Low-Back Pain Patients Classification Based on sEMG Activation Patterns Detection

Abstract

Low back pain (LBP) diagnostic challenge to assigning patients into more specific groups has been detected as one of the key obstacles in ensuring successful rehabilitation treatments. Novel approaches treat LBP patients as a non-homogeneous group where a more individualized approach is proposed, thus leading to a need for an effective subgrou**. In this paper, the possibility to differentiate chronic low back patients (CLBP) from LBP patients with radiculopathy (RLBP), by means of surface electromyography (sEMG), was examined. A feature model based on muscle activation patterns was proposed by measuring time distances between consecutive pairs of distinguished myoelectric events. For each sEMG channel, median and 90^th percentile values of time distances were calculated. This procedure was applied to six simple raw features: zero crossing (ZC), signal slope change (SSC), Willison amplitude (WAMP), variance (VAR), relative variance difference (RVD), and permutation entropy (PE). The classification differentiation task (CLBP vs. RLBP) was accomplished by employing support vector machines (SVM) and nearest neighbor (kNN) classifier variants. The presented approach demonstrated consistent high-accuracy classification results (0.90) for the given CLBP and RLBP subgrou**. The obtained results suggest that muscle activation patterns are worth further exploring as part of interpretable LBP patients’ subgrou** efforts.