In-vivoverified anatomically aware deep learning for real-time electric field simulation

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging

Background:

• Transcranial magnetic stimulation (TMS) is a key non-invasive brain modulation technique for mental health disorders.
• Accurate simulation of electric fields (E-fields) is crucial for effective, targeted TMS.
• Current E-field simulation methods are time-consuming and complex, hindering rapid coil placement optimization.

Purpose of the Study:

• To develop an efficient and accurate deep learning model for simulating TMS-induced E-fields.
• To improve the speed and precision of determining optimal TMS coil placements.
• To facilitate personalized TMS treatment strategies.

Main Methods:

• An attentional deep learning network was designed to simulate E-fields.
• The network utilizes individual MRI scans and coil configurations as input.
• It transforms MRI data into explicit brain tissues to generate local E-field distributions.

Main Results:

• The deep learning method reduced mean relative error in gray matter E-field strength by 21.1% compared to prior deep learning approaches.
• Correlation between regional E-field strengths and electrophysiological responses increased by 35.0%.
• In-vivo TMS experiments showed comparable stimulation performance to traditional computational methods, reducing optimization time from hours to minutes.

Conclusions:

• The developed deep learning network offers a precise and efficient method for simulating TMS E-fields.
• This approach significantly accelerates the optimization of TMS coil placement.
• The method shows strong potential for enabling individualized coil placements in clinical settings for personalized TMS therapy.