Feasibility of estimating isokinetic knee torque using a neural network model

Area of Science:

• Biomechanics and Motor Control
• Computational Neuroscience
• Rehabilitation Engineering

Background:

• Electromyography (EMG) and joint torque relationships are extensively studied.
• Previous research utilized artificial neural networks (ANNs) for elbow joint torque estimation.
• A need exists for accurate knee joint torque estimation methods.

Purpose of the Study:

• To develop and evaluate artificial neural network (ANN) models for estimating net isokinetic knee torque.
• To compare the accuracy of ANN models against stepwise regression models.

Main Methods:

• Collected isokinetic knee extensor and flexor torque data alongside agonist and antagonist EMG.
• Utilized concentric and eccentric contractions at 30°/s and 60°/s joint velocities.
• Trained a three-layer ANN model using an adjusted back-propagation algorithm with variables including age, gender, height, body mass, EMG, joint position, and velocity.

Main Results:

• ANN models achieved significantly higher torque estimation accuracy (R=0.96) compared to stepwise regression (R=0.71).
• Stepwise regression identified body mass, height, joint position, and agonist EMG as key predictors.
• ANN model accuracy improved with an increased number of hidden units, demonstrating good generalization capabilities.

Conclusions:

• Artificial neural network (ANN) models provide a feasible and accurate technique for estimating net isokinetic knee torque.
• ANNs offer superior performance over traditional regression methods for this application.
• The developed ANN model shows potential for generalization to larger populations.