NeuroMotion: Open-source platform with neuromechanical and deep network modules to generate surface EMG signals during voluntary movement

Shihan Ma, Irene Mendez Guerra, Arnault H. Caillet, Jiamin Zhao, Alexander Kenneth Clarke, Kostiantyn Maksymenko, Samuel Deslauriers-Gauthier, Xinjun Sheng, Xiangyang Zhu, Dario FarinaPLOS Computational Biology 2024 · DOI · OpenReview

Introduction

In the field of neuromechanics, understanding how the nervous system controls the musculoskeletal system to produce movement is a fundamental challenge. A critical component of this research is the electromyography (EMG) signal—the electrical activity recorded from muscles. While robust tools exist for simulating biomechanics (e.g., OpenSim) and others for simulating EMG during static (isometric) contractions, a significant gap has persisted: the ability to simulate realistic EMG during voluntary, dynamic movement.

Existing EMG simulators are often computationally prohibitive or limited to static conditions where muscle length and electrode position remain constant. This limitation hinders the development and validation of "decoders"—the algorithms used in prosthetics and exoskeletons to translate muscle signals into intended joint angles—without relying on massive amounts of expensive, difficult-to-collect experimental data.

Methodology

NeuroMotion addresses this gap by employing a modular, three-part architecture designed to bridge the transition from neural intent to electrical output. The platform integrates musculoskeletal modeling with deep learning to capture the non-stationary properties of muscle signals during motion.

Overview of the NeuroMotion architecture and signal generation pipeline, showing the transition from motion to EMG signals through musculoskeletal and deep learning modules.

Core Modules

  1. Musculoskeletal (MSK) Module: Utilizing the OpenSim API, this module takes joint kinematics (such as wrist angles) as input. It estimates the muscle fiber lengths and muscle activations required to execute the specified movement.
  2. Motor Unit Pool Module: This module transforms muscle activations into "spike trains"—the specific discharge timings of individual motor units. It supports both classical models (Fuglevand) and more complex Leaky Fire-and-Integrate (LIF) neuron models.
  3. BioMime (Deep Network Module): This is the platform's core innovation. BioMime is a deep conditional generative neural network that efficiently generates Motor Unit Action Potentials (MUAPs). Unlike static models, BioMime receives non-stationary physiological inputs (like changing muscle fiber lengths and depths) and outputs how the MUAP "morphs" in real-time as the muscle moves.

The final surface EMG is generated by convolving the dynamic MUAPs from BioMime with the spike trains from the Motor Unit Pool.

Key Insights and Results

The "Morphing" MUAP

A significant contribution of NeuroMotion is the demonstration of how MUAP waveforms change during movement. As muscles shorten or move deeper under the skin, the duration and amplitude of the electrical signal shift. NeuroMotion captures these non-stationary properties, which are often ignored in simpler models but are vital for realistic simulation.

Computational Efficiency and Validation

BioMime achieves ultra-fast inference, taking only 0.287 seconds to simulate a muscle's activity for a given condition, making large-scale data generation feasible. The platform was validated against experimental data from six human subjects performing wrist and finger flexion/extension tasks.

Zero-Shot Regression and Data Augmentation

Remarkably, the authors demonstrated that a Ridge regressor trained entirely on synthetic data generated by NeuroMotion could be directly applied to real-world experimental data to predict joint angles with significant accuracy (PCC of 0.52 for wrist and 0.61 for finger angles). Furthermore, when synthetic data was used to augment limited experimental datasets, prediction accuracy improved significantly (e.g., MCP joint PCC increased from 0.66 to 0.72).

Conclusion and Future Work

NeuroMotion provides a powerful tool for "in-silico" experiments, allowing researchers to see exactly how specific physiological changes, such as muscle fatigue or electrode shifts, affect the resulting EMG. While currently limited by anatomical specificity to the forearm and simplified biomechanical assumptions (like rigid tendons), future iterations aim to incorporate subject-specific training and custom electrode placements. The platform is fully open-source, with code available on GitHub and experimental data hosted on Figshare, inviting broad community collaboration in neural interface design.

Summary updated

Abstract

Neuromechanical studies investigate how the nervous system interacts with the musculoskeletal (MSK) system to generate volitional movements. Such studies have been supported by simulation models that provide insights into variables that cannot be measured experimentally and allow a large number of conditions to be tested before the experimental analysis. However, current simulation models of electromyography (EMG), a core physiological signal in neuromechanical analyses, remain either limited in accuracy and conditions or are computationally heavy to apply. Here, we provide a computational platform to enable future work to overcome these limitations by presenting NeuroMotion, an open-source simulator that can modularly test a variety of approaches to the full-spectrum synthesis of EMG signals during voluntary movements. We demonstrate NeuroMotion using three sample modules. The first module is an upper-limb MSK model with OpenSim API to estimate the muscle fibre lengths and muscle activations during movements. The second module is BioMime, a deep neural network-based EMG generator that receives nonstationary physiological parameter inputs, like the afore-estimated muscle fibre lengths, and efficiently outputs motor unit action potentials (MUAPs). The third module is a motor unit pool model that transforms the muscle activations into discharge timings of motor units. The discharge timings are convolved with the output of BioMime to simulate EMG signals during the movement. We first show how MUAP waveforms change during different levels of physiological parameter variations and different movements. We then show that the synthetic EMG signals during two-degree-of-freedom hand and wrist movements can be used to augment experimental data for regressing joint angles. Ridge regressors trained on the synthetic dataset were directly used to predict joint angles from experimental data. In this way, NeuroMotion was able to generate full-spectrum EMG for the first use-case of human forearm electrophysiology during voluntary hand, wrist, and forearm movements. All intermediate variables are available, which allows the user to study cause-effect relationships in the complex neuromechanical system, fast iterate algorithms before collecting experimental data, and validate algorithms that estimate non-measurable parameters in experiments. We expect this modular platform will enable validation of generative EMG models, complement experimental approaches and empower neuromechanical research.
BibTeX
@article{ma2024neuromotion,
  author = {Shihan Ma and Irene Mendez Guerra and Arnault H. Caillet and Jiamin Zhao and Alexander Kenneth Clarke and Kostiantyn Maksymenko and Samuel Deslauriers-Gauthier and Xinjun Sheng and Xiangyang Zhu and Dario Farina},
  title = {{NeuroMotion: Open-source platform with neuromechanical and deep network modules to generate surface EMG signals during voluntary movement}},
  journal = {PLOS Computational Biology},
  year = {2024},
  doi = {10.1371/journal.pcbi.1012257},
}
Figure