A new wearable wrist device developed by University of Massachusetts Amherst researchers uses a machine-learning algorithm to continuously track arm movement impairment in stroke patients. The technology allows clinicians to measure motor severity outside the clinic and make real-time adjustments to therapy programs, shifting rehabilitation away from the current one-size-fits-all standard.
Continuous Monitoring Replaces Traditional Assessments
Every year, upwards of 795,000 people in the United States suffer strokes, and as many as 77% of these patients encounter upper-limb mobility problems right after the occurrence. About 40% of patients continue to have chronic issues that limit independent living. While physical therapy helps, progress is typically measured by a 30-minute observational assessment conducted by a clinician pre- and post-rehabilitation.
“That means during that therapy process, neither the patient nor the therapist has a clear idea of how patients are responding to the treatments that they’re receiving,” says Sunghoon Ivan Lee, associate professor in the Manning College of Information and Computer Sciences at UMass Amherst and corresponding author on the paper describing the technology. Ongoing data collection through a wearable monitor lets therapists adjust treatment strategies in a timely manner.
Algorithm Outperforms Standard Clinical Evaluations
The device uses an accelerometer sensor to capture upper-limb movement. Ryan Wang, a graduate student and lead author on the paper, developed the machine-learning algorithm that interprets the data. The model was trained on accelerometer data and clinician assessment scores from subacute stroke patients, who are between one week and six months post-stroke, as well as healthy individuals.
Researchers found that the algorithm was 40% to 50% more accurate in reflecting a patient’s true condition compared to standard clinician evaluations. The team discovered that using their digital biomarker instead of clinician observations generated statistically significant results with 50% fewer participants, which can accelerate research and reduce costs.
Commercialization and Clinical Trials
The National Institutes of Health supported the work, which was completed alongside researchers from Washington University in St. Louis, Shirley Ryan AbilityLab, and Harvard Medical School/Mass General Brigham. A provisional patent has been filed, and Lee is pursuing commercialization through the startup Lumid Health. Support for this phase comes from the UMass Amherst Institute for Applied Life Sciences’ Translational Seed Award, the UMass Office of Technology Commercialization & Ventures Technology Development Fund, and participation in the NSF I-Corps Training Program.
Research partners are currently recruiting stroke patients for an ongoing study at Spaulding Rehabilitation Hospital in Boston to further develop the technology.
How sensors and machine learning track stroke recovery continuously
How does the wearable device track stroke recovery?
An accelerometer sensor captures upper-limb movement throughout the day, and a machine-learning algorithm interprets the data to track changes in arm movement impairment.
Why is continuous tracking better than traditional clinic assessments?
Traditional assessments take about 30 minutes and typically occur only before and after rehabilitation, leaving a gap where neither patients nor therapists know if the treatment is working. Continuous monitoring captures real-life performance at all times of day.
Who developed the machine-learning algorithm?
The algorithm was developed by Sunghoon Ivan Lee and his graduate student, Ryan Wang, at the University of Massachusetts Amherst.
Where is the technology currently being tested?
Research partners are recruiting stroke patients for a study at Spaulding Rehabilitation Hospital in Boston.
Keep reading