Efficient, Robust, and Accurate CNN Predictor for Neuronal Activation in Directional Deep Brain Stimulation

Area of Science:

• Neuroscience
• Biomedical Engineering
• Artificial Intelligence

Background:

• Deep brain stimulation (DBS) programming involves complex parameter adjustments.
• Manual DBS programming is increasingly impractical with advanced electrode designs.
• Current volume of tissue activated (VTA) calculations are time-consuming.

Purpose of the Study:

• To develop a faster and more accurate method for VTA calculation in DBS.
• To replace traditional, slow axonal modeling with a machine learning approach.
• To improve the efficiency and robustness of DBS programming.

Main Methods:

• Utilized finite element models (FEM) to simulate electric fields from DBS systems.
• Generated a dataset of axonal responses using multicompartment cable models.
• Trained a convolutional neural network (CNN) to predict neural activation thresholds.

Main Results:

• The CNN model achieved high accuracy in predicting nerve fiber activation thresholds (MAE of 0.032V).
• CNN demonstrated superior stability and accuracy compared to existing activation function (AF) methods.
• Computation time was reduced by five orders of magnitude versus standard methods.

Conclusions:

• CNN-based neural fiber prediction offers a quick, accurate, and robust solution for DBS programming.
• This AI-driven approach enhances the efficiency of VTA calculation.
• The method shows significant potential for clinical applicability in DBS parameter selection.