A machine learning approach for clinical gait analysis and classification of polymyalgia rheumatica using myoelectric sensors

Abstract

The study focuses on Polymyalgia Rheumatica (PMR), an autoimmune musculoskeletal disease primarily affecting the shoulder blade and hip muscles in older adults, particularly women aged 50 and above. The research aims to address two main challenges: the need for more clarity on the disease's pathophysiology and the challenge of identifying disease severity in patients. The study introduces a novel approach involving movement assessment, by designing a low-cost MyoTracker system, and using electromyography (EMG) features to understand the impact on patients' hip muscles. A clinical trial was conducted at Komfo Anokye Teaching Hospital in Ghana, where the study employed a qualitative research approach to monitor movement patterns. Participants were tasked to perform exercises comprising of gait, knee lifting, and knee extension with sensors attached to the hip muscles. This research unfolds in three iterations, the first investigation involved hip muscular imbalances where the significant difference between patients and healthy controls in the maximum voluntary contraction (MVC) values was recorded. The bilateral difference computed between the left and right hip in patients exhibited 15% MVC on average compared to the healthy control group's 6%, indicating substantial hip muscular imbalances. The second iteration involved a movement assessment to identify specific movement patterns in patients. Support Vector Machine (SVM) achieves 85% accuracy for gait exercises, while Decision Tree (DT) performs less efficiently at 70%. SVM also excels in knee lifting exercises (70% accuracy), outperforming DT (60%). Based on hip muscle activation, patients' movement patterns significantly differ from healthy controls. In the third iteration, deep learning techniques, specifically RNN-LSTM and Vision Transformer (ViT), classify PMR disease severity based on EMG features. The study's results carry significant clinical implications with the evidence of hip muscular imbalances aiding in designing tailored rehabilitation protocols. Importantly, this study uses a cost-effective method for determining disease severity, enabling predictions about patients with severe PMR conditions. The key contribution of this thesis is the identification of patients’ specific movement patterns and the determination of PMR severity among patients. Other contributions are the detection of hip muscular imbalance in patients and the design of rehabilitation protocols to address hip muscular imbalances and improve patients' range of motion, enhancing overall well-being. In conclusion, this comprehensive study leverages innovative approaches, from a MyoTracker system for movement assessment to deep learning models, to unravel the complexities of PMR disease. The collaboration with medical experts emphasises the potential real-world impact of this research in enhancing the treatment and recovery processes for individuals.