Wearable Sensor Measures Arm Impairment During Stroke Recovery

Continuous Wrist Sensing Tracks Arm Impairment After Stroke

A wearable wrist device uses machine learning to track post-stroke arm movement impairment during everyday activities. The technology measures how motor severity changes throughout the rehabilitation process rather than relying on brief visual tests.

Researchers at University of Massachusetts Amherst developed the sensor in partnership with Washington University in St. Louis, Shirley Ryan AbilityLab, and Harvard Medical School/Mass General Brigham. The findings appeared in Science Translational Medicine.

Standard In-Clinic Motor Exams Create Monitoring Gaps

More than 795,000 Americans experience a stroke each year, and upper-limb mobility limitations affect up to 77% of patients immediately afterward. About 40% of those individuals continue to manage chronic arm deficits that impair independent daily living.

Physical therapists typically evaluate arm motor function through manual observation sessions that last roughly 30 minutes. Because these manual evaluations require substantial clinical time, clinicians generally schedule them only before and after a completed course of rehabilitation.

This infrequent schedule leaves patients and physical therapists without objective data on how movement severity responds to specific therapy exercises between sessions.

Machine Learning Separates Movement Volume From Impairment Severity

The wrist sensor houses an accelerometer to record arm motion as individuals carry out standard routines at home. An algorithm interprets the data to separate raw movement volume from underlying motor impairment severity.

Patients can deliberately move an impaired arm more often throughout the day without changing their underlying motor impairment score. The machine learning model accounts for that difference by estimating clinical severity scores directly from continuous motion patterns.

The development team trained the algorithm on motion data and clinical ratings from healthy volunteers and patients in the subacute stage of stroke recovery. The study defined the subacute phase as one week to six months following a stroke.

Trial Results Show Improved Estimation Accuracy and Smaller Sample Needs

The study authors reported that their digital biomarker was 40% to 50% more accurate at reflecting true clinical condition than intermittent visual assessments alone. The sensor gathers data across full days instead of capturing a single performance window inside a clinic.

Continuous monitoring also reduced the cohort sizes needed to evaluate post-stroke interventions in clinical trials. When the authors re-evaluated data from a past stroke trial using the sensor model, they reached statistically significant results with 50% fewer study participants.

Therapists could use continuous impairment data to modify home exercise prescriptions before a multi-week rehabilitation block ends.

Development Status and Clinical Limitations of the Wearable Sensor

The device remains an investigational tool and is not currently distributed for home medical use. Patients recovering from a stroke cannot purchase the monitor over the counter or request it from standard outpatient clinics today.

The algorithm was trained on patients within one week to six months of a stroke. The published data do not demonstrate whether the same model accurately rates arm severity in individuals living with chronic stroke impairments past the six-month window.

Patients undergoing stroke rehabilitation should discuss their current exercise routines and functional progress with their neurologist or physical therapist.