Cumulative link mixed models and clustering of EMG signals for pain assessment in lateral epicondylitis

Abstract

This Master’s Thesis comprises the formulation, inference and post-hoc analysis of ordinal regression models to describe elbow pain reporting from people with a lateral epicondylitis condition. Furthermore, an experiment is designed to collect pain reports and electromyography signals from 38 individuals in a strictly controlled setup. Explanatory variables are extracted from the forearm muscles electrical activity and converted into input features for the regression model. PRECURE ApS provided a data set with pain reports as well as an IoT device commercially called MLI Elbow to record muscle activity of target individuals. A cumulative link mixed model with scale effects is proposed for the PRECURE’s pain reports data set while a cumulative link model with scale effects is proposed for the pain reports gathered in the experiment. A signal processing pipeline and a clustering algorithm are designed for an exploratory analysis of the electromyography signals collected in the experiment. The final models provides interesting and valuable findings about the understanding of factors affecting pain reports. Moreover, the study of elbow pain, being this the physiological and psychological manifestation of lateral epicondylitis, disclose patterns in the risk factors involved in developing this condition.