Predicting RF-Induced Heating for Deep Brain Stimulator System Using an Artificial Neural Network

Different deep brain stimulators (DBSs) implanted along different trajectories can lead to significant differences in radiofrequency (RF)-induced heating. Accurate prediction of the RF-induced heating for a certain implant trajectory would help to mitigate the RF-induced heating hazards. Based on typical implantation trajectories of DBSs, an artificial neural network (ANN) model is developed and applied to predict the RF-induced heating for the DBSs at 3T MRI. A dataset of 19800 DBS exposure samples was generated based on a) a generic RF birdcage coil, b) a detailed anatomical male model, c) thousands of DBS lead trajectories, and d) four artificial transfer function (TF) models. An ANN regression model with 8 fully connected layers is developed to predict the RF-induced heating. With a well-designed ANN, the RF-induced heating for the DBS system could be efficiently and accurately predicted. The mean absolute error (MAE) performance of the ANN regression model was less than 0.81 °C, and the coefficient of determination (R2) value was 0.997.

I. Introduction

Deep brain stimulator (DBS) implants are used for patients with Parkinson's disease, essential tremor, dystonia, and other neurological disorders [1]. Due to the RF-induced heating hazards associated with the DBS implants, however, these patients may have limited access to magnetic resonance imaging (MRI). To ensure patient safety, restrictions/limits based on the whole-body averaged specific absorption rate (SAR) and/or B₁⁺ are often required [2]–[4].