MR Fingerprinting: Introduction
Martijn A Cloos1

1NYU School of Medicine, United States


Magnetic Resonance Fingerprinting (MRF) is an exciting new framework to rapidly acquire simultaneous quantification of multiple tissue properties, but what is it that distinguishes MRF from other quantitative MR techniques?


Over the years, many quantitative MR techniques have been proposed [1-7 and many others]. Each of which strives to strike an optimal balance between acquisition speed, model simplicity, accuracy, and precision [7-10]. These methods, almost universally, prioritize signal localization over the characterization of spin dynamics. The DESPOT technique [5], for example, typically considers only three distinct states of the magnetization, each of which is sampled hundreds of times to form three complete images. The longitudinal relaxation time (T1) and transverse relaxation time (T2) are then calculated from the ratio between the images. MRF [11], on the other hand, emphasizes the spin-dynamics, and sacrifice much in terms of signal localization to rapidly sample hundreds of slightly different states [12].

Seeing through the artifacts

A typical MRF sequence may only capture about ~1/50th of k-space per measured state of the magnetization [11,13,14,15]. Such sacrifices in signal localization inevitably lead to under-sampling artifacts. However, it is possible to see through these artifacts, provided that they are sufficiently incoherent from one another.

Suppose, for a moment, that we somehow captured a fully sampled image for each measured state of the magnetization in the MRF sequence. We could then follow how the signal evolves in each voxel as the sequence perturbs the spin-system from one state to the next. This evolution is what is called an MR “fingerprint”. Just like the whorls on your finger, these fingerprints contain unique features that enable us to identify the “owner”. In this case of an MRF exam, the measured fingerprints are compared to the entrees in a dictionary of simulated fingerprints with known properties. If the simulations are representative of the actual experiment, the best match will identify the underlying tissue properties in the voxel.

With this in mind, let us now return to the undersampled case. If the under-sampling artifacts are spatiotemporally incoherent (each image contains different under sampling artifacts), they will add a noise like component to the measured fingerprint. Such rapidly varying noise like features can easily be filtered out by the dictionary matching process. Intuitively, one could visualize the idea using a sliding window (low pass filter) to smooth out the rapid noise like fluctuations produced by incoherent artifacts (and applying the same filter to the dictionary) [14,16,17]. This is not the full story, however, and may not work well in all cases. A more general analysis can be made using a principle component analysis of the [18,19].

Noise and pseudo noise

In a way we may be able to say that truly incoherent undersampling artifacts add a noise like component to each fingerprint. This “pseudo noise” comes on top of the thermal-noise from the sample and electronics. In traditionally imaging techniques, the SNR is expected to increase with the square root of the number of measurements (signal averaging). If we repeat the exact same MRF measurement twice, we may expect to see a similar behavior (provided that the pseudo-noise is not the dominant factor). However, if instead we acquire complementary k-space samples in each measurement (while making sure to retain incoherence), we both increase the “raw” SNR and reduce the noise like contribution from undersampling artifacts.

In the above, the signal localization and quantification are divided into two separate steps. Although computationally daunting, the reconstruction can be reformulated to jointly estimate both the signal origin and tissue properties [20].

Matching vs fitting

Although matching is often highlighted as a key concept in MRF, it may not be the most distinct feature. For example, a dictionary matching procedure can also be used quantify the T2 a traditional turbo spin echo based sequence [21]. Moreover, on some fundamental level fitting and matching are not that different. Whether fitting an analytic solution or searching a dictionary for the optimal match, the same fundamental equations are used to describe the spin-dynamics. The only difference is that fitting is easier when a compact analytical solution is available. In fact, traditional fitting-based techniques are often tailored to take advantage of sequence designs that produce relatively simple dynamics. Unfortunately, only a small subset of all possible MR sequences can easily be described using simple analytical solutions. The signal evolution produced by more elaborate sequences which rapidly sample many distinct magnetization states is often too complex to describe in a tractable way. However, it is usually relatively straightforward to numerically simulate the signal evolution. Such simulations also make it easier to incorporate experimental imperfections, such as B0 variations [11], B1+ non-uniformities [14, 15] and slice profile effects [14, 22]. By switching to matching based reconstruction, it becomes possible to access a much larger design space of possible quantitative MR sequences. This new-found freedom can be used to create more patient friendly sequences [23] and circumvent the need for perfect experimental conditions [14].


No acknowledgement found.


[1] Look, David C., and Donald R. Locker. "Time saving in measurement of NMR and EPR relaxation times." Review of Scientific Instruments 41.2 (1970): 250-251.

[2] Pykett, I. L., et al. "Measurement of spin-lattice relaxation times in nuclear magnetic resonance imaging." Physics in medicine and biology 28.6 (1983): 723.

[3] Carr, Herman Y., and Edward M. Purcell. "Effects of diffusion on free precession in nuclear magnetic resonance experiments." Physical review 94.3 (1954): 630.

[4] Meiboom, Saul, and David Gill. "Modified spin‐echo method for measuring nuclear relaxation times." Review of scientific instruments 29.8 (1958): 688-691.

[5] Deoni, S., et al. "Determination of optimal angles for variable nutation proton magnetic spin‐lattice, T1, and spin‐spin, T2, relaxation times measurement." Magnetic resonance in medicine 51.1 (2004): 194-199.

[6] Winkelmann, S., Schaeffter, T., Koehler, T., Eggers, H. & Doessel, O. An Optimal Radial Profile Order Based on the Golden Ratio for Time-Resolved MRI. IEEE Trans. Med. Imaging 26, (2007) 68–76.

[7] Ehses, P. et al. IR TrueFISP with a golden‐ratio‐based radial readout: Fast quantification of T1, T2, and proton density. Magn. Reson. Med. 69, (2013):71-81.

[8] Velikina, J. V., et al. "Accelerating MR parameter mapping using sparsity‐promoting regularization in parametric dimension." Magn. Reson. Med 70, (2013):1263-1273.

[9] Zhang, Tao, John M. Pauly, and Ives R. Levesque. "Accelerating parameter mapping with a locally low rank constraint." Magnetic resonance in medicine 73.2 (2015): 655-661.

[10] Block, Kai Tobias, Martin Uecker, and Jens Frahm. "Model-based iterative reconstruction for radial fast spin-echo MRI." IEEE transactions on medical imaging 28.11 (2009): 1759-1769.

[11] Ma, Dan, et al. "Magnetic resonance fingerprinting." Nature 495.7440 (2013): 187.

[12] Based on this definition one could consider the valuable work done by [6] and [7] (among others) to describe proto-fingerprinting techniques.

[13] Jiang, Y., Ma, D., Seiberlich, N., Gulani, V. & Griswold, M. A. MR fingerprinting using fast imaging with steady state precession (FISP) with spiral readout. Magn. Reson. Med. 74, (2015):1621–1631.

[14] Cloos, Martijn A., et al. "Multiparametric imaging with heterogeneous radiofrequency fields." Nature communications 7 (2016).

[15] Buonincontri, G. and Sawiak, S. J. MR fingerprinting with simultaneous B1 estimation. Magn. Reson. Med. 76 (2016):1127-35.

[16] Liao, et al., 3D MR fingerprinting with accelerated stack-of-spirals and hybrid sliding-window and GRAPPA reconstruction. Neuroimage. 15;162: (2017):13-22.

[17] Lattanzi R, et al. Phase unwinding for dictionary compression with multiple channel transmission in magnetic resonance fingerprinting. Magn Reson Imaging. 24 (2017):32-38.

[18] McGivney, D. F. et al. SVD compression for magnetic resonance fingerprinting in the time domain. IEEE Trans. Med. Imaging 33, (2014):2311–2322.

[19] Yang M, et al. Low rank approximation methods for MR fingerprinting with large scale dictionaries. Magn. Reson. Med. DOI:10.1002/mrm.26867

[20] Ben‐Eliezer N., et al. Rapid and accurate T2 mapping from multi–spin‐echo data using Bloch‐simulation‐based reconstruction. Magn. Reson. Med. 73, (2015): 809-817.

[21] Sbrizzi A., et al. Fast quantitative MRI as a nonlinear tomography problem. Magn Reson Imaging. 46, (2018): 56-63.

[22] Ma D., et al., Slice profile and B1 corrections in 2D magnetic resonance fingerprinting. Magn. Reson. Med. 78(5) (2017):1781-1789.

[23] Ma D., et al. Music-based magnetic resonance fingerprinting to improve patient comfort during MRI examinations. Magn. Reson. Med. 75, (2016):2303–2314.

Proc. Intl. Soc. Mag. Reson. Med. 26 (2018)