Real-time estimation of the optimal coil placement in transcranial magnetic stimulation using multi-task deep...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Transcranial magnetic stimulation (TMS) is a key neuromodulation technique requiring precise coil placement for effective therapeutic and diagnostic outcomes.
• Current computational models for optimizing TMS coil positioning are accurate but computationally intensive, limiting real-time application.
• Accurate coil positioning is crucial for focal stimulation and maximizing electric fields at the cortical target.

Purpose of the Study:

• To develop a rapid, accurate computational framework for optimizing TMS coil placement and predicting induced electric fields.
• To create a neural surrogate model capable of real-time coil optimization, overcoming the limitations of traditional numerical simulations.
• To validate the developed model in both healthy subjects and glioblastoma patients, considering tumor heterogeneity.

Main Methods:

• Development of a novel multi-task deep neural network (Attention U-Net based) trained on extensive numerical optimization data.
• Simultaneous prediction of optimal coil placement and TMS-induced electric fields.
• Validation of the model's accuracy and speed against state-of-the-art numerical methods and in clinical patient data.

Main Results:

• The neural surrogate achieved coil optimizations in 35 ms, a significant speed improvement over existing methods.
• Position estimates showed mean errors below 2 mm, surpassing manual coil positioning accuracy.
• Predicted electric fields demonstrated high correlation (r > 0.97) with numerical references.
• The model performed accurately in glioblastoma patients, highlighting the importance of heterogeneous tumor conductivities.

Conclusions:

• The developed deep neural network offers a computationally efficient and accurate solution for real-time TMS coil optimization.
• This approach facilitates subject-specific and potentially real-time electric field-optimized TMS applications.
• Findings support the use of realistic conductivity models and demonstrate the model's applicability in complex patient cases.