SHAPE VARIABILITY ANALYSIS OF SEMG AMPLITUDE DISTRIBUTION S. Boudaoud 1 , F. Ayachi 2 , C. Marque 1 1 Université de Technologie de Compiègne, Compiègne, France 2 Université de Picardie Jules Verne, Amiens, France Abstract Classically, the assessing of muscle activation is done using several amplitude estimators (RMS, ARV) supposing the Gaussianity of the sEMG signal distribution for formalization convenience. According to recent studies done by simulation and in experimental conditions, the sEMG signal distribution shape seems to be modified by many factors as contraction level, fatigue state, muscle anatomy and used instrumentation, and also motor control parameters. In this work, we propose to investigate the shape (or non-Gaussianity) variability of the sEMG amplitude distribution, according to contraction level, using High Order Statistics (HOS) and a recent functional formalism, the Core Shape Modelling (CSM). For this purpose, a large scale simulation (25 muscle anatomies) is performed, using parallel computing, to classify sEMG data generated during three contraction levels (20%, 50%, and 80% MVC). This classification is based solely on shape discrimination. From the results screening, it appears that the CSM method obtains, using Laplacian electrode arrangement, the higher classification scores even in presence of additional noise and anatomical variability. However, when some critical confounding parameters are modified, these scores decrease. These results demonstrate that shape information from sEMG amplitude distribution, if taken with caution, should be relevant for assessing the neuromuscular system. Keywords: surface electromyography, muscle force classification, non-Gaussianity, Motor Unit recruitment, shape analysis, core shape model, high order statistics. I. INTRODUCTION For amplitude estimation task, sEMG signal is often considered as a Gaussian random process for formalization convenience in the literature [1]. But, in experimental conditions, Laplacian distribution has been observed at low contraction level [1]. In fact, we also found recently using HOS and specific shape analysis method that the non-Gaussianity evolutes, in simulation and experimental conditions, according to contraction level, several anatomical (fat thickness), physiological (MU number and type), instrumental (electrode disposition) and control parameters (MU recruitment strategy and firing synchrony) [2], [3]. This study will bear on the analysis of the shape of the sEMG amplitude distribution for non-Gaussianity screening. This use is motivated by the assumption that information of MUs recruitment schemes are possibly contained is subtle modification of the distribution shape. This amplitude distribution shape depends on both active MUs firing statistics and corresponding Motor Unit Action Potentials (MUAPs) shapes. In fact, its guessed that the cumulative effect, induced by the MU recruitment, modifies enough the distribution shape when a suitable positioned electrode arrangement, that favours MUAPs asymmetry and reduces auto-cancellation phenomenon, is employed (monopolar or Laplacian). In this shape analysis, direct and indirect shape parameters will be used. The HOS parameters can be considered as indirect since the distribution shape monitoring (skewness or peakness) is done directly on the sEMG amplitude and not on its distribution. Another possibility is the direct access to the distribution shape using estimation and adapted shape analysis method. Recently, a functional formalism, namely the Core Shape Modelling (CSM) [4], has been