Using Wearable Inertial Sensors to Estimate Clinical Scores of Upper Limb Movement Quality in Stroke

This study explores the use of wearable sensors to assess upper limb motor function in stroke patients, specifically focusing on the Action Research Arm Test (ARAT). The aim is to enable remote and objective assessments. The study involved stroke patients performing ARAT tasks while wearing wrist-worn inertial sensors. Machine learning algorithms were used to estimate ARAT scores from the sensor data. The results showed that ARAT scores could be estimated with reasonable accuracy using data from just two wrist worn sensors, suggesting potential for widespread use of wearable sensors in neurorehabilitation.

• 1
  ARAT task scores were classified with approximately 80% accuracy using machine learning algorithms applied to data from two wrist-worn inertial sensors.
• 2
  Linear regression between summed clinical task scores and estimated sum task scores yielded a good fit (R2 = 0.93).
• 3
  Estimates of the sum scores showed a mean absolute error of 2.9 points, which is smaller than the minimally detectable change of the ARAT.