A type of data-acquisition sequence in magnetic resonance imaging has been developed that rapidly and robustly quantifies properties of imaged tissue by elucidating a characteristic signal fingerprint. See Article p.187
A conventional magnetic resonance imaging (MRI) scan is designed to create contrast between tissue types. It does so on the basis of various physical properties that are related to the microenvironment of water molecules in the tissues. The resulting images are typically qualitative and merely represent a contrast that is weighted towards one tissue property or another. Although such qualitative images certainly contain clinically useful information, MRI scientists continue to search for methods that can efficiently quantify tissue properties. On page 187 of this issue, Ma et al.1 present a kind of MRI data-acquisition sequence and image-formation algorithm that avoids the standard approach of acquiring an image weighted towards a single tissue property. This innovative method is based on the collection of magnetic resonance 'fingerprints' that reflect the combined effect of multiple underlying tissue properties. The approach has the potential to make quantitative mapping of tissue properties feasible in the clinical setting and to lead to improved diagnoses.
The pattern of curves on the human fingertip does not directly encode specific information such as an individual's name or sex, but its uniqueness does provide a way to recognize an individual and so link it to known attributes for that person. In other contexts, the test fingerprint can be used to describe any unique and recognizable signal that can be linked back to a specific set of properties. Conventional MRI methods that create quantitative maps of tissue properties are time consuming to collect and susceptible to corruption if the subject moves. The magnetic resonance fingerprinting approach described by Ma et al. is both fast and resilient to inconsistencies in the acquired data. The strength of the method comes from the use of a specialized data-acquisition sequence that elucidates a characteristic signal, or fingerprint, for a given spatial tissue location. That signal can then be compared with a set of precalculated fingerprint signals that are associated with known tissue properties. Once the fingerprint for a given spatial location has been recognized, that same location can be assigned the associated tissue properties in quantitative maps (Fig. 1).
In medical imaging, a scanner collects signals of emitted or transmitted energy after some type of input energy — such as that of X-ray photons, ultrasonic waves and radio-frequency waves — interacts with the object being imaged, for example the organs and tissues of the human body. In general, the physical properties of the tissue and the microenvironment surrounding it determine the intensity of the collected signals. The signals also depend on the selection of data-acquisition parameters made by the scanner operator. To form an image with an MRI scanner, a powerful magnet is used to generate a strong magnetic field that affects the magnetic spins of protons in the water molecules prevalent throughout the body. A small, but detectable, majority of the magnetic spins align themselves with the strong magnetic field. The spins precess (rotate) about the strong magnetic field at a characteristic frequency known as the Larmor frequency2, much like a gyroscope precesses about the force vector created by gravity. The precessing magnetic spins can be excited from an equilibrium energy state to a state of higher energy by using a radio-frequency wave that has a frequency matching the Larmor frequency. Once the input radio-frequency energy is deactivated, the excited spins return to the equilibrium energy state and emit radio-frequency energy, which can be detected to form an image.
The intensity of the emitted radio-frequency energy is a function of several tissue properties, including the proton density (PD) at a given spatial location and how quickly the excited spins return (relax) to the equilibrium energy state. The rate of relaxation is described by two time constants3 (T1 and T2) that reflect relaxation in the geometric planes respectively parallel and perpendicular to the strong magnetic field. The signal intensity for a given spatial location in almost all magnetic resonance images can be defined as a function of PD, T1, T2 and a few data-acquisition sequence parameters. Obtaining maps of PD, T1 and T2 is important in a clinical setting because pathological changes to tissue often alter one or more of the values for these three parameters. The importance of these tissue properties is apparent when one considers that most conventional clinical magnetic resonance images are weighted towards PD, T1 or T2 to reveal pathological conditions.
Since the introduction of MRI in the early 1970s4, magnetic resonance physicists and engineers have continued to develop approaches to show information about the tissues and organs being scanned. The clinical utility of rapidly and robustly mapping PD, T1 and T2 makes the technique described by Ma et al. extremely valuable. The authors use a customized version of a data-acquisition protocol known as an inversion-recovery balanced steady-state free-precession (IR-bSSFP) sequence that is sensitive to all the tissue properties of interest5. Unlike the conventional approach of keeping sequence parameters constant throughout the repeated data readouts of the IR-bSSFP protocol, the authors pseudorandomly varied the amplitude of radio-frequency excitation and time spacing between repeated excitations to cause the collected MRI signal at a given spatial location to behave in a pseudorandom manner. However, the pseudorandom MRI-signal response pattern is actually a function of PD, T1 and T2 at that spatial location. The authors refer to this signal pattern as the magnetic resonance fingerprint for a particular combination of tissue properties.
Ma and colleagues present compelling evidence for the success of the strategy using 'imaging phantoms' with known properties and for in vivo brain images. The method is robust in the presence of bulk motion of the object being imaged and in the case of data undersampling. The reduction of acquired data markedly decreases scan time by a factor of nearly 50-fold compared with the time of the fully sampled scan. The approach should be adaptable to other MRI data-acquisition sequences, and could be extended to include the effects of other tissue and microenvironment properties that are relevant to the adapted MRI sequence types. Perhaps the most exciting possible future application of the method is the collection of magnetic resonance fingerprints in patients with confirmed pathology, such as brain tumours, that can be difficult to grade and classify. A library of fingerprints that are correlated to known pathology could be used to improve diagnosis for such patients.
Ma, D. et al. Nature 495, 187–192 (2013).
Kellogg, J., Rabi, I., Ramsey, N. & Zacharias, J. Phys. Rev. 56, 728–743 (1939).
Levitt, M. H. Spin Dynamics 2nd edn (Wiley, 2008).
Lauterbur, P. C. Nature 242, 190–191 (1973).
Schmitt, P. et al. Magn. Reson. Med. 55, 177–186 (2006).