A neural network model for simulation of torso muscle coordination

Area of Science:

• Biomechanics
• Computational Neuroscience
• Biomedical Engineering

Background:

• Understanding lumbar muscle function is crucial for addressing back pain and improving biomechanical models.
• Previous artificial neural network (ANN) models often rely on predefined muscle activity targets.
• Simulating motor control requires accurately representing muscle responses to external forces.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for simulating lumbar muscle responses to static moment loads.
• To investigate if muscle activity patterns can emerge from moment equilibrium constraints rather than explicit training targets.
• To explore the role of muscle competition in motor control and recruitment strategies.

Main Methods:

• An artificial neural network (ANN) was designed to mimic a motor control system focused on moment equilibrium.
• A modified backpropagation algorithm with moment equilibrium constraints was used for iterative training (parameter optimization).
• Assumptions on muscle moment-generating capacity and inter-muscle competition were incorporated.

Main Results:

• The ANN model successfully predicted realistic lumbar muscle activity patterns, validated against experimental electromyographic (EMG) data (r²: 0.4-0.9).
• Muscle activity patterns emerged as a consequence of satisfying moment equilibrium constraints.
• Varying competition parameters allowed prediction of alternative recruitment strategies and co-contraction levels.

Conclusions:

• Artificial neural network (ANN) simulations can effectively mimic motor recruitment plans using simplified systems.
• Competitive interactions between muscles appear to be fundamental to the motor learning process.
• The model provides insights into the intrinsic mechanisms governing muscle recruitment and coordination during load-bearing tasks.