Primary Multiparametric Quantitative Brain MRI: State-of-the-Art Relaxometric and Proton Density Mapping Techniques

Abstract

This review on brain multiparametric quantitative MRI (MP-qMRI) focuses on the primary subset of quantitative MRI (qMRI) parameters that represent the mobile ("free") and bound ("motion-restricted") proton pools. Such primary parameters are the proton densities, relaxation times, and magnetization transfer parameters. Diffusion qMRI is also included because of its wide implementation in complete clinical MP-qMRI application. MP-qMRI advances were reviewed over the past 2 decades, with substantial progress observed toward accelerating image acquisition and increasing mapping accuracy. Areas that need further investigation and refinement are identified as follows: (a) the biologic underpinnings of qMRI parameter values and their changes with age and/or disease and (b) the theoretical limitations implicitly built into most qMRI mapping algorithms that do not distinguish between the different spatial scales of voxels versus spin packets, the central physical object of the Bloch theory. With rapidly improving image processing techniques and continuous advances in computer hardware, MP-qMRI has the potential for implementation in a wide range of clinical applications. Currently, three emerging MP-qMRI applications are synthetic MRI, macrostructural qMRI, and microstructural tissue modeling.

Graphical abstract

Figure 1:

Spatial scales in MRI theory. (A) Three-dimensional rendering of the theoretical voxel shape in Fourier MRI as represented by the voxel sensitivity function. (B) Three spatial scales and the associated theories: from quantum physics of dipolar interactions between individual spins (bottom) to semiclassical physics describing spin packets with use of the Bloch equation (middle) to imaging theory describing voxels (top). Microstructural tissue modeling (MTM) is a venue for deconstructing voxels into spin packets. Knowing the voxel sensitivity function can be useful for modeling the actual intravoxel structure and providing a link between the mesoscopic and voxel scales.

Figure 2:

Bloembergen, Purcell, and Pound–Dixon theory. Computer simulations show T1 and T2 curves as functions of the correlation time (τ_c) for several values of the main magnetic field in the range of 50 mT to 7.0 T. In the short correlation time regimen of cerebrospinal fluid (CSF) and the free pool of tissue, the T1 and T2 curves are independent of B₀ and decrease as functions of increasing correlation time. Following a transition range, the T1 and T2 curves show opposite dependencies on the correlation time. These curves are valid for ¹H nucleus in any tissue environment at the spatial scale of spin packets, but not necessarily at the much larger spatial scale of voxels, depending on the spatial scalability of the voxel. A typical tissue would have a distribution of correlation times such as the one shown in cyan: free (narrow peak) and potentially broad macromolecular feature. Interpool magnetization transfer (MT) and/or magnetization exchange (ME) effects lead to B₀ dependencies of the relaxation times. Horizontal lines indicate the detection limits for conventional MRI (T2, >1 msec) and ultrashort echo time (UTE-MRI) (T2, >15 µsec). Mag = magnetic.

Figure 3:

The four-pool model of white matter. The four ¹H-nuclei pools are depicted with use of a transmission electron micrograph of a myelinated axon in cross-section. The myelin layer (concentric) surrounds the axon of a neuron, showing cytoplasmatic organs inside. The nonaqueous pools 1 and 4 are detectable only with ultrashort echo time pulse sequences due to the extremely short T2s. The micrograph of this cadaveric tissue sample was generated and deposited into the public domain by the Electron Microscopy Facility at Trinity College, Hartford, Connecticut http://www.trincoll.edu). IEW = intra- and extracellular water, M = myelin protons, MW = myelin water, NM = nonmyelin protons.

Figure 4:

Magnetization transfer (MT) basics. (A, B) Graphs of the free and the restricted-motion pools illustrate the simplest MT imaging experiment in which an off-resonance radiofrequency pulse (red arrow) is used to saturate the semisolid pool (A), resulting subsequently in a net decrease of the liquid pool proton density (B) available for signal generation, hence additional MT contrast. M_z = z-component of the magnetization. (C–G) Quantitative MT images of the adult human brain in the axial plane. The parameter maps shown are the pool size ratio F, forward exchange rate k_f, spin-lattice relaxation rate of the free pool, spin-spin relaxation rate of the free pool, and spin-spin relaxation rate of the restricted-motion pool. (Brain images reproduced from reference . Note: The referenced article is available under the Creative Commons CC-BY-NC-ND license, which permits noncommercial use of the work as published, without adaptation or alteration, provided the work is fully attributed.)

Figure 5:

Multiparametric quantitative MRI (MP-qMRI) example with the triple turbo spin-echo pulse sequence. Same-section directly acquired images and quantitative MRI parameter maps in the axial plane. The left column shows the three images directly acquired with (A) proton density weighting (PDw), (B) T2 weighting (T2w), and (C) T1 weighting (T1w). The insert on the right shows the same-section MP-qMRI maps: (D) the cerebrospinal fluid–normalized proton density map (nPD) and (E, F) relaxation rate maps. Note that the inverse white matter (WM)–to–gray matter contrast in the relaxation rate maps versus (G, H) the relaxation time maps, in which WM is not only brighter but also shows marked texture. Furthermore, the WM texture of the longitudinal relaxation rate (R1) map is different from that in the transverse relaxation rate (R2) map. All images are in a 35-year-old healthy male volunteer and were acquired with the 9-minute scan (80 contiguous sections; reconstructed voxel = 0.5 × 0.5 × 2 mm) used for the ELGAN-ECHO (Extremely Low Gestational Age Newborns–Environmental influences on Child Health Outcomes) study (103).

Figure 6:

Proton density (PD) mapping. (A) PD map generated with the triple turbo spin-echo pulse sequence and the processing steps outlined in the text. Note that the superimposed single-line profile shows negligible residual coil profile effects. (B) Whole-brain histograms of PD, T1, and T2 for generated with triple turbo spin-echo (80 consecutive sections; voxel = 0.5 × 0.5 × 2 mm; scan time, approximately 9 minutes). Accuracy of the PD map is further ascertained by the very similar bimodal PD histogram shape compared with the T1 relaxogram below, which is consistent with the empirical 1/PD ∝ 1/T1 relationship that holds for healthy brain tissue (see text). This map is in a 41-year-old healthy female volunteer and was acquired with the same protocol used for the ELGAN-ECHO (Extremely Low Gestational Age Newborns–Environmental influences on Child Health Outcomes) study (103). CSF = cerebrospinal fluid, GM = gray matter, WM = white matter.

Figure 7:

Voxel types according to spatial scalability. Three different voxel types with increasing inner complexity from left to right (simulation colors: blue = cerebrospinal fluid [CSF], red = gray matter [GM], green = white matter [WM]). Type 1, which contains pure CSF (left), is fully scalable, meaning that the quantitative MRI (qMRI) parameters at the spin-packet and voxel scales are equal. Type 2 voxels contain a typical structured soft tissue (eg, WM or GM), leading to a distribution of correlation times and magnetization transfer effects, as reviewed in the text. Type 3 voxels are difficult for qMRI algorithms because of partial volume effects.

Figure 8:

Partial volume artifacts. (A) T1-weighted image is compared with (B) the corresponding T1 map of the American College of Radiology MRI accreditation phantom. The arrows and outlines in A and B correspond to the magnifications in (C) and (D), respectively. Even though the T1 map has less intensity variation, bright pixels near the edges (see magnifications) result in a less natural appearance. As discussed in the text, improving the visual appearance of quantitative MRI (qMRI) maps is very important for the clinical adoption of qMRI.

Figure 9:

Schematic overview of the convolutional neural networks model with two modules for tissue property mapping. (A) The feature extraction module consists of four fully connected layers (FNNs), which is designed to mimic singular value decomposition to reduce the dimension of signal evolutions. The U-Net structure was used for the spatially constrained quantification module to capture spatial information from neighboring pixels for improved quantification of tissue properties. MR fingerprinting (MRF) images of three contiguous sections (red) were used as input, and the corresponding reference T1 and T2 maps from the central section were used as output for the network. (B) Reformatted quantitative maps in axial, coronal, and sagittal views from the prospectively accelerated scan (three-dimensional MRF, standard; R = 2; 192 time points). The acquisition time for 176 sections was about 7 minutes. (From reference . Note: The referenced article is available under the Creative Commons CC-BY-NC-ND license, which permits noncommercial use of the work as published, without adaptation or alteration, provided the work is fully attributed.)

Figure 10:

Representative in vivo T1, T2, and apparent diffusion coefficient (ADC) mapping of three sections with use of multitasking and the respective reference protocols in a healthy volunteer. Multitasking provides T1, T2, and ADC maps with good qualitative agreement with the references and without image distortion (white arrows), which can be observed on single‐shot echoplanar imaging ADC maps. (Reprinted, with permission, from reference .)

Figure 11:

R1 weighting. (A) Series of R1-weighted synthetic MR (SynthMR) images exposing the underlying white matter (WM) texture. (B) Upon projection along the superior-to-inferior direction, a rendering resembling a WM fibrogram was generated. This association needs validation with ex vivo experiments. These synthetic images are in the same 41-year-old healthy female participant as in Figure 6. SL = section number.