DeeptDCS: Deep Learning-Based Estimation of Currents Induced During Transcranial Direct Current Stimulation.

Area of Science:

• Neuroscience
• Computational Biology
• Artificial Intelligence

Background:

• Transcranial direct current stimulation (tDCS) is a non-invasive brain stimulation method.
• Accurate real-time evaluation of tDCS-induced current density is crucial for research and clinical applications.
• Current simulation methods are computationally intensive, limiting rapid analysis.

Purpose of the Study:

• To develop a deep learning-based emulator, DeeptDCS, for rapid, near real-time evaluation of tDCS-induced current density.
• To incorporate complex electrode configurations into the simulation without increasing input dimensionality.
• To enhance the accuracy and generalizability of tDCS current density modeling.

Main Methods:

• An Attention U-net architecture was employed as the core of the DeeptDCS emulator.
• Volume conductor models (VCMs) of head tissues were used as input, with electrode configurations integrated directly.
• The model was trained and tested using VCMs and various electrode parameters.

Main Results:

• DeeptDCS achieved higher accuracy compared to standard U-net and its variants.
• The emulator demonstrated strong generalization capabilities to non-trained electrode configurations after fine-tuning.
• Computational time for emulation was reduced to a fraction of a second, significantly faster than physics-based simulators.

Conclusions:

• DeeptDCS offers a computationally efficient and accurate solution for emulating tDCS-induced current density.
• The speed of DeeptDCS enables its use in time-intensive applications like uncertainty quantification and optimization studies.
• This deep learning approach accelerates the evaluation of tDCS effects, advancing brain stimulation research.