A factorization‐based algorithm to predict EMG data using only kinematics information

Abstract

EMG analyses have several applications, such as identifying muscle excitation patterns during rehabilitation or training plans, or controlling EMG‐driven devices. However, experimental measurements can be time consuming or difficult to obtain. This study presents a simple algorithm to predict EMG signals that can be applied in real time during running, given only the instantaneous vector of kinematics. We hypothesize that the factorization of the kinematics of the skeleton together with the EMG data of calibration subjects could be used to predict EMG data of another subject only using the kinematic information. The results showed that EMG signals of lower‐limb muscles can be predicted accurately in less than a second using this method. Correlation coefficients between predicted and experimental EMG signals were higher than 0.7 in 10 out of 11 muscles for most prediction trials and subjects, and their overall median value was higher than 0.8. These values confirm that this method could be used to accurately predict EMG signals in real time when only kinematics are measured.

Effects of simulated hypo-gravity on lower limb kinematic and electromyographic variables during anti-gravitational treadmill walking

“…The results of this study provide additional evidence that kinematic and electromyographic features do not scale across load levels proportionally with each other ( Cappellini et al, 2006 ; Ivanenko et al, 2002 ). While it possible to accurately predict muscle activity from kinematics alone ( Manzano and Serrancoli, 2021 ), work by Mauntel et al (2017) did find kinematic-EMG decoupling depending on the type of movement being performed. In that way, the findings of this study—that kinematics scale linearly down to 20% of body weight, while muscle activity displays non-linear scaling as weight is decreased similarly—are not particularly surprising.…”