M2M-InvNet: Human Motor Cortex Mapping From Multi-Muscle Response Using TMS and Generative 3D Convolutional Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Transcranial magnetic stimulation (TMS) targets the motor cortex to elicit motor evoked potentials (MEPs) in muscles.
• The precise location of TMS application influences the resulting MEPs, indicating a causal link between stimulation site and neural activation.

Purpose of the Study:

• To investigate if motor evoked potentials (MEPs) can be used to infer the stimulated regions on the motor cortex.
• To develop and evaluate deep learning models for this inverse imaging task, aiming to guide TMS coil placement for desired muscle responses.

Main Methods:

• Leveraged a previously developed 3D convolutional neural network (CNN) for predicting MEPs from electric fields.
• Developed and assessed five distinct inverse imaging CNN architectures, including conventional and generative models.
• Evaluated model performance using multiple reconstruction accuracy metrics.

Main Results:

• One proposed architecture, M2M-InvNet, demonstrated superior performance in estimating stimulated cortical regions from MEP data.
• The study successfully tackled the inverse imaging problem, mapping MEPs back to their cortical origins.

Conclusions:

• The M2M-InvNet architecture shows significant promise for accurately identifying TMS-stimulated areas on the motor cortex based on MEPs.
• This approach could enable more precise TMS coil positioning to achieve specific patterns of muscle activation.