Related Experiment Video
Updated: Aug 8, 2026
09:27
An Emerging Target Paradigm to Evoke Fast Visuomotor Responses on Human Upper Limb Muscles
Published on: August 25, 2020
Machine Learning Methods to Predict Unmeasured Muscle Activation in Upper Limb Reaching Task for Assessing
Baivab Bhandari1, Shadman Tahmid2, James Yang1
1Human-Centric Design Research Lab, Department of Mechanical Engineering, Texas Tech University, Lubbock, TX 79409, USA.
Journal of Biomechanical Engineering
|August 7, 2026
Summary
This study introduces a deep learning framework to predict muscle activation using surface electromyography (sEMG) alone. The CNN-LSTM model accurately estimates muscle activation patterns, even for unmeasured muscles.
Area of Science:
- Biomechanics
- Neuroscience
- Machine Learning
Background:
- Estimating muscle activation is crucial for neuromuscular modeling but faces challenges with limited sensor data and noisy surface electromyography (sEMG) signals.
- Traditional methods like Inverse Dynamics and Static Optimization require extensive kinematic and kinetic data, limiting their application.
- Interpreting sEMG signals for direct muscle activation is complex due to non-linear signal characteristics.
Purpose of the Study:
- To investigate the feasibility of a deep learning framework for learning generalizable inter-muscular activation relationships using only sEMG data.
- To evaluate the effectiveness of different deep learning architectures (CNN, LSTM, CNN-LSTM) in capturing spatial and temporal dynamics of EMG signals.
- To assess the impact of training data size and input muscle selection on prediction accuracy.
Main Methods:
- A deep learning framework was developed using sEMG data from 30 participants performing a standardized forward-reaching task.
- Three model architectures were evaluated: Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM) networks, and hybrid CNN-LSTM models.
- Systematic variations included the number of unmeasured output muscles, training dataset size, and Leave-One-Muscle-Out (LOMO) analysis.
Main Results:
- The hybrid CNN-LSTM model demonstrated superior performance compared to standalone CNN and LSTM models, especially for multi-muscle prediction.
- The CNN-LSTM model achieved high accuracy (RMSE=0.103, r=0.866) in predicting the activation of the triceps long head muscle.
- LOMO analysis indicated the importance of specific input muscles for accurate prediction.
Conclusions:
- Deep learning, particularly the CNN-LSTM architecture, is feasible for scalable, non-invasive prediction of muscle activation patterns from sEMG.
- This approach enables simultaneous prediction of superficial muscle activations and provides a foundation for proxy prediction of difficult-to-measure muscles.
- The findings support the use of sEMG-based deep learning models for advancing neuromuscular modeling and biomechanical analysis.
