MRI Safety Assessment of Implantable Medical Devices using Neural Networks



Journal Title

Journal ISSN

Volume Title



Magnetic resonance imaging (MRI) is one of the most effective and widely used noninvasive imaging techniques for disease diagnosis, due to its superior performance in soft tissue imaging without harmful ionizing radiation. However, the radiofrequency (RF)-induced heating which causes temperature rises and tissue burns, is a major hazard for patients with implantable medical devices to have an MRI. Recently, the RF-induced heating for passive implantable medical devices (PIMDs) has been carefully assessed to clarify the safety conditions for MRI examination in standard and fully controlled environments. However, these assessment needs costly measurements or numerical simulations which can take a relatively long time. Therefore, it is not applicable in providing the RF-induced heating of all available configurations for diverse configurations, for fast predicting the RF-induced heating in the design stage, or estimating the potential risks for patients with unlabeled implantable devices in emergency situations, etc. It is necessary to provide a fast prediction method of RF-induced heating in standard and fully controlled conditions or environments for different kinds of implantable medical devices. Numerical modeling and simulations are conducted to study the RF-induced heating for general PIMDs, such as commonly used plate systems, and external fixators, in a 1.5 T or 3 T magnetic resonance (MR) environment. RF-induced heating for various configurations of the implantable medical devices in the phantom that covers possible clinical scenarios is investigated to be the ground truth data. Then, the neural networks (NNs) can be used as the surrogate model to train and predict the RF-induced heating against various configurations for different kinds of devices. To get accurate prediction performance, different architectures of NNs are applied to predict the RF-induced heating of the implantable medical devices based on numerical or measure results. To validate the NNs, part of the ground truth data was used for training, while the rest were used to test the performance. Once the NNs had been trained, the possible hazard of the new implantable medical devices with predefined configurations would be clarified.



MRI Safety, Neural Networks