CHARACTERIZATION AND ESTIMATION OF MUSCULOSKELETAL PAIN USING MACHINE LEARNING

<p>Traditional scales utilized for recording pain are known to be highly subjective and biased due to inaccuracies in recollecting actual pain intensities. As a result, machine learning (ML) models that are trained using these scores as ground truth are reported to have low performance for objective pain classification because of the huge disparity between what was felt in moments of pain and the scores recorded afterward.</p> <p>In the present study, two devices were designed for gathering real-time, continuous in-session subjective pain scores and the recording of the autonomic nervous system (ANS) altered endodermal (EDA) activity. 24 participants were recruited to undergo a post-exercise circulatory occlusion (PECO) with muscle stretch experiment for creation of discomfort. Concomitant EDA data and in-session scores emanating from each participant were ingested into a custom-built pain platform. Additionally, post-experiment subjective pain scores were retrieved from each participant using the visual analog scale (VAS). The collected data were analyzed in the time domain to extract corresponding EDA objective features and in-session ground truth targets for the development of ML models. Subsequently, random forest (RF) and multi-layer perceptron (MLP) ML models were developed for a 3-way classification of skeletal muscle pain intensities. Model performance was assessed using the macro-averaged geometric mean metric.</p> <p>Results obtained over 10-fold cross-validation indicate that ML models trained with the continuous in-session ground truth scores and objective features achieved a performance of 75.9 % and 78.3 % for MLP and RF respectively. Whereas ML models trained with VAS scores and objective features achieved 70.3 % and 74.6 % for MLP and RF respectively. Overall, this study eliminates the problem of high variance that causes data point imbalance and also provides evidence that ML performance for pain intensity characterization is vastly improved when continuous in-session ground truth scores are used over traditional or post-experiment ground truth scores (VAS).</p> <p>To the best of the author’s knowledge, this is the first study to collect both the ground truth scores from a continuous in-session device and an objective physiological marker of pain over the same time spectrum to yield better ML prediction performance.</p>

Name thesis:degree_name

Master of Science (MS)

Level thesis:degree_level

Masters Thesis

Year dc:date.available

2023

Author dc:creator

• Faremi, Boluwatife

Contributors dc:contributor

• Andrew H. Sung
• Zhaoxian Zhou
• Jonathon Stavres

• Objective
• Pain
• Machine learning
• Characterization
• Prediction
• EDA
• Artificial Intelligence and Robotics
• Bioelectrical and Neuroengineering
• Biomedical
• Biomedical Devices and Instrumentation
• Databases and Information Systems
• Data Science
• Data Storage Systems
• Digital Communications and Networking
• Electrical and Electronics
• Exercise Physiology
• Exercise Science
• Hardware Systems
• Laboratory and Basic Science Research
• Other Kinesiology
• Signal Processing
• Software Engineering

Repository record dc:identifier

https://aquila.usm.edu/masters_theses/985

OAI identifier oai:identifier

oai:aquila.usm.edu:masters_theses-2066