Improving EMG-Based Muscle Force Estimation by Using a High-Density EMG Grid and Principal Component Analysis

Abstract

Abstract-The accuracy of predictions of muscle force based on electromyography (EMG) is an important issue in biomechanics and kinesiology. Since human skeletal muscles show a high diversity and heterogeneity in their fiber architecture, it is difficult to properly align electrodes to the muscle fiber direction. Against this background, we analyzed the effect of different bipolar configuration directions on EMG-based force estimation. In addition, we investigated whether principal component analysis (PCA) can improve this estimation. High-density surface-EMG from the triceps brachii muscle and the extension force of the elbow were measured in 11 subjects. The root mean square difference (RMSD) between predicted and measured force was determined. We found the best bipolar configuration direction to cause a 13% lower RMSD relative to the worst direction. Optimal results were obtained with electrodes aligned with the expected main muscle fiber direction. We found that PCA reduced RMSD by about 40% compared to conventional bipolar electrodes and by about 12% compared to optimally aligned multiple bipolar electrodes. Thus, PCA contributes to the accuracy of EMG-based estimation of muscle force when using a high-density EMG grid.Index Terms-Force estimation, heterogeneous muscle fiber architecture, human, principal component analysis, redundancy, surface electromyography.

Redundancy or heterogeneity in the electric activity of the biceps brachii muscle? Added value of PCA-processed multi-channel EMG muscle activation estimates in a parallel-fibered muscle

“…1): proximal or distal single bipolar channels over BB (1p, 1d), single proximal and distal rows of multiple bipolar channels (2p, 2d), multiple rows of bipolar channels over the entire proximal or distal part (3p, 3d), and multiple rows of bipolar channels over the entire BB (4). Alternatively, the same monopolar channels, which were used for the different configurations described above, were processed with PCA as proposed in a previous study (Staudenmann et al, 2006), yielding the conditions: 2pPCA, 2dPCA, 3pPCA, and 3dPCA. The first four principal components were discarded, as a means to eliminate common information from the monopolar EMG data in an unbiased way (Staudenmann et al, 2006(Staudenmann et al, , 2009a.…”

“…Although muscle activation is not a physiological variable, it is a commonly used term to represent the combined effect of recruitment and rate coding in a scalar value, ranging from 0% to 100%. Estimates of muscle activation based on conventional bipolar EMG, however, have limitations (Farina et al, 2004;Staudenmann et al, 2006). Importantly, activity can be heterogeneous within muscles during voluntary contractions, as was shown with multi-channel EMG (Eriksson et al, 1984;Holtermann et al, 2005;Staudenmann et al, 2005Staudenmann et al, , 2009avan Zuylen et al, 1988, Vieira et al, 2011Wolf et al, 1998), as well as with ultrasound imaging (Bojson-Moller et al, 2004;Darby et al, 2012;Finni et al, 2003;Pappas et al, 2002).…”

“…A bipolar configuration should preferably be aligned to the muscle fiber direction (De Luca, 1997;Farina et al, 2004;Hermens et al, 2000). Otherwise the measured EMG amplitude (Merletti et al, 1999) and the quality of muscle activation estimates are reduced and more sensitive to changes in joint angle (Staudenmann et al, 2006). Recently, it was shown that the problem of electrode alignment can be overcome with principal component analysis (PCA), which was applied on multiple monopolar EMG channels and is thus unbiased with respect to the muscle fiber direction (Staudenmann et al, 2006).…”

“…Otherwise the measured EMG amplitude (Merletti et al, 1999) and the quality of muscle activation estimates are reduced and more sensitive to changes in joint angle (Staudenmann et al, 2006). Recently, it was shown that the problem of electrode alignment can be overcome with principal component analysis (PCA), which was applied on multiple monopolar EMG channels and is thus unbiased with respect to the muscle fiber direction (Staudenmann et al, 2006). So far, PCA-processed monopolar multi-channel EMG has been applied on the triceps brachii, and triceps surae (Staudenmann et al, 2006(Staudenmann et al, , 2009a, both mainly pennate and architecturally complex muscles (Chow et al, 2000;Kawakami et al, 1993).…”

“…Recently, it was shown that the problem of electrode alignment can be overcome with principal component analysis (PCA), which was applied on multiple monopolar EMG channels and is thus unbiased with respect to the muscle fiber direction (Staudenmann et al, 2006). So far, PCA-processed monopolar multi-channel EMG has been applied on the triceps brachii, and triceps surae (Staudenmann et al, 2006(Staudenmann et al, , 2009a, both mainly pennate and architecturally complex muscles (Chow et al, 2000;Kawakami et al, 1993). A substantial improvement in muscle activation estimates was achieved with PCA as compared to multiple bipolar EMG using the same electrodes (Staudenmann et al, 2006(Staudenmann et al, , 2009a, suggesting that the misalignment of (multiple) bipolar EMG is a cause of errors in the estimation.…”