Deep Muscle EMG construction using A Physics-Integrated Deep Learning approach

TL;DR

This work introduces a physics-integrated neural musculoskeletal model (NMM) that reconstructs non-measurable deep muscle EMG envelopes $e_b$ and $e_t$ from joint kinematics and load by embedding musculoskeletal forward dynamics within a transformer-based PINN. The model jointly optimizes measured EMG data and physics-based torque constraints, while personalizing 32 MSK parameters for each subject to achieve subject-specific EMG reconstructions. Compared with a muscle synergy extrapolation (MSE) baseline using NMF, the NMM demonstrates superior temporal and spatial EMG accuracy, more physiologically plausible activation magnitudes, and improved joint-torque predictions across multiple loading conditions. The approach, validated on five subjects performing elbow flexion-extension with varying loads, offers a non-invasive pathway to reconstruct deep-muscle activity, with potential applications in rehabilitation, ergonomics, and wearable robotics; future work includes extending to gait, incorporating IMUs for velocity data, and expanding to additional joints and muscles.

Abstract

Electromyography (EMG)--based computational musculoskeletal modeling is a non-invasive method for studying musculotendon function, human movement, and neuromuscular control, providing estimates of internal variables like muscle forces and joint torques. However, EMG signals from deeper muscles are often challenging to measure by placing the surface EMG electrodes and unfeasible to measure directly using invasive methods. The restriction to the access of EMG data from deeper muscles poses a considerable obstacle to the broad adoption of EMG-driven modeling techniques. A strategic alternative is to use an estimation algorithm to approximate the missing EMG signals from deeper muscle. A similar strategy is used in physics-informed deep learning, where the features of physical systems are learned without labeled data. In this work, we propose a hybrid deep learning algorithm, namely the neural musculoskeletal model (NMM), that integrates physics-informed and data-driven deep learning to approximate the EMG signals from the deeper muscles. While data-driven modeling is used to predict the missing EMG signals, physics-based modeling engraves the subject-specific information into the predictions. Experimental verifications on five test subjects are carried out to investigate the performance of the proposed hybrid framework. The proposed NMM is validated against the joint torque computed from 'OpenSim' software. The predicted deep EMG signals are also compared against the state-of-the-art muscle synergy extrapolation (MSE) approach, where the proposed NMM completely outperforms the existing MSE framework by a significant margin.

Figures (12)

• Figure 1: Schematic architecture of the proposed Neural Musculoskeletal Model for constructing EMG signals from joint kinematics.
• Figure 2: Illustration of the experimental setup used for data collection.
• Figure 3: Training loss history of the proposed neural musculoskeletal model.
• Figure 4: Description of the measured and constructed EMG envelopes at different loads. The EMG envelopes at the third and sixth muscles are constructed from the proposed neural musculoskeletal model, whereas EMG envelopes at other muscles are directly observed using sensors. The EMG signals at the third and sixth muscles at different loads are constructed using a single NMM.
• Figure 5: Heatmaps displaying the comparison of EMG profiles for the NMM and MSE methods across subjects (S-1 to S-5) and different load conditions (0 kg, 2 kg, and 4 kg). The plots represent two muscles: EMG2 (Brach) and EMG5 (Tri med) at each load. Each subplot provides insights into the temporal variations of muscle activation across subjects and methods under varying loads.