Predicting MRI RF Exposure for Complex-shaped Medical Implants Using Artificial Neural Network

In this paper, a fast prediction method using a 3-layer feed-forward artificial neural network (ANN) based on complex shape descriptors was proposed to estimate the radio-frequency (RF) exposure for complex-shaped implantable devices under magnetic resonance imaging (MRI). The inputs of the ANN consist of 20 parameters based on dimensions and surface-based features of the complex-shaped devices, and the output is the peak MRI RF exposure, in terms of 1-gram averaged specific absorption rate (SAR1g), for the device. This method is implemented and validated with 408 complex-shaped trauma implantable devices. These accurate 3D CAD models are used to numerically calculate the peak SAR1g of the devices by using a full-wave electromagnetic solver based on finite-difference time-domain (FDTD) method. Among the 408 devices, 326 were used to train the neural network, while 82 were utilized to examine the validity of the SAR1g values predicted by the ANN. The results have shown that the correlation between the estimation and the target values was larger than 0.993, and the root mean square error was less than 6.21 W/kg.