Mapping Motor Cortex Stimulation to Muscle Responses: A Deep Neural Network Modeling Approach.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Understanding coordinated motor control is crucial for basic science and clinical applications like stroke rehabilitation.
• Modeling the relationship between brain stimulation and muscle response can elucidate neurological injury mechanisms and inform therapeutic interventions.

Purpose of the Study:

• To explore and recommend an optimal deep neural network (DNN) model for mapping transcranial magnetic stimulation (TMS) of the motor cortex to muscle responses (M2M-Net).
• To analyze the trade-off between model complexity and performance for enhanced understanding of motor control.

Main Methods:

• Utilized a combination of finite element simulation, empirical neural response profiles, a convolutional autoencoder, a deep network mapper, and multi-muscle activation recordings.
• Investigated various DNN architectures and employed information criteria for comparative performance analysis.

Main Results:

• Identified a specific DNN model architecture that minimizes squared errors for the M2M-Net.
• The optimal model effectively maps motor cortex stimulation to a combination of direct and synergistic muscle connections.

Conclusions:

• The developed M2M-Net provides a reliable method for modeling muscle responses from brain stimulation, advancing the study of motor control.
• The findings support the use of DNNs in understanding neurological disorders and developing targeted neurostimulation therapies.