Fast prediction of MRI RF-induced heating for implantable plate devices using neural network

In this paper, a fast prediction method using a 3-layer feed-forward neural network was proposed to estimate the radiofrequency (RF) - induced heating for generic implantable plate devices under magnetic resonance imaging (MRI). The peak local specific absorption rate averaged over 1 gram (SAR1g) was calculated to represent the heating level. A total number of 130 plate devices have been numerically studied with various length, width, thickness, number of screws, length of screws, etc. Among the 130 devices, 91 were used to train the neural network, while 39 were utilized to examine the validity of the SAR1g values predicted by the neural network. The results have shown that the correlation between the estimation and the target values was larger than 0.995 and the mean square error was less than 2.9 W/kg with a mean peak SAR of 93.3 W/kg.