Vessel wall characterization using quantitative MRI: what’s in a number?

Blood suppression is essential for achieving accurate delineation of the vessel wall, which would otherwise be compromised by smearing of the bright-blood lumen signal. Two-dimensional (2D) T2-weighted (T2w) spin-echo sequences have inherent blood suppression due to outflow effects at long echo times; however, this mechanism is not compatible with short echo times needed for T1w imaging.

The first robust technique allowing for 2D T1w black-blood imaging of the arterial vessel wall was proposed by Edelman et al. [19]. This spin-echo-based method achieves blood suppression by using a pair of non-selective and slice-selective inversion pulses (double-inversion recovery axial image recovery, or DIR), thereby effectively inverting only the tissue and blood outside the imaging slice (Fig. 1a). The inversion time (TI) between the inversion pulses and imaging readout is chosen such that blood longitudinal magnetization is nulled (as a result of T1 relaxation), while at the same time non-inverted blood flows out of the imaging slice. The slightly higher signal-to-noise ratio (SNR) and robust blood suppression make T1w imaging the preferred choice for 2D vessel wall thickness measurements.

Unfortunately, the dependency on blood outflow from the imaging slice renders a multi-slice or three-dimensional (3D) implementation ineffective [20]. Another problem obviously arises when post-contrast measurements are performed, where the T1 of blood decreases significantly, and the optimal TI is difficult to determine beforehand. Yarnykh, however, has proposed an elegant solution by introducing a second inversion pair (quadruple inversion recovery, or QIR), making blood suppression effective over a much larger range of T1 values [21].

Increasing interest in isotropic 3D imaging protocols has led to the development of 3D black-blood imaging sequences that do not rely on outflow effects and are therefore more robust against slow flow artifacts. 3D turbo spin-echo (TSE) sequences actually have inherent black-blood properties themselves, caused by the buildup of intravoxel dephasing as a result of the positive gradient moment of the readout gradient during the echo train. While this was presented in the early literature as an alternative to bright-blood brain angiography [22], successful implementation for vessel wall imaging required novel variable-flip-angle (VFA) refocusing schemes [23‐25] that also allowed a stable signal response for longer echo trains (Fig. 1b). This ensures a favorable point-spread function to prevent blurring of the thin vessel wall. Naturally, this can only be achieved for specific values of T1 and T2, and the exact choice of flip angle scheme will always be a trade-off between tissue contrast and effectiveness of flow suppression.

Another class of black-blood methods achieves blood suppression independent of the acquisition scheme through the use of black-blood preparation modules. One method, motion-sensitized driven equilibrium (MSDE, Fig. 1c), combines T2 preparation with flow-sensitizing gradients [26, 27]. If the first moment of the preparation module is sufficiently high, intravoxel dephasing for flowing blood occurs, while the static tissue effectively only undergoes T2 relaxation during the time interval TE_prep .

The final tip-up pulse restores the static tissue longitudinal magnetization for subsequent readout using turbo field-echo (TFE) or TSE acquisition schemes. As shown by Fan et al. [23], the latter seems the more effective strategy, as it adds the black-blood properties of the TSE readout.

An alternative preparation method uses so called-DANTE (delay alternating with nutation for tailored excitation), consisting of a non-selective train of small-flip-angle RF pulses [28, 29]. While this preparation is much longer than MSDE (~100–150 ms), static tissue signal is better preserved and suppression occurs even at low velocities. DANTE is particularly promising for intracranial imaging, where it also enhances suppression of cerebrospinal fluid closely surrounding the vessels [30‐33].

Figure 2a shows examples of 3D MSDE TFE and VFA TSE carotid scans in the same volunteer, illustrating the ability to acquire black-blood coronal images covering the common and internal carotid arteries. Because of the isotropic voxels, these data can easily be reformatted into sagittal and axial views. Figure 2b clearly illustrates that 2D compared to 3D imaging may result in overestimation of wall thickness in regions near or within the bifurcation, where the vessel diameter changes quickly. On the other hand, we also know that in locations with less curvature, ECG-gated 2D DIR scans with in-plane resolution of 0.25 mm have optimal vessel wall delineation, as shown in Fig. 2c. These measurements were validated against histology and ultrasound in a pig model [34], and resulted in a mean carotid vessel wall thickness of 0.49 mm [35].

Given the current resolution in 3D MRI protocols, overestimation of vessel wall thickness is likely to occur [36]—especially considering smaller vessels such as the intracranial or coronary arteries [37, 38]—although this issue may be less problematic in later stages of atherosclerosis, where plaque formation has caused significant wall thickening [39].

While excellent reproducibility values for 2D thickness measurements have been reported for carotid and aortic vessel walls [40, 41], a great advantage of 3D imaging is that it requires no tedious, accurate planning of the imaging slices, while still allowing reconstructions in arbitrary planes. This may partly explain the good reproducibility of 3D black-blood methods reported for various applications, such as the carotid arteries [42, 43], thoracic aortic wall [8, 24, 44], abdominal aorta [44] and even the coronary arteries [38]. While generally not applied, ECG and/or respiratory triggering can further minimize blurring due to vessel wall pulsation or breathing motion [8].

Although different black-blood mechanisms can be easily described, their exact performance in terms of SNR, tissue/lumen contrast-to-noise ratio (CNR), and effective resolution are highly dependent on the exact sequence parameters, including both the preparation module and the specific acquisition scheme (e.g. TFE vs. TSE).

While comparisons between different methods have been reported [29], the large number of sequence parameters makes a fair comparison very challenging. Consequently, no consensus on optimal vessel wall imaging protocols has been reported thus far.

Results from multi-contrast imaging studies show that LRNC has lower T2 than healthy intima/media. Although IPH may show large changes in T2 over time [51], the most commonly found “recent IPH” is associated with elevated T2 values. A traditional way of measuring T2 is by exponential fitting of signal intensities acquired with multi-echo spin-echo sequences (Fig. 3a). While the early literature had already reported in vivo vessel wall T2 quantification from dual spin-echo sequences [52], Biasiolli et al. have only recently improved this approach, allowing for acquisition of multiple echoes between 25 and 100 ms [53]. Since the need for relatively short echo times reduces the outflow effect compared to regular T2w spin-echo sequences, blood suppression was enhanced using DIR preparation (Fig. 1a). Figure 3b nicely illustrates a T2w image of an advanced plaque, along with corresponding quantitative T2 maps. In this study, the mean values for FIB and LRNC were 56 ± 9 and 37 ± 5 ms, while various patches of IPH had T2 values >90 ms. By training a Bayes classifier using expert readings of co-registered multi-contrast data, the authors developed an automated segmentation algorithm that discriminates the various plaque components based on their absolute T2 values. A significant drawback of the method is that it is limited to 2D imaging, with a relatively long acquisition time of 320 R–R intervals despite the use of 5/8 partial Fourier acceleration. Nonetheless, these results show that all relevant plaque components may be discriminated on the basis of T2 alone.

Instead of using multi-echo spin-echo-based sequences, T2* can be determined using a blood-suppressed multi-echo gradient echo approach, with typical echo times of 3–40 ms. Unlike T2, T2* is strongly affected by the presence of magnetic field inhomogeneities and thus might not solely reflect tissue structure. At the same time, this makes T2* imaging very sensitive in detecting protein-bound iron. Raman et al. [54] were the first to conduct an extensive study of the role of iron in atherosclerosis using quantitative T2* measurements. They found a significant decrease in T2* between asymptomatic (34.4 ± 2.7 ms) and symptomatic patients (20.0 ± 1.8 ms). Furthermore, ex vivo iron quantification in endarterectomy specimens showed equal total iron content in both groups, but greatly reduced levels of paramagnetic Fe(III) complexes. Overall, these results strongly suggest that symptomatic plaques are associated with higher amounts of ferritin-bound iron, which was sensitively assessed using quantitative T2* MRI. In addition to endogenous iron, T2*-weighted imaging has also been frequently used to detect macrophage-mediated uptake of intravenously injected superparamagnetic iron oxide (SPIO) particles as a surrogate marker of plaque inflammation [55, 56]. However, such studies still suffered from the use qualitative MRI methods [57]. In more recent studies, quantitative T2* mapping has been applied in combination with QIR blood suppression in order to increase the accuracy in assessing SPIO accumulation by calculating ΔT2* or ΔR2* between pre- and post-contrast scans [58, 59]. While T2* values can be prone to magnetic field inhomogeneities, both studies did report similar baseline T2* values of approximately 25 ms, indicating good reproducibility with this approach. Unfortunately, 3D implementation of these multi-echo techniques has not yet proven feasible, most likely due to the inevitable increase in repetition time, leading to clinically unacceptable acquisition times.

T1 mapping methods are well validated in areas such as brain or cardiac imaging, and in these areas they are generally based on steady-state imaging at multiple flip angles or sampling of the inversion recovery longitudinal relaxation curve [60, 61]. Unfortunately, these are all bright-blood techniques that compromise vessel wall delineation and are susceptible to flow artifacts. We recently developed a black-blood version of the DESPOT1 approach [60] by using MSDE blood suppression with short pre-pulses of 11.5 ms and maintaining steady-state conditions of the 3D RF spoiled gradient echo train by adding dummy pulses after signal acquisition [62]. This enabled 3D vessel wall T1 mapping with isotropic resolution of 0.7 mm. Moreover, T2 mapping was possible using the same sequence by varying the TE_prep time of the MSDE pre-pulse. While good reproducibility of this 3D approach was shown, both carotid T1 and T2 values (844 ± 96 ms/39 ± 5) were lower than with single-slice TSE-based measurements (1227 ± 47/55 ± 11 ms) [63]. Furthermore, the use of MSDE pre-pulses causes T1 modulation in k-space, which can lead to some blurring of the resulting T1 maps. Figure 4a shows quantitative T1 mapping results from a carotid plaque where significantly reduced T1 is indicative of IPH. Simultaneous non-contrast angiography and intraplaque hemorrhage (SNAP) is a recently published method for detecting IPH, which can identify regions of IPH by phase-sensitive reconstruction of inversion recovery data [64]. Figure 4b shows SNAP⁺ and SNAP⁻ images corresponding to the region shown in Fig. 4a, representing IPH and arterial blood, respectively. Since the sign of the SNAP signal changes below a specific T1 value, we can show that the IPH region visible in the SNAP⁺ image can be recreated from the quantitative T1 map by simple thresholding.

While the main benefit in using quantitative T1 and T2^(*) mapping—compared to multi-contrast T1w and T2^(*)w imaging—is to improve reproducibility and facilitate longitudinal monitoring of changes in plaque composition, it still does not allow for direct assessment of the relative contribution of different tissues in each voxel. In contrast, for instance, Koppal et al. were able to assess quantitative maps of fat content based on Dixon imaging, showing significant differences between the lipid core (12.6%) and surrounding tissue (9.2%) [65]. On the other hand, the use of multi-contrast imaging to detect different plaque components (i.e. LRNC, IPH) has been well validated against histology [45, 51]. Quantitative relaxation parameter mapping extends this concept and might enable better definition of thresholds for discriminating between these different tissue types. Indeed, a recent study reported that LRNC detection based on T2 mapping—pixels with T2 <42 or T2 >90 ms when IPH was included—showed very good correlation with histology (R = 0.85) and had good sensitivity (AUC = 0.79) for detecting recently symptomatic plaques [66].

MRI is also able to quantify water diffusion within tissues. Strong field gradients applied on each side of a 180° refocusing pulse cause diffusion-mediated signal attenuation due to phase dispersion of spins. Conversely, in static tissue, the effect of both gradients cancels out and the signal is maintained. The degree of diffusion weighting is given by the b-value, which depends on the gradient strength, duration and spacing [67]. Similar to varying TE to quantify T2^(*), the apparent diffusion coefficient (ADC) can be estimated by an exponential fit of signals acquired at different b-values, which determines the amount of diffusion weighting. In vessel wall imaging, diffusion weighting is of particular interest for detecting the presence of an LRNC, which from ex vivo studies has been known to have a strongly decreased ADC [68, 69]. In fact, Clarke et al. showed that, compared to T1 and T2 quantification, ADC was the parameter that could best distinguish LRNC from FIB [70].

Diffusion-weighted imaging (DWI) is typically performed using 2D single-shot echo-planar imaging (EPI) sequences, which provides time efficiency for sufficient averaging of the low DW signal (because of the additional need for long TEs). However, EPI is very susceptible to B₀ inhomogeneities, because phase errors accumulate for each additional phase encoding step. Kim et al. [71] were the first to apply inner-volume imaging for vessel wall applications, reducing the effective field of view (FOV) and thereby the echo train length by a factor of 4. In this way, good-quality ADC maps of 2-mm slices were obtained with a resolution of 1 × 1 mm². This spatial resolution, however, is still inferior to the resolution used for vessel wall thickness measurements.

A novel approach enabling 3D vessel wall DWI was recently presented by Xie et al. [72]. They decoupled diffusion weighting from the imaging readout through the use of a motion-compensated diffusion-weighted pre-pulse (Fig. 5a). Combined with 3D TSE acquisitions, this resulted in high-resolution imaging of 0.6 × 0.6 × 2 mm³ (Fig. 5b). Additional double-inversion recovery preparation and low-b-value motion-sensitized gradient on top of diffusion weighting further guaranteed efficient blood suppression. A large decrease in LRNC ADC values compared to FIB (0.6 × 10⁻³ vs. 1.27 × 10⁻³ mm²/s) was shown, indicating that ADC mapping may serve as a sensitive and quantitative non-contrast technique for plaque characterization (Fig. 5c). Using a similar approach but with a stimulated echo pathway, Zhang et al. [73] recently showed that phase errors arising from eddy currents could be prevented. This resulted in stable carotid artery ADC values across subjects of 1.4 ± 0.23 × 10⁻³, which is comparable to results of earlier studies.

While challenging, extension of DWI to a diffusion tensor imaging (DTI) protocol using multiple gradient directions would allow calculation of the fractional anisotropy (FA) of the vessel wall microscopic fiber structure. Opriessnig et al. [74] were recently the first to apply a 2D DTI sequence using four b-values and 18 diffusion directions on a 10-mm carotid artery segment. In 12 healthy volunteers, the authors found a significant correlation between FA and age, indicating possible alterations of the vessel wall microstructural integrity. Moreover, the reproducibility of FA measurements appeared very high, with CV values no higher than approximately 5%.

In addition to DCE-MRI, various other techniques have been proposed for quantifying endothelial permeability in atherosclerotic plaques. Delayed-enhancement imaging with either low molecular weight gadolinium chelates or albumin-binding agents has been used as a semi-quantitative measure of plaque permeability in both animal models (mice [108], rabbits [109‐111]) and humans [112, 113]. Phinikaridou et al. [114] and Bar et al. [115] quantified endothelial permeability in the brachiocephalic artery of atherosclerotic mice as a change in the vessel wall relaxation rate (R1, s⁻¹) 30 min after injection of an albumin-binding contrast agent (gadofosveset trisodium). This technique has also been used successfully to quantify changes in permeability in the murine brachiocephalic artery after therapeutic intervention [111, 116]. More recently, Phinikaridou et al. [117] used this same technique to quantify endothelial permeability in aortic plaques in atherosclerotic rabbits, and demonstrated higher R1 (indicative of higher permeability) in aortic segments more prone to disruption after injection of Russell’s viper venom. Unlike quantitative dynamic imaging with DCE-MRI, these techniques measure signal enhancement or quantify tissue relaxation time only at a fixed point after contrast agent injection. This approach is particularly well suited for higher molecular weight or albumin-binding contrast agents, whose plaque uptake kinetics are intrinsically slower. While failing to capture the dynamic uptake of contrast agents over time, these techniques offer a simpler and robust alternative to DCE-MRI for quantification of plaque microvascularization and permeability.

Atherosclerosis originates predominantly at regions with perturbed flow that can occur at the outer edges of vessel bifurcations. In these regions, hemodynamic wall shear stress (WSS), the frictional force sensitized by endothelial cells forming the inner lining of blood vessels, is weaker than in protected regions and can even exhibit direction reversal. The atherogenic endothelial phenotype resulting from low WSS mediates recruitment and activation of monocytes, which can subsequently lead to plaque formation [118]. WSS is also known to increase with increasing blood flow. In response, the vessel dilates to reduce blood flow such that WSS returns to normal values. Regions where WSS is chronically elevated, such as the apices of bifurcations in the cerebral vasculature, are predisposed to the formation of aneurysms [119]. WSS therefore represents a key link between blood flow and alterations in biomechanical vessel wall parameters.

Flow measurement using MRI has traditionally been performed using 2D phase-contrast imaging in most vascular beds. With gating to the cardiac cycle, a time-resolved (cine) measurement can be performed [120]. Currently, a comprehensive assessment of flow in an entire 3D volume (3D cine phase-contrast MRI or 4D flow MRI) is feasible [121], enabling the measurement of full time-varying 3D velocity fields in a wide variety of cardiovascular regions [122].

The importance of WSS in vascular disease has led to widespread interest among researchers in obtaining reliable estimates of WSS from MRI-measured velocity data. Oshinski et al. were the first to develop a method based on linear fitting through velocity values to obtain the velocity derivative (the shear rate) at the wall [123]. Multiplication of the shear rate by dynamic viscosity yields the WSS. Other groups developed WSS estimation based on parabolic fitting, which showed greater accuracy than linear fitting [124, 125]. Stalder et al. used cubic B-splines to derive the shear rate at the wall [126]. Another important hemodynamic parameter shown to correlate with atherosclerosis is the oscillatory shear index (OSI) [127]. OSI represents the temporal oscillation of WSS during the cardiac cycle, the deviation of WSS from its predominant direction parallel to the vessel. Thus, the OSI can be calculated using methods to estimate time-resolved WSS.

These techniques applied to MRI-measured flow data have provided valuable insight into the relation between abnormal WSS and pathophysiology. Duivenvoorden et al. showed that WSS was an independent predictor of carotid wall thickness, lumen area and vessel size [128]. Mutsaerts et al. found that WSS was associated with periventricular white matter lesions and cerebral infarcts [129]. Markl et al. observed that low WSS and high OSI, which are potentially atherogenic wall parameters, were predominantly concentrated at the posterior wall of the internal carotid artery in normal controls, a region known to be prone to atherosclerosis [130]. Wentzel et al. reported that the presence of atherosclerotic plaques in the descending aorta was associated with low WSS [131]. Other studies showed that WSS on the ascending aorta was elevated compared to healthy controls in bicuspid valve disease, which implicates a relationship between elevated WSS and aortic dilation [132, 133].

However, these algorithms were based on 2D phase-contrast MRI or the manual placement of planes in 4D flow MRI data perpendicular to the vessel of interest. With 2D techniques highly focused on specific vascular landmarks, focal abnormal WSS expression on the vessel of interest may be missed. Also, the manual placement of 2D planes for WSS analysis can be a laborious task. Bieging et al. were the first to develop a WSS algorithm based on 4D flow MRI that was capable of estimating WSS along the entire wall of the ascending aorta [134]. The linear least squares method was used to fit a line through three velocity vectors along the inward normal. Potters et al. expanded on this method with the use of smoothing spline fitting to mitigate the influence of noise. The feasibility of estimating WSS along the entire aorta (the arch and descending part included) and the carotid bifurcation was also demonstrated (see Fig. 8a [135]). Cibis et al. recently found an inverse relationship between wall thickness and time-averaged WSS in patients with asymptomatic plaque, calculated with the algorithm described in Potters et al. in the carotid bifurcation (Fig. 8b) [136].

Some important considerations should be kept in mind when estimating 4D flow MRI-derived WSS. First, several studies showed that an accurate definition of the vessel wall is paramount [125, 135]. Nonetheless, low inter-observer variability in WSS was found for both 2D (<10%) and 3D (<5%) WSS algorithms [137, 138]. Second, the absolute value of WSS decreases with spatial resolution [126, 135]. Thus, 4D flow MRI-derived WSS is always underestimated compared to computational fluid dynamics where fine meshes are used [139]. Qualitatively, however, regions of high and low WSS and the direction of WSS tend to correspond well [140‐142](Fig. 7).

Another parameter that can be assessed using phase-contrast MRI is pulse wave velocity (PWV), which characterizes the speed of the arterial pulse through a specific part of the circulation. PWV is a measure of arterial stiffness and can be directly correlated with the vessel wall elastic modulus E using the Moens-Korteweg equation: where h is the vessel wall thickness, ρ is the blood density and d is the vessel diameter. The elasticity of the vessel wall decreases as part of natural aging. MRI-based PWV measurements have indeed shown significant increases in PWV as a function of age in the carotid arteries and the aorta [143], ranging from roughly 5 m/s for young adults to 7–8 m/s in the elderly. More importantly, arterial stiffness has also been associated with atherosclerosis [144], and studies have shown that the measurement of PWV in various vascular beds significantly improves the prediction of future cardiovascular events [145‐147].

While Doppler ultrasound is the clinical workhorse for determining PWV, MRI might offer several advantages. Particularly in 3D anatomical imaging, MRI enables better visualization of each vessel segment, independent of angle and depth. This likely improves reproducibility due to ease of planning, as well as by a more accurate determination of the path length of the pulse wave [148, 149]. Many approaches to calculate PWV are mentioned in the literature and unfortunately no clear consensus exists on which method produces most reliable PWV values. The most widely used method is by measuring flow-time curves at two different slices along the vessel. The so-called foot–foot method can then be used to calculate PWV by dividing the path length between the slices by the time difference between the initial up-slopes of the two flow curves (Fig. 9a). The exact location of the foot is slightly affected by the baseline definition and the number of points used to define the linear part of the up-slope (usually 20–80% of the maximum flow). Another analysis method finds the time shift at which both curves have the highest correlation; however, this can be affected by the correlation window and presence of wave reflections starting from the systolic downslope [150]. Instead of relying on phase-contrast acquisitions in two slices, PWV could be calculated from measurements in a single or multiple oblique sagittal slices [151, 152]. Flow–time curves could then be plotted for each pixel along a centerline of the vessel, which produces a set of time shifts as a function of distance, presumably resulting in a more reliable estimate of PWV. For the aorta, this has been extended to volumetric flow measurements using highly accelerated 4D flow acquisitions, which further facilitates the planning of the imaging volume (see [153], Fig. 8). Tortuous vessels such as the carotids could benefit from this methodology, although temporal and spatial resolution still appear to be too low for this specific application. A third method provides even more local vessel wall PWV values by assessing changes in lumen diameter (A) as a function of flow (Q) at different time points during the cardiac cycle, where PWV is given by ΔQ/ΔA during the initial up-slope of this relation (Fig. 9b). In a study comparing the above-mentioned methods for the aorta, Ibrahim et al. found the best agreement between the foot–foot and cross-correlation methods, which also showed the highest reproducibility [148]. While analytical methods vary widely among studies, results for group-averaged aortic PWV have been quite consistent (4–5 m/s) in young adults [149‐151, 154]. Unfortunately, less data is available on carotid PWV (~5.5 m/s) and femoral arteries (~7 m/s) [143, 155], for which the latter interestingly seems to decrease with age. Finally, while repeated measurements on the same day have shown good repeatability [149, 156], day-to-day physiological variations in blood pressure and blood flow might limit the applicability of PWV measurements for sensitive monitoring of gradual changes over time.

Recent years have seen the introduction of 7T imaging in clinical research, with the direct advantage of an increase in SNR directly proportional to the strength of the magnetic field [166].

While challenging at 3T [167], this makes 7T MRI of particular interest for characterization of intracranial atherosclerosis, where the even smaller vessel size requires sufficient signal for accurate delineation of the vessel wall. Researchers from UMC Utrecht have been pioneers in this field by developing 7T intracranial vessel wall imaging protocols [168], as well as evaluating the benefit of 7T versus 3T MRI in detecting atherosclerotic plaques (see also Fig. 10) [18]. On the one hand, they found that 7T MRI resulted in better wall definition. This was mainly achieved by exploiting the increased SNR for higher acceleration factors, enabling the use of very robust but inherently time-inefficient inversion recovery-based cerebrospinal fluid (CSF) suppression. Although in many cases lesions were detected only at 7T, the authors also noted a significant number of cases showing the opposite. Similarly, studies to date on carotid plaque characterization have shown no strong indications of the superiority of 7T over 3T [169, 170], which might be related to the more inhomogeneous transmit field at 7T compromising non-selective blood suppression techniques, as well as signal homogeneity. In this respect, there is a considerable need for better-designed high-field carotid coils in order to achieve improved transmit and receive characteristics [171, 172]. Similarly, developments in cardiac 7T coil design may boost the field of coronary imaging by exploiting the increase in SNR for obtaining higher spatial resolution [173].

Quantitative imaging generally comes with the need to acquire multiple images, either having different sensitivity to the parameter of interest (T1, T2, ADC) or sampling a dynamic process over time (DCE, flow). Long acquisition times are therefore one of the big hurdles in this field, and complicate the translation of such methods to a clinical environment. What may be able to change this in the near future is the rapid improvements in reconstruction and post-processing techniques. In recent years, mathematics has played an increasingly important role in advancing MRI technology. In particular, the introduction of “compressed sensing” has taught us that images can be reconstructed with far less data than was considered necessary using assumptions of image sparsity in combination with iterative reconstruction algorithms [174]. As this is quite a generally applicable concept, many researchers have already experimented with compressed sensing to improve vessel wall MRI. For example, 3D MSDE acquisitions were accelerated to a factor of 5 without significant deviations in assessing vessel wall or plaque component dimensions [175, 176]. Gong et al. realized that multi-contrast vessel wall MRI could be significantly accelerated by assuming significant shareable information between the different contrast acquisitions [177]. The use of their proposed reconstruction algorithm allowed for highly accelerated T2w and PDw scans (up to a factor of 6) once a moderately accelerated T1w scan (e.g. using SENSE) was available. Using regular CS techniques, Yuan et al. reduced a 3D MSDE-based T2 mapping protocol to a clinically acceptable imaging time of 7 min [178]; reconstruction of separate images was done independently. A more promising technique related to quantitative relaxation time (T1/T2) mapping are model-based algorithms that use prior knowledge of the relaxation equations to reconstruct undersampled scans at multiple inversion or echo times [179, 180]

For dynamic or time-resolved imaging, temporal relations between images are used in order to achieve high undersampling factors for the individual time frames. This technique shows great promise for achieving higher-temporal-resolution 3D DCE imaging [86] or allowing interleaved sampling of black- and bright-blood images for the simultaneous measurement of the arterial input function and vessel wall signal response [106]. For 4D flow applications, the use of k-t GRAPPA has enabled acceleration factors of up to 5 without substantial errors in derived WSS parameters [181].

Perhaps the greatest benefit of freely undersampling k-space data is the ability to select the optimal data retrospectively. This can be used to circumvent the need for cardiac and respiratory gating, without losing scan efficiency, and order the data in the reconstruction process based on available data of cardiac and respiratory motion from either sensors or MR navigators. Using appropriate motion correction algorithms, this even allows for 100% scan efficiency [182, 183]. Recent impressive data from Ginami et al. [184] showed high-quality coronary vessel wall imaging by making reconstructions using different timings of the acquisition window within the cardiac cycle in order to retrospectively achieve the optimal vessel wall delineation.

In addition to the development of novel MR methods and contrast mechanisms, the integration of MRI with other imaging modalities that interrogate different aspects of plaque physiology is quickly becoming a reality. As a notable example, the introduction of simultaneous PET/MRI systems allows for the seamless combination of anatomical and physiological imaging with MRI, and metabolic/functional imaging with PET. PET is a highly sensitive modality, and has already been extensively validated for quantification of plaque macrophages with ¹⁸F-FDG [185, 186]. However, PET is an imaging modality with intrinsically low spatial resolution, and it is traditionally combined with computed tomography (CT) for improved anatomical localization. Combining PET with MRI instead of CT offers several advantages. MRI provides high-spatial-resolution imaging and better soft tissue contrast than CT, and enables the quantification of several physiological parameters in the vessel wall, as previously described in this review. The better anatomical definition of MRI can be used to apply partial volume and motion corrections to improve the localization of the PET signal [187, 188].

Last but not least, combining PET with MRI instead of CT reduces patient exposure to ionizing radiation—a highly desirable feature for longitudinal, repeated imaging in patients with chronic diseases (such as atherosclerosis). The fact that the two modalities are intrinsically co-registered allows for easier image interpretation, image analysis and experimental design [188, 189]. However, there are also specific challenges that may arise when combining these two modalities, such as the effective conversion of MR images to accurate PET attenuation maps, which are still the subject of active investigation [189].