Automated estimation of clinical parameters by recurrence quantification analysis of surface EMG for agonist/antagonist muscles in amputees

Abstract

The surface electromyogram (sEMG) of amputees’ remnant muscles can be an appropriate candidate to estimate clinical parameters, such as DASH (disabilities of the arm, shoulder, and hand) score or PLSI (phantom limb sensation intensity) automatically which can be used instead of the patients’ self-report or specialist assessments. In this study, to assess the DASH score and PLSI, two artificial neural networks (ANNs) were established as non-linear models to map the dynamic behavior of sEMG to the clinical measures. Since the amputees miss some muscles, the muscle activation of the healthy subjects was used to find the synergistic muscles in wrist flexion and extension movements of the trans-radial amputees’ residual limb. Due to sEMG variability and its chaotic nature, the complex patterns of the synergistic agonist/antagonist muscles were extracted via the recurrence quantification analysis (RQA). The inverse of Determinism (DET) known as Complexity (CPX) was measured from RQA. The average of CPX in synergistic muscles was fed to the two ANNs to estimate the DASH score and PLSI. The performance of mapping the sEMG complexity features to clinical measurements were acceptable and reliable. The correlation between actual DASH score and PLSI and their estimated values were 96 % and 89 %, respectively. The interclass correlation coefficients of the estimated DASH score and PLSI were 94.9 % and 81.9 %. The automated DASH score and PLSI assessment based on muscle dynamics can construct a potential approach to the setting of the unsupervised rehabilitation training and tracking the improvements, objectively.