Development of novel parallel transmit RF pulse design algorithms for ultra-high field magnetic resonance imaging

Eberhardt, Boris; Shah, Nadim Joni (Thesis advisor); Stahl, Achim (Thesis advisor)

Aachen : RWTH Aachen University (2022)
Dissertation / PhD Thesis

Dissertation, RWTH Aachen University, 2022


Magnetic resonance imaging (MRI) enables the non-invasive investigation of both brain anatomy and function. While clinical MRI scanners operate mostly at static fields B0 of 3T or below, current research is focused on higher fields of 7T or above due to the improvements in the achievable signal-to-noise ratio (SNR). The resonance frequency of the time-dependent magnetic excitation fields B1(t) increases linearly with the static magnetic field. Within the considered ultra-high field (UHF) region of the radio-frequency (RF) band, a shortening of the RF wavelengths inside biological tissue can be observed with the consequence of inhomogeneous excitation patterns and even signal voids due to complex interferences. Furthermore, the existence of concomitant electric fields inside the conductive tissue results in an increased power deposition that can cause dangerous elevations in temperature. A proposed solution to these problems is to employ multiple RF transmitters that are driven in parallel and independently. The new degrees of freedom enabled by this parallel transmission (pTx) can be utilized to control the magnitude and phase of each individual transmit channel, with the goal to shape the resulting excitation field superposition in a way that simultaneously compensates for inhomogeneous excitation patterns and increased power deposition. It is investigated in this work both with simulations and in vivo measurements at a B0 field of 9.4T how state of the art slice-selective pTx RF pulse algorithms can be improved by making use of all degrees of freedom that were not considered in previously published methods. This includes the optimization of the magnetic field gradients that are played out during the RF excitation phase, and the search for optimal complex pTx RF scaling weights. Nature-inspired evolution strategies are applied to find robust solutions that improve upon previously published methods. A multi-step algorithm is developed that includes both safety-relevant and hardware constraints. The investigation was carried out with clinical applicability in mind, and thus a fast running time of only a few seconds per slice with only minimal computational resources was achieved. Any pTx RF pulse algorithm still requires an initial subject-specific calibration procedure that involves the measurements of all B1 and B0 fields. For the employed 16 channel pTx coil array this can take up to five minutes, and even longer calibration times have been reported previously. Such long calibration times with subsequent pulse calculations could be an impediment for clinical adoption. In this work, generative machine learning models are utilized to generate missing field maps from only minimal seed data. This synthesized data can be used to calculate pTx RF pulses. Additionally, multiple machine-learning models have been investigated to predict RF pulses with the aim to reduce the pulse calculation time. The motivation behind these methods is to reduce complexity and save time by automation, and thus to increase their potential applicability.