EasiSMR: Recognizing Non-Specific Low Back Pain Symptomatic Muscles Using Multi-Muscles Fusion based Machine Learning

Non-specific low back pain (NLBP) is a symptom of the unknown pathoanatomical cause. The treatment focuses on reducing pain and disability based on doctor's experiences, which is usually with the insufficient support of objective evidence. Helping doctors recognize symptomatic muscles and giving patients more precise treatment accordingly are big practical challenges. Surface electromyography (sEMG) is the collective electric signal from muscles, which is a widely used and noninvasive way to have an insight into the muscles. With the rapid development of machine learning and signal processing in biomedical engineering, many problems requiring individual judgment can possibly be solved by building appropriate computing models based on the heterogeneous and complex dataset, as the approach we used to address the practical challenges. In this work, a novel NLBP Symptomatic Muscles Recognition method (EasiSMR) is proposed to accurately locate the symptomatic muscles, and an objective evaluation is then provided based on wearable sEMG measurements. EasiSMR is designed to build machine learning models for six low back muscles separately which enables to recognize the symptomatic muscles via an isometric exercise. EasiSMR possesses two main contributions to enhance its recognition performance: a Multi-Muscles Fusion method (MMF) which takes the muscle compensation information into account, and a set of new features are extracted for isometric exercise which reflects the characteristics of muscle firing phase. Intensive experiments were carried out to verify the performance of EasiSMR on the real-world NLBP sEMG dataset collected in hospital. Results demonstrate that the proposed method achieves an average recognition accuracy of 86.16% for all six low back muscles and shows potential value in NLBP treatment.