Using generative deep learning to shorten calibration time in ultra-high field MRI

7th February 2022

Boris Eberhardt, Benedikt A. Poser, N. Jon Shah and Jörg Felder

Magnetic resonance imaging at ultra-high field (UHF) is hugely beneficial due to its higher signal-to-noise ratio (SNR) and increased spatial resolution. However, these benefits come at a cost, as increased field strengths lead to increased RF inhomogeneities and higher energy deposition with a correspondingly increased specific absorption rate (SAR) burden.

Spoke trajectory parallel transmit (pTX) excitation can be used to mitigate some of these issues, and to this end, current RF excitation pulse design algorithms either employ the acquisition of field maps with subsequent non-linear optimisation or a universal approach applying robust pre-computed pulses.

In this research, an intermediate method that uses a subset of acquired field maps combined with generative machine learning models to reduce the pulse calibration time while offering more tailored excitation than robust pulses (RP) is presented and evaluated.

The possibility of employing image-to-image translation and semantic image synthesis machine learning models based on generative adversarial networks (GANs) to deduce the missing field maps is examined. Additionally, an RF pulse design that employs a predictive machine learning model to find solutions for the non-linear (two-spokes) pulse design problem is investigated. As a proof of concept, simulation results obtained with the suggested machine learning approaches that were trained on a limited data-set, acquired in vivo, are presented.

The results show that the achieved excitation homogeneity based on a subset of half of the B+1maps acquired in the calibration scans and half of the B+1maps synthesised with GANs is comparable with state-of-the-art pulse design methods when using the full set of calibration data while halving the total calibration time. By employing RP dictionaries or machine-learning RF pulse predictions, the total calibration time can be reduced significantly as these methods take only seconds or milliseconds per slice, respectively.

By substituting missing data with synthesised data generated by machine learning models, it is possible to save time on the initial calibration and to either optimise RF pulses or make improved choices when pre-calculated pulses are used. In this way, it is anticipated that deep learning will aid more streamlined clinical throughput in the future. 

Original publication: B1 field map synthesis with generative deep learning used in the design of parallel-transmit RF pulses for ultra-high field MRI