MRI RF-Induced Heating Prediction of Complex-Shaped Passive Implantable Medical Devices Using Mesh-Based Convolutional Neural Network

In this article, a convolutional neural network (CNN) model is proposed to predict the in vitro radio frequency (RF)-induced heating of complex-shaped passive implantable medical devices (PIMDs) under magnetic resonance imaging (MRI). The electromagnetic (EM) simulation meshes and incident electric field on the mesh grids are used as the input of the CNN model while the network output corresponds to the RF-induced 1-g specific absorption rate (SAR). A convergence analysis is performed to understand the effectiveness of the method. A discussion on selecting the training dataset size is presented using a principal component analysis (PCA) algorithm. Results demonstrate the robustness of the CNN model for the prediction of RF-induced heating from complex-shaped PIMDs under MR exposure.