Recent developments in pulse sequence design, signal processing techniques, model-based reconstructions, and improved MR hardware and computing capabilities have revitalized the field of quantitative MRI. Magnetic Resonance Fingerprinting is one example of a technology that has emerged out of these advances, and one that could change the way the field thinks about the nature of signal encoding and information in our data. It is important to remain cognizant of the challenges to clinical adoption of such quantitative MR techniques and for scientists to work closely with physicians to ensure that they are implemented and used appropriately. However, the high level of interest within the MR community, coupled with the strength of these new technologies, may pave the way for these techniques to become essential components of the clinical workflow for tissue characterization in the near future.
The collection of quantitative information using MRI is potentially beneficial in many clinical situations, as it may enable earlier and more definitive diagnosis of disease. Indeed, the generation of numerical values from MR images, including but certainly not limited to information such as lesion size, perfusion parameters, ejection fraction, apparent diffusion coefficient, and flow, has become more commonplace as MR imaging has matured. However, most clinical MR images are still only qualitatively weighted by underlying tissue properties such as T1, T2, or diffusion. The resulting grayscale pixel values are only meaningful in relation to one another, making comparisons between different subjects, the same subject imaged on different scanners, or even the same subject imaged on the same scanner on different days nearly impossible. Clinicians have continued to rely on these weighted images because the benefits of collecting quantitative tissue property maps have not outweighed the costs, which include lengthy scan times and remaining inconsistencies even in so-called “quantitative parameter maps” due to acquisition or model errors.
However, in recent years a number of advances have occurred which bring the promise of rapid quantitative MR imaging closer to clinical reality. This talk will focus on advances in relaxometry, although similar techniques have been applied to quantify perfusion, fat fraction, and microstructural information. These advances rely first of all on efficient pulse sequences (i.e. Look-Locker [1]), which can be used to collect data more rapidly than conventional techniques as in DESPOT [2] or MOLLI [3]. Instead of attempting to force the signal evolution to be sensitive only to one main underlying parameter, other types of pulse sequences such as inversion recovery bSSFP [4] take advantage of the sensitivity of the sequence to multiple tissue properties [5]. In this way, several tissue properties can potentially be accessed simultaneously. At the same time, new rapid imaging technologies, including advanced parallel imaging techniques and sparse reconstruction methods, have been emerging ([6-10] are just a few examples). Although the idea of using model-based reconstructions to constrain signal time courses is not a recent development, the application of these methods to relaxometry (as in [11]) in conjunction with efficient pulse sequences and new image reconstruction strategies has led to the reinvigoration of relaxation parameter mapping. Finally, although non-Cartesian k-space sampling trajectories such as spiral or radial have been employed in research for decades, improvements to MRI hardware and advances in computational speed have gradually encouraged the use these data collection strategies in practice.
Magnetic Resonance Fingerprinting (MRF) is one example of the new techniques which leverage these developments for quantitative MRI. The ultimate goal of MRF is to move from collecting weighted images to instead collect data sensitive to tissue properties, out of which quantitative maps can be extracted. The first demonstration of MRF showed that this method can be employed to simultaneously map T1, T2, and proton density values in a single short (<10s) scan [12]. In MRF, the pulse sequence parameters (typically the flip angles and TRs) are varied during the scan, which prevents the magnetization from reaching a steady state. Tissues with different underlying T1 and T2 values will exhibit signal evolutions that are different from one another. These signal evolutions are complex and difficult to fit to a simple signal model, such as an exponential recovery or decay, as would be done in conventional parameter mapping. Instead, the experimental signal evolution in each pixel is matched to a dictionary of possible signal evolutions calculated using the Bloch equations, knowledge of the pulse sequence, and an input range of T1 and T2 values. While it would be possible to perform tissue property mapping in this way using fully-sampled data, the acquisition time would be long (several minutes for a 2D scan). In MRF, a highly undersampled spiral trajectory is employed to dramatically reduce the scan time. As long as the aliasing artifacts produced by the undersampling are incoherent, pattern matching can still be used to extract the underlying T1 and T2 values despite the additional interference from aliasing. Note that as mentioned above, MRF is made possible through the use of rapid pulse sequences which can sensitize the magnetization to multiple tissue properties, a model-based non-linear reconstruction which relies on sparsity, reliable gradient systems and fast computer processing. MRF has been used to generate T1 and T2 maps in the brain [12,13], prostate [14], body [15], and heart [16]. While originally used to map relaxation values, MRF can be employed to map any tissue property to which a sequence is sensitive, including diffusion [17], perfusion [18,19], partial volume effects [20], and microvascular parameters such as blood oxygenation [21].
While MRF and other new quantitative MRI techniques show promise for rapid tissue property mapping, practical hurdles to ubiquitous clinical use remain. A number of groups have been founded to help tackle these challenges, for instance the RNSA Quantitative Imaging Biomarker Alliance (QIBA) and the new ISMRM Quantitative MR Study Group. Firstly, these methods must be vigorously tested for stability across different vendors, in different environments, and over long periods of time. The need for globally accepted references has led the National Institute for Standards in Technology (NIST) to develop new phantoms which can be used by different groups to test their implementations and validate their results. Such testing is essential before a technique can be claimed to be quantitative; for instance, T1 and T2 values measured using MRF have been found to be precise on several different scanners using this phantom [22]. Once a quantitative imaging method has been found to yield reproducible numbers, a bigger challenge remains: to demonstrate that the parametric maps have real clinical value that justifies their use. The value can take many forms (improved diagnostic accuracy, scan time savings, etc), but it must be clearly shown for specific applications. This demand goes beyond an understanding of “normal” values, as well as an idea of the range of “abnormal” values. Real life, important clinical questions must be asked in a way where the quantitative imaging can actually provide a meaningful answer – either by objectively depicting that information which must currently be inferred from the images, or providing reliable information that is not available any other way. Finally, acceptance by radiologists and other physicians is crucial, as without their buy-in these techniques will never be adopted. User-friendly reconstructions in clinically acceptable time frames are essential, and the ability to generate weighted images from the parametric maps may be useful while physicians acclimate to the new appearance of these maps.
In conclusion, recent developments in pulse sequence design, signal processing techniques, model-based reconstructions, and improved MR hardware and computing capabilities have revitalized quantitative MRI. MRF is one example of a technology that has emerged out of these advances, and one that could change the way the field thinks about the nature of signal encoding and information in our data. It is important to remain cognizant of the challenges to clinical adoption of such quantitative MR techniques and for scientists to work closely with physicians to ensure that they are implemented and used appropriately. However, the high level of interest within the MR community, coupled with the strength of these new technologies, may pave the way for these techniques to become essential components of the clinical workflow for tissue characterization in the near future.