Rehabilitation Robotics and Machine Learning for Stroke Severity Classification - Thesis (Master of Science)

Aug 2, 2026

Upload by Jiven

Stroke therapy reduces impairments and supports motor recovery by leveraging neuroplasticity. This thesis applies supervised machine learning to classify stroke severity using clinician-labeled data. Thirty-three chronic-stroke patients completed rehabilitation activities while Motus Nova technology recorded upper and lower body motion. Features derived from sensor data included range-of-motion and pressure statistics and movement counts (e.g., force flexion/extension). Using a harmonized dataset of about 32,000 sessions and light gradient boosting with 10-fold cross-validation, the study achieved about 94% average accuracy, enabling objective identification of severity classes to support improved future outcomes.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages69

Reading Time2 h 54 min

AccessFree to download

Introduction

Background

Methods and Procedure

Data Collection and Harmonization

Data Collection

Clinician Labeling

Data Harmonization

Exploratory Data Analysis

Methodologies

ScholarWorks@GSU Rehabilitation Robotics and Machine Learning for Stroke Severity Classification | Authors | Greenfield , Raymond | | --- | --- | | Citation | Greenfield , Raymond. "Rehabilitation Robotics and Machine Learning for Stroke Severity Classification." 2022. Thesis , Georgia State University. [https://doi.org/10.57709/32980067](https://doi.org/10.57709/32980067) | | DOI | [https://doi.org/10.57709/32980067](https://doi.org/10.57709/32980067) | | Download date | 2026-03-07 07:55:25 | | Link to Item | [https://hdl.handle. net/20.500.14694/10481](https://hdl.handle. net/20.500.14694/10481) | Rehabilitation Robotics and Machine Learning for Stroke Severity Classification by Raymond McCoy Greenfield Under the Direction of Committee Chair’s Igor Belykh, Ph.D. A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree of Master of Science in the College of Arts and Sciences Georgia State University 2022 ABSTRACT Stroke therapy is essential to reduce impairments and improve motor movements by engaging autogenous neuroplasticity. This study uses supervised learning methods to address an autonomous classification via stroke severity labeled data by a clinician. Thirty-three patients with chronic stroke performed a variety of rehabilitation activities while utilizing the Motus Nova rehabilitation technology to capture upper and lower body motion. Based on the minimum, maximum, and mean of the range of motion and pressure as well as the number of movements, force flexion, and extension for each game and session provided from the sensor data. Supervised learning methods were applied to a harmonized dataset of roughly 32,000 patient sessions based on the maximum score per session per game. With this approach using light gradient boosting methods we achieved an average of 94% accuracy with 10-fold cross-validation to prevent overfitting. This thesis shows objectively-measured rehabilitation training, enabling the identification of the stroke severity class with the hopes to have patients have a less severe class in the future. Over the last 10 years robotic rehabilitation has been utilized in inpatient therapy. Robotic rehabilitation has been shown to be effective in improving the severity of stroke in some cases. In particular, robotic devices can be used to help stroke survivors regain movement, improve their functional abilities and improve depression (11) . These devices can provide a high level of precision and repeatability, allowing patients to perform therapeutic exercises with greater accuracy and consistency (1) . Additionally, because robotic devices can be programmed to provide different levels of assistance, they can be tailored to the individual needs of each patient. This allows for a more personalized and effective rehabilitation in-home program (21) . INDEX WORDS: Machine Learning, Artificial Intelligence, Deep Neural Network, Gradient Boosting, Physical Therapy, Neuroplasticity Copyright by Raymond McCoy Greenfield 2022 Rehabilitation Robotics and Machine Learning for Stroke Severity Classification by Raymond McCoy Greenfield Committee Chair: Committee: Alexandra Smirnova Digitally signed by Alexandra Smirnova DN: cn=Alexandra Smirnova, o, ou, [email=asmirnova@gsu.edu](email=asmirnova@gsu.edu), c=US Date: 2022.12.17 21:19:02-05'00' Electronic Version Approved: Igor Belykh Vladimir Bondarenko Alexandra Smirnova Office of Graduate Studies College of Arts and Sciences Georgia State University December 2022 iv DEDICATION This thesis is dedicated to the stroke victims of the world in an effort to have a higher quality of life. v ACKNOWLEDGMENTS Above all, I offer my genuine gratitude to my major advisor Professor Igor Belykh and Dr. Rusell Jeter, both of who inspired me with their support and guidance. Without their advice and help, I could not have developed the knowledge and accomplished this thesis. Dr. Jeter has not only guid

FAQs

How is stroke severity classification performed in this study?

The study uses supervised learning with clinician-labeled stroke severity data and sensor-derived features from rehabilitation sessions.

What data source and device are used to capture patient movement?

Participants performed rehabilitation activities using Motus Nova rehabilitation technology, which records upper and lower body motion for feature extraction.

What machine learning approach and evaluation method were used to measure performance?

Light gradient boosting models were trained on a harmonized dataset of roughly 32,000 sessions and evaluated with 10-fold cross-validation to reduce overfitting.

Aug 2, 2026

Upload by Jiven