Cortical surface electric field estimation for real-time TMS with graph neural networks

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Transcranial magnetic stimulation (TMS) is a non-invasive technique used for neurostimulation and neuromodulation.
• Accurate estimation of electric fields (E-fields) induced by TMS is crucial for effective and safe clinical applications.
• Current methods for E-field estimation are computationally intensive and require extensive preprocessing.

Purpose of the Study:

• To develop a real-time method for estimating E-fields on the cortical surface using 3D head MR images.
• To bypass the need for constructing a 3D anatomical head model for E-field estimation.
• To improve the efficiency and accuracy of E-field prediction in TMS.

Main Methods:

• A deep learning-based cortical surface reconstruction method was employed to generate a cortical mesh.
• A graph neural network (GNN) with a 2D cortical surface mesh topology was utilized for E-field estimation.
• A U-Net architecture was integrated with the GNN to extract multiscale features from volumetric MR images, enabling efficient computation by restricting estimation to a 2D surface.

Main Results:

• The proposed method achieved E-field estimation in 29 ms per coil configuration, significantly faster than existing methods.
• The method demonstrated higher accuracy in E-field estimation compared to conventional voxel-wise approaches.
• Real-time E-field estimation was achieved without the need for 3D anatomical model construction or extensive preprocessing like tissue segmentation.

Conclusions:

• The developed method enables real-time, accurate E-field estimation on the cortical surface for TMS.
• This technique eliminates the need for complex 3D modeling and preprocessing, making it clinically practical.
• Real-time E-field estimation can aid in optimizing coil placement and ensuring precise stimulation of target brain regions.