Clinical applications of the kT-points method to homogenise spin excitation in 3T MRI
Raphaël Tomi-Tricot
2018-09-26
Abstract:High field magnetic resonance imaging (MRI) is often used in clinical practice, for the high signal-to-noise ratio it offers.However, at high field, the radiofrequency (RF) wavelength used for imaging is shorter, which can induce loss of signal and contrast when it is close to or shorter than the dimensions of the irradiated objects. At 3T, RF wavelength is about 30 cm in human tissues,and such artefacts are frequently observed in the abdomen, as well as in the thighs or in the breasts. This is certainly one of the main reasons why high field MRI has failed to establish itself as the gold standard in hospital, where 1.5T scanners are more frequent.Recent 3T scanners usually come with a two-RF-channel parallel transmission setup: in principle, each channel can transmit completely independent waveforms. However, this technology is not exploited fully in practice, as only the static RF shimming is implemented: a single waveform is used, with adjusted amplitude and phase on each channel. This allows to create interference patterns, calculated to counteract transmission inhomogeneities measured beforehand in the patient (RF and possibly static field).This method works best when many channels are available, but shows its limits when good homogeneity is expected over a large field of view. The kT-points method, developed at CEA-NeuroSpin for brain imaging at ultra-high field (7T) relies on a succession of short rectangular RF pulses interleaved with static gradient “blips” to modulate magnetisation at will, thus producing homogeneous excitation in spite of an imperfect RF field. Those composite pulses are even more effective as they can take advantage of parallel transmission (eight channels at 7T). In this thesis, the kT-points technique is applied at 3T. The objective is to demonstrate its usefulness and its compatibility with a clinical routine workflow. First, several changes made to the kT-points pulse design algorithm and to static field mapping in order to adapt them to body imaging (liver, breasts) are described. Indeed, the presence of lungs and fat add further constraints compared to the brain. Then, several clinical studies are detailed. The first one regards T₂-weighted breast MRI on a single-channel scanner. It shows that in that case the default transmit mode is satisfactory,and only slightly improved by kT-points. A second study focuses on T₁-weighted dynamic contrast-enhanced imaging of the liver,with two transmit channels. Static RF shimming and kT-points were compared on a large cohort. For some “difficult” patients,acquisitions were quantitatively and qualitatively better with kTpoints,which therefore offer a more uniform diagnostic quality among the population. Finally, a novel method is proposed:SmartPulse. It is based on the universal pulse concept, developed in NeuroSpin for brain imaging, whose principle is to design pulses (e.g. kT-points) for a given application, that homogenise excitation in the whole population, and not only for one subject.Thus, there is no more need for cumbersome mapping and inline pulse design. SmartPulse extends the range of universal pulses to body imaging, by adequately clustering the population, designing different pulses for each cluster, and applying a machine learning classifier to assign the most appropriate pulse to any new subject.Proof of concept was undertaken in abdominal imaging, whereinter-subject variability is considerable. We hope this thesis will give a new outlook on RF inhomogeneity handling in routine 3T MRI, and in the long run will help making body imaging moreaccessible at high and ultra-high field.
Computer Science,Engineering,Medicine,Physics