A machine learning-based approach for RF transfer function modeling of active implantable medical electrodes at 3T MRI

Abstract

Abstract Objective. The objective of this work is to propose a machine learning-based approach to rapidly and efficiently model the radiofrequency (RF) transfer function of active implantable medical (AIM) electrodes, and to overcome the limitations and drawbacks of traditional measurement methods when applied to heterogeneous tissue environments. Approach. AIM electrodes with different geometries and proximate tissue distributions were considered, and their RF transfer functions were modeled numerically. Machine learning algorithms were developed and trained with the simulated transfer function datasets for homogeneous and heterogeneous tissue distributions. The performance of the method was analyzed statistically and validated experimentally and numerically. A comprehensive uncertainty analysis was performed and uncertainty budgets were derived. Main results. The proposed method is able to predict the RF transfer function of AIM electrodes under different tissue distributions, with mean correlation coefficients r of 0.99 and 0.98 for homogeneous and heterogeneous environments, respectively. The results were successfully validated by experimental measurements (e.g. the uncertainty of less than 0.9 dB) and numerical simulation (e.g. transfer function uncertainty <1.6 dB and power deposition uncertainty <1.9 dB). Up to 1.3 dB in vivo power deposition underestimation was observed near generic pacemakers when using a simplified homogeneous tissue model. Significance. Provide an efficient alternative of transfer function modeling, which allows a more realistic tissue distribution and the potential underestimation of in vivo RF-induced power deposition near the AIM electrode can be reduced.