Linear vs. non-linear mapping of peak power using surface EMG features during dynamic fatiguing contractions

Area of Science:

• Biomechanics and Motor Control
• Exercise Physiology
• Biomedical Engineering

Background:

• Surface electromyogram (sEMG) signals reflect muscle activity during exercise.
• Estimating power output changes during fatiguing exercise is crucial for performance analysis.
• Comparing linear and non-linear modeling approaches for sEMG-based power estimation is needed.

Purpose of the Study:

• To compare the effectiveness of non-linear (neural network) and linear (linear regression) power mapping using sEMG features.
• To evaluate the accuracy of these models in estimating power output changes during fatiguing knee extension exercise.

Main Methods:

• Fifteen healthy participants performed leg press exercises to volitional fatigue.
• sEMG data from vastus medialis and lateralis muscles were recorded.
• Various sEMG variables (MAV, Fmed, FInsm5, MFM, WIRM, WIRW) were extracted and used as inputs for linear regression and neural network models.

Main Results:

• The non-linear (neural network) model showed higher correlation coefficients and signal-to-noise ratios compared to the linear model.
• These differences were not statistically significant.
• Both linear and non-linear approaches demonstrated comparable validity in estimating power output changes.

Conclusions:

• Non-linear mapping of sEMG signals offers a potentially more accurate method for estimating power output during fatiguing exercise.
• However, linear mapping remains a valid and equally effective approach given the lack of significant differences.
• Both methods can reliably estimate changes in peak power during repetitive, fatiguing knee extension exercises.