Â鶹´«Ã½Ó³»­

Â鶹´«Ã½Ó³»­

Explore the latest content from across our publications
Join The Â鶹´«Ã½Ó³»­

Log In

Forgot Password?
Create New Account

Loading... please wait

Abstract Details

Machine Learning for Rehabilitation Exercise “Dose” Measurement: protocol description and feasibility of inpatient rehabilitation data collection
Neuro-rehabilitation
P8 - Poster Session 8 (8:00 AM-9:00 AM)
15-001

To evaluate the feasibility of using biometric sensors and machine learning strategies to create an objective system for quantifying inpatient rehabilitation exercise participation post-stroke.
Prior research suggests that patients show best rates of motor recovery with a greater number of activity-based interventions (usually in the form of rehabilitation exercises).  However, the exact amount and optimal timing is, as of yet, unclear partly because it has been very difficult to objectively monitor the "dosing" of an exercise program. The ultimate goal of this project is to create an automated system using biometric superficial skin sensors and machine learning to track rehabilitation exercise participation by patients with arm weakness due to stroke and other neurological diseases. 

In this project, sensors are placed on the upper extremity of patients with weakness from stroke currently admitted in a hospital acute rehabilitation facility. The sensors collect information about accelerometry, gyroscopy, and/or surface electromyography throughout the subject’s daily activities, and information about their rehabilitation sessions (physical or occupational therapy) is obtained with video recordings. Information about sensor tolerability, technical issues related to sensor adhesion, and feasibility and content of therapy session recordings is collected as well.
Preliminary data shows high patient interest in study participation and good tolerance of study protocols. Protocol limitations include observed variability in rehabilitation exercises, as well as logistical and technical issues with regards to sensor adhesion, scheduling, and video recordings.
Superficial sensors are well tolerated and can be used to collect information about patient movements and muscle activation during acute rehabilitation admission. Thus, this project has shown great promise and may prove to be, with time and effort, ultimately successful. However, technical issues related to sensor adhesives and high variability in therapy session activities present additional challenges in developing machine learning algorithms to quantify post-stroke rehabilitation therapy “dose”.
Authors/Disclosures
Jeronimo Cardona
PRESENTER 		No disclosure on file
Ania Busza, MD, PhD (University of Rochester) Dr. Busza has received research support from NIH/NINDS. Dr. Busza has received personal compensation in the range of $500-$4,999 for serving as a Grant reviewer with NIH.
No disclosure on file
No disclosure on file