn what patients have done to perform these exercises. In doing so, treatment should be better suited to each patient and more successful.
Sara Kiesler (Co-chair)
Daniel P. Siewiorek (Co-chair)
Asim Smailagic (Institute for Complex Engineered Systems)
Patrick J. Sparto (Dept. Physical Therapy, University of Pittsburgh)
qkravitz [atsymbol] cs.cmu.edu