Muscle Force Estimation Using Data Fusion from High-Density SEMG Grid S. Allouch, M. Al Harrach, S. Boudaoud, J. Laforet UMR CNRS 7338, Biomechanics and Bio-engineering University of Technology of Compiegne (UTC) Compiègne, France mariam.harrach@utc.fr, samar.allouch@utc.fr, sofiane.boudaoud@utc.fr, jeremy.laforet@utc.fr F.S. Ayachi Multimodal Interaction Laboratory, SIS-McGill University Montréal, QC, Canada sofiane.ayachi@mcgill.ca R. Younes Faculty of engineering, Lebanese university Beirut, Lebanon, ryounes@ul.edu.lb Abstract— The aim of the proposed work is to evaluate, by simulation, the introduction of a data fusion process from a HD-sEMG grid (8X8) to improve the muscle force estimation from sEMG signal. For this purpose, twelve electrode arrangements are combined to dimension reduction technique (PCA or channel averaging) to obtain a monodimensional sEMG signal. After, this signal is used in a sEMG-force relationship model to estimate the muscular force. In fact, two models, with different complexity, and used in the biomechanics community are studied. In the simulation, three isometric contractions are simulated (20%, 50% and 80% MVC) using a recent sEMG-force generation model. Finally, the Normalized RMS Difference (NRMSD) between the estimated force and the simulated force by the sEMG-force generation model is calculated for each combination (electrode arrangement and dimension reduction technique, force estimator). According to the obtained results, the combination PCA and Laplacian arrangement gave the best fitting using the second force estimator while the best result obtained for the first force estimator is with the Right Diagonal Bipolar (DBR) arrangement combined with channel averaging. In future works, these force estimators, combined to HD-sEMG data fusion, will be experimentally evaluated. Keywords—Data fusion, muscle force estimation, HD-sEMG, PCA, sEMG-Force relationship modeling. I. INTRODUCTION Surface electromyography (sEMG) is an important tool in biomechanics, neurophysiology and physical rehabilitation, it is widely used to estimate muscle activation and force from the signal amplitude [1]. Many studies on muscle joint systems have shown that sEMG signals depend on the variability at the level of individual muscles [2], the level of muscle parts [3] and the level of motor units (MU) [4], in addition it varies over time. To improve the prediction of muscle force based on EMG, it important to reduce any annoying variability. Many studies have shown that improving estimation of muscle force based on sEMG can be by using bipolar electrodes pairs [5] and with a grid of monopolar electrodes which are densely spaced (i.e., high-density EMG grid) [6]. With high-density EMG grids many monopolar EMG signals are collected over a relatively small collection surface. The EMG signals can be organized in different ways. e.g., bipolar configurations (vertical, horizontal and diagonal), higher order derivations (the Laplacian configuration) [7] and using the principal component analysis (PCA) as an unbiased statistical method to detect the redundancy which can be in the recorded signals. The aim of this article is to compare the performance of several methods of merging data in order to obtain the best estimate of muscle force based on the synthesized data provided by a realistic sEMG-force generation model [8] developed in our team. Assessing muscle force estimation accuracy in simulation is interesting since we have access to a simulated intrinsic force which is very complex to obtain in experimentation. However, the used models suffer from several simplifications that must be taken into account. In this study, twelve electrode arrangements combined to reduction dimension techniques (PCA or averaging) will be coupled to two force estimators from sEMG signal. The first one is based on the work of Staudenmann [6] using simple filtering and rectification on the sEMG signal. The second one employed the Hill type muscle model [1],[9], widely used in biomechanics, that takes into account muscle activation and deformation and possible nonlinear behavior. After briefly recalling the concept of sEMG-Force simulation and data fusion, we present, in this study, the accuracy results in force estimation using the twelve electrode arrangements using PCA or channel averaging. Finally, the obtained results are discussed and some perspectives are given. A. sEMG-Muscle force Relationship modeling Method 1: According to Staudenmann [6], generated EMG signals are high-pass filtered (10 Hz), compensate for an estimated electromechanical delay (100 ms). After, the 64 sEMG signals are merged with a specific electrode configuration combined to PCA or channel averaging, the result signals are full wave rectified, averaged over the EMG channels (classical or PCA), to get one dimension 2013 2nd International Conference on Advances in Biomedical Engineering 978-1-4799-0251-4/13/$31.00 ©2013 IEEE 195