Pushing MP2RAGE boundaries: Ultimate time‐efficient parameterization combined with exhaustive T1 synthetic contrasts

MP2RAGE parameter optimization is redefined to allow more time‐efficient MR acquisitions, whereas the T1‐based synthetic imaging framework is used to obtain on‐demand T1‐weighted contrasts. Our aim was to validate this concept on healthy volunteers and patients with multiple sclerosis, using plug‐and‐play parallel‐transmission brain imaging at 7 T.


INTRODUCTION
The MP2RAGE (magnetization prepared 2 rapid acquisition gradient echoes) sequence 1 is used widely used for 3D structural T 1 -weighted imaging of the brain at ultrahigh field (UHF; i.e., 7 T and above) but also increasingly at clinical 3T magnetic fields. 2 MP2RAGE parameterization relies on the choice of several key parameters, notably including the TR, flip angle (FA), and the two TIs (TI 1 and TI 2 ) when gradient-echo (GRE TI1 & GRE TI2 ) images are acquired.These parameters were initially optimized by Marques et al. to obtain the highest and most uniform T 1 -weighted brain tissue contrast when combining the two acquired images in a complex ratio.][14] In addition to this main UNI contrast, the original contrasts obtained with each inversion of the MP2RAGE (GRE TI1 and GRE TI2 ) could also provide new and specific information.Thus, it turns out that the contrast obtained during the first inversion (GRE TI1 ) of the MP2RAGE nullifies the voxels at the boundary between gray and white matter, a characteristic that has shown its relevance in focal cortical dysplasia detection at 3 T 15 and 7 T 16 (resulting in the renaming of this contrast as "EDGE" for edge-enhancing gradient echo).The FLAWS (fluid and white-matter suppression) parameterization has been proposed with the goal of obtaining clinically relevant acquired contrasts.FLAWS is a MP2RAGE-derived sequence 17 that is further optimized at diverse magnetic fields by Beaumont et al. [18][19][20] and is characterized by two specific TI values: allowing the suppression of CSF and white-matter (WM) signals.Therefore, FLAWS provides a WM-suppressed imaging (FLAWS 1 , close to the FGATIR contrast [fast gray-matter acquisition T 1 inversion recovery]) that is useful for targeting deep structures before deep brain stimulation 21 ) and CSF-suppressed imaging (FLAWS 2 , a MPRAGE-like contrast), nevertheless still dependent on transmission and reception bias.These two acquired images can be combined to suppress both WM and CSF, resulting in a gray matter (GM)-specific contrast (FLAWS MIN , which could be relevant in epilepsy, notably for focal cortical dysplasia detection 22 ).Finally, by further combining FLAWS 1 and FLAWS 2 , a B − 1 -free contrast close to UNI with high GM/WM contrast could also be generated (FLAWS HCO ).FLAWS HCO already demonstrated high potential in MS imaging for lesions detection in the cortex 23 and the cervical spinal cord 24 at 3 T.All these specific FLAWS contrasts, not available when using the conventional MP2RAGE parameterization, could therefore be useful in several clinical contexts at 7 T and 3 T.
Nevertheless, both MP2RAGE and FLAWS parameterizations are determined by the choice of a specific pair of TI delays yielding the desired image contrasts.This implies that MP2RAGE and FLAWS contrasts (i.e., UNI, EDGE, FGATIR, MPRAGE, FLAWS MIN , and FLAWS HC&HCO ) cannot all be obtained in a single acquisition.A novel MP2RAGE optimization 25 using extended phase graph (EPG) has recently been proposed at 7 T by Dokumaci et al. to simultaneously provide several key contrasts, such as UNI and FLAWS MIN , as well as T 1 mapping, in a single acquisition.However, this MP2RAGE parameterization choice came at the cost of inevitable tradeoffs, notably regarding achieved contrast-to-noise ratio (CNR), achievable spatial resolution, and visual outputs of clinical contrasts.
A fundamentally different way to obtain multiple clinically relevant contrasts using a single MP2RAGE acquisition is through T 1 -based synthetic imaging. 26Massire et al. recently demonstrated how to generate synthetic UNI and FLAWS MIN images (sUNI, sFLAWS MIN ) based on a T 1 map acquired from a conventional MP2RAGE imaging and signal equation reintegration at 7 T. Pushing this formalism further, it becomes possible to completely separate the MP2RAGE parameterization choice (notably the TIs) from the desired output contrasts, which will be generated exclusively based on the obtained T 1 map.The resulting freedom on MP2RAGE parameter choice could immediately be reinvested to allow a more time-efficient acquisition scheme, opening the way for faster scan time or increased spatial resolution for a given time budget.
This approach is supported by the use of parallel transmission 27 (pTx) technology and the universal pulses 28 technique, which enable overcoming the transmit field (B + 1 ) inhomogeneities, one of the main obstacles affecting the MP2RAGE sequence at UHF.Although this bias had to be mitigated using B + 1 maps in the original optimizations of MP2RAGE, 1 FLAWS, 20 or MP2RAGE-FLAWS, 25 this is no longer necessary when using pTx, making this new parameterization framework even more attractive.
Consequently, the aim of this study was to investigate a newly redefined and efficient MP2RAGE parameterization, subsequently used to generate on-demand online clinically relevant synthetic contrasts using pTx at UHF.The proposed method with its large portfolio of contrasts was experimentally validated on healthy volunteers and compared in quality and speed with whole-brain UNI images generated with a conventional MP2RAGE parameterization on a large range of spatial resolutions, from conventional clinical resolution (1 × 1 × 1 mm 3 ) to ultra-high resolution (0.45 × 0.45 × 0.45 mm 3 ).The opportunities provided by this new framework were assessed in terms of clinical relevance on imaging of patients with MS for a given time budget, from 0.80 × 0.80 × 0.80 mm 3 up to 0.67 × 0.67 × 0.67 mm 3 resolution at 7 T.

Theory of MP2RAGE signal and consequences on parameterization
MP2RAGE is an inversion recovery-based 3D sequence in which two GRE images are acquired with different FAs ( 1 / 2 ).A purely T 1 -weighted UNI image, numerically bound within the [−0.5; +0.5] range, can be generated with the following equation 1 : where GRE TI1 and GRE TI2 are the complex signal intensities of the GRE volumes acquired at TI 1 and TI 2 , and * is the complex conjugate operator.Note that proton density, T * 2 , and B − 1 dependencies are naturally canceled out in UNI.A direct relationship between UNI signal intensity and T 1 values can be numerically calculated for any voxel. 1n addition, assuming a perfect magnetization inversion, it could be observed 26 that UNI signal is only a function of T 1 and the following MP2RAGE parameters: TI 1 , TI 2 , α 1 , α 2 , n, ES, and TR (where n is the number of partition-encoding steps, ES is the echo spacing of GRE readout modules, and TR is the sequence repetition time that directly drives the acquisition duration).
In this work, the central idea consists of not focusing on the radiological relevance of acquired or combinatory contrasts, but rather using a T 1 mapping as an intermediate step for generating all the desired synthetic contrasts (see Section 2.3).Consequently, T 1 -based synthetic imaging opens the way to redefine how MP2RAGE parameterization is performed.MP2RAGE parameterization now breaks down like this: (1) Spatial resolution is chosen to set up n; (2) ES is adjusted through the receiver bandwidth to set a desired SNR level and to define GRE train durations; (3) because specific TI values are not imposed anymore, these are chosen as short as possible to fit the two GRE trains back-to-back in the chronogram; and (4) TR is set as short as possible to limit idle time and reduce acquisition total duration.Only the choice of FA values (which do not affect duration) should be optimized to maximize the SNR of the quantitative T 1 map, conveniently with EPG.

FA optimizations with EPG simulations
After TI 1&2 , ES, and TR were fixed to shorten the total acquisition time, a relevant target consisted of maximizing the WM/GM CNR obtained in the original UNI image used to compute the T 1 map by adjusting the only remaining parameters (i.e., α 1 and α 2 ).This criterion was purposely chosen, as the synthetic MP2RAGE framework requires reliable T 1 mapping on the physiological T 1 range of interest.To do this, a MP2RAGE EPG simulation tool implemented by Dokumaci et al. 25 on MATLAB (R2016a; MathWorks, Natick, MA, USA) was used to generate WM and GM signals.An FA range of 1 -10 was considered with independent steps of 1 for both α 1 and α 2 .The chosen T 1 values for WM/GM/CSF were 1150/1900/3350 ms, respectively, whereas original code implementation, as well as T 2 and proton density values, were kept unchanged.UNI WM/GM CNR values were computed based on the equations detailed in appendix 2 of Ref. 1.For each spatial resolution, an "optimal" FA combination was chosen, keeping in mind that a bijective T 1 -to-UNI signal relationship in the T 1 range of interest is nonetheless necessary to avoid any ambiguity in T 1 calculation.To verify this, the theorical curve of UNI signal vs quantitative T 1 was plotted for each set of parameters (see Figure S1A-C).The potential bias induced by B + 1 variations possibly remaining despite pTx use was simulated for the time-efficient (0.65 mm) 3 protocol as follows: modeling T 1 curves calculated based on UNI signal for a range of B + 1 variation between 85 and 115%, estimating the possible error range in the WM/GM contrast calculation, and then applying this range to the curve modeling the synthetic UNI contrast as a function of T 1 (Figure S2).

Synthetic imaging processing and online reconstruction
The core idea of T 1 -based synthetic imaging, as described by Massire et al, 26 consists in creating additional mathematical relationships between T 1 values and the desired synthetic UNI signal.It relies on the simplified forward solution of the MP2RAGE UNI signal equation (see appendix 2 of Marques et al. 1 ) based on (1) an already acquired T 1 map and (2) by retrospectively changing the MP2RAGE sequence key parameters.Indeed, as a reminder, UNI signal is only a function of T 1 and the following MP2RAGE parameters: TI 1 , TI 2 , α 1 , α 2 , n, ES, and TR (assuming perfect inversion).Consequently, by changing any of these parameters, the nonlinear dynamic of signal with respect to T 1 values can be fully controlled to tune the tissue contrast (i.e., nulled, compressed, extended, or Abbreviations: ES, echo spacing (turbo-FLASH TR); n, number of partition-encoding steps before/after the k-space center; sEDGE, synthetic edge-enhancing gradient echo; sFLAWS HCO , synthetic fluid and white-matter suppression-high contrast opposite; sFLAWS MIN , synthetic fluid and white-matter suppression-minimum intensity contrast; sFGATIR, synthetic fast gray-matter acquisition T 1 inversion recovery.a For sFLAWS MIN , two synthetic parameterizations were used (one being sFGATIR) and subsequently combined to generate a minimum intensity projection contrast.
b Sequence parameters were taken from the literature 1,20 and used in the reintegration of FLAWS 1&2 signal equations.The sFLAWS HCO ratio is then computed with the formula written in the table.
highlighted) and generate synthetic T 1 -weighted images according to any specific clinical needs.Interestingly, any choice of sequence parameter is theoretically possible, including those not achievable experimentally (such as with gradient constraints).In addition, this approach offers more flexibility compared with simulated inversion contrasts (which simulate exponential relaxation based on a T 1 map), because it involves all MP2RAGE sequence parameters (and not only the TIs) and it keeps the "uniform" signal distribution.The synthetic MP2RAGE parameters were chosen empirically, or based on the literature, to generate five different synthetic contrasts with uniform signal intensity (Table 1).The design of desired relationships between the synthetic output images and T 1 values was prototyped on MATLAB using the code previously described by Massire et al. 26 Online synthetic reconstruction was then directly implemented on the MR system in a custom sequence reconstruction pipeline (ICE language).The different steps implemented were virtual coil combination, 29 UNI ratio computation, T 1 mapping capabilities, and synthetic reconstructions of all contrasts.The following contrasts were generated (Figure 1B): • sUNI denotes "synthetic UNI" (i.e., the "canonical UNI" T 1 -weighted contrast with a high contrast between GM-WM and CSF-GM as described in the original MP2RAGE work 1 ).This choice was made because of its consensual use in the literature, notably for brain segmentation 3 and for delineation of (juxta)cortical MS lesions. 5,6,8sFGATIR denotes "synthetic FGATIR" and consists of nulling the WM signal.
• sEDGE denotes "synthetic EDGE" and consists of nulling voxels that contain about 50%/50% of GM/WM to highlight cortex boundaries.
• sFLAWS MIN denotes "synthetic FLAWS MIN " and consists of nulling voxels that primarily contain either WM or CSF, to highlight the cortex.
• sFLAWS HCO denotes "synthetic FLAWS HCO " and was notably generated as a sanity check to evaluate how this contrast compares with UNI.

Methodological validation: MRI on healthy volunteers
Experiments were performed on 4 healthy volunteers (3 males/1 female, mean age 34.5 ± 13 [23-53] years) using a whole-body, investigative, 7T MR system (Magnetom 7 T; Siemens Healthcare, Erlangen, Germany), equipped with an 8-transmit/32-receive head coil (Nova Medical, Wilmington, MA, USA).This study was approved by the local institutional review board (ID: 2018-A01761-54), and the volunteers provided informed written consent before examinations.Acquisitions were run within the manufacturer's so-called protected mode (i.e., with a peak power limit per channel of 540 W, average power limits of 1.5 W per channel, and 8 W total at the coil plug).Universal pulses 28 using a GRAPE parameterization 30 were used for all the MP2RAGE acquisitions (for both inversion and excitation RF pulses), allowing calibration-free pTx imaging.The MR sequence was an in-house sequence playing universal RF pulses with a custom image reconstruction.The MP2RAGE protocols covering the whole brain (Table 2) were acquired with four different spatial resolutions: • "Clinical" (1 × 1 × 1 mm 3 ) and "clinical research" (0.8 × 0.8 ×0.8 mm 3 ) spatial resolutions, to illustrate sizeable acquisition-time reduction while ensuring equivalent image quality.

F I G U R E 1
Simulated relationship between UNI signal and T 1 values.(A) Evaluated T 1 values (output) obtained based on UNI signal (input) for conventional (black) and time-efficient (pink) MP2RAGE parameterizations acquired on healthy volunteers at three different spatial resolutions.Sequence parameters are found in Table 2.Note how all curves monotonously sample the brain T 1 range.(B) Generated synthetic UNI signal intensities (output) versus input T 1 values for synthetic MP2RAGE protocols.These synthetic contrasts curves are obtained by reintegrating the MP2RAGE signal equation with the parameters chosen in Table 1.sEDGE, synthetic edge-enhancing gradient echo; sFGATIR, synthetic fast gray-matter acquisition T 1 inversion recovery; sFLAWS HCO , synthetic fluid and white-matter suppression-high contrast opposite; sFLAWS MIN , synthetic fluid and white-matter suppression-minimum intensity contrast; sUNI, synthetic UNI contrast.

T A B L E 2
Sequence parameters used on healthy volunteers.• "Research" high spatial resolution (0.65 × 0.65 × 0.65 mm 3 ), to illustrate moderate acquisition time reduction in addition to spatial-resolution improvement through partial Fourier removal.
The CNRs between WM/GM, WM/CSF, and GM/CSF were measured by drawing, in a reproducible manner (through image coregistration), regions of interest (ROIs) in WM (semioval centers and temporal WM, 4 ROIs per subject), GM (bilateral frontoparietal cortex, 8 ROIs per subject), and CSF (lateral ventricles, 4 ROIs per subject) for all data sets on selected conventional and synthetic UNI contrasts, using the following equation: ) , where A and B correspond to signal intensity values and σ corresponds to respective SDs.Image quality of acquired and synthetic volumes was assessed by a single neuroradiologist with 9 years of experience, notably looking at the delineation of subtle anatomical structures (deep gray nucleus, claustrum).

Clinical validation: MRI on MS patients
To evaluate the applicability of accelerated MP2RAGE imaging associated with synthetic contrast imaging, a small cohort of MS patients was included in the same investigative 7T MR system (also with pTx-protected mode).This study was approved by the local institutional review board (IDs: 2017-A02703-50 and RCB 2019-A01931-56), and patients provided informed written consent before the examinations.Based on a conventional MP2RAGE acquisition at 0.8 × 0.8 × 0.8 mm 3 (11 min 36 s duration), a time-efficient protocol with matching spatial coverage and comparable "time budget" of about 10 min was designed by adjusting TI 1&2 , ES and TR, as described in Section 2.1.FAs were then optimized as described in Section 2.2.A 0.67 × 0.67 × 0.67 mm 3 spatial resolution protocol with an expected higher diagnosis accuracy and a duration of 9 min 50 s was finally obtained (Table 3).Image analysis was based on two criteria: • Lesion depiction was evaluated by an experimented neuroradiologist, with an open side-by-side comparison of the two sequences.
• CNRs between (juxta)cortical MS lesion, WM, GM, and CSF were calculated (four measurements per sequence and per subject for WM, GM, and CSF, and from two to four measurements per sequence and per subject for (juxta)cortical lesions, depending on the patients' lesion load).

Simulation results and synthetic contrast generation
Figure 1A shows the relationships between T 1 values and UNI signal intensities for conventional (black) and time-efficient (pink) MP2RAGE protocols, at all considered spatial resolutions.All these functions are monotonous for the brain and CSF T 1 range, granting reliable T 1 mapping.Figure 1B shows signal dynamics versus T 1 values for all synthetic contrasts: sUNI, sFGATIR, sEDGE, sFLAWS MIN , and sFLAWS HCO .The corresponding parameters for these synthetic MP2RAGE protocols are listed in Table 1.An example of FA optimization results is provided in Figure S1.Simulations of the effect of B + 1 inhomogeneities on CNR WM/GM on the UNI synthetic contrast with time-efficient parametrization are shown in Figure S2B.In the worst-case scenario, the contrast ratio between WM and GM could decrease of up to 20% in the brain regions most affected by B + 1 inhomogeneities.Finally, a graphical summary of the synthetic contrast generation workflow is illustrated in Figure 2.

In vivo scans on healthy volunteers
Two subjects were scanned with both conventional and time-efficient 1 × 1 × 1 mm 3 MP2RAGE protocols (Table S1).Time-efficient MP2RAGE combined with synthetic T 1 -weighted imaging allowed a 40% reduction of the acquisition time (5 min 06 s vs. 8 min 30 s), while preserving the original visual aspect of brain-tissue contrast (Figure 3).Measured CNR WM/GM values of sUNI and conventional UNI were relatively similar: 6.4 ± 0.8 versus 6.6 ± 0.4 (mean values ± SD), respectively (see Table 4 for all CNR results).

F I G U R E 2
Synthetic T 1 -weighted imaging using time-efficient MP2RAGE parameterization workflow.(A) MP2RAGE sequence parameterization is chosen to maximize white-matter/gray-matter contrast-to-noise ratio (CNR WM/GM ), while keeping the acquisition at the minimum for a given spatial resolution.Consequently, gradient-recalled echo (GRE TI1 ) and GRE TI2 are not used for radiological analysis.(B) Resulting UNI image is generated with standard MP2RAGE signal equation.As the brain-tissue contrast might be suboptimal, this volume is not used for radiological analysis but only to generate the quantitative T 1 map.(C) Signal equation is re-integrated using the T 1 map to generate on-demand, clinically relevant, synthetic contrasts to be used for radiological analysis.sEDGE, synthetic edge-enhancing gradient echo; sFGATIR, synthetic fast gray-matter acquisition T 1 inversion recovery; sFLAWS HCO , synthetic fluid and white-matter suppression-high contrast opposite; sFLAWS MIN , synthetic fluid and white-matter suppression-minimum intensity contrast; sUNI, synthetic UNI contrast.
Three subjects were scanned with both conventional and time-efficient (0.80 × 0.80 × 0.80 mm 3 ) MP2RAGE protocols, corresponding to a 30% reduction of the acquisition time (6 min 32 s vs. 9 min 20 s), while preserving the original visual aspect of brain-tissue contrast (Figure 3).Measured CNR WM/GM values of sUNI and conventional UNI were relatively similar: 7.5 ± 0.6 versus 8.4 ± 2.6, respectively (Table 4).

F I G U R E 3
Comparison of synthetic UNI and conventional UNI contrasts.(A) Synthetic UNI contrast (sUNI) obtained from a time-efficient (1 mm) 3 MP2RAGE acquisition (5 min 06 s).(B) UNI contrast obtained from a conventional (1 mm) 3 MP2RAGE acquisition (8 min 30 s). (C) sUNI contrast obtained from a time-efficient (0.8 mm) 3 MP2RAGE acquisition (6 min 32 s).(D) UNI contrast obtained from a conventional (0.8 mm) 3 MP2RAGE acquisition (9 min 20 s).High similarity between conventional and synthetic contrasts could be observed at both resolutions.(E) sUNI contrast obtained from a time-efficient (0.65 mm) 3 MP2RAGE acquisition (9 min 09 s).(F) UNI contrast obtained from a conventional (0.65 mm) 3 MP2RAGE acquisition (11 min 10 s).The difference in true spatial resolution (due to partial Fourier removal) is well visible on zoomed images; arrows show the improved delineation of striatal gray matter on the proposed time-efficient MP2RAGE imaging.Although the decrease in contrast-to-noise ratio on the 9 min 09 s acquisition is noticeable compared with the 11 min 10 s acquisition, the image remains perfectly informative.
Three subjects were scanned with both conventional and time-efficient (0.65 × 0.65 × 0.65 mm 3 ) MP2RAGE protocols, corresponding to a 19% reduction of the acquisition time (9 min 02 s vs. 11 min 10 s).A 0.65 × 0.65 × 0.65 mm 3 conventional MP2RAGE without 6/8 partial Fourier in the partition-encoding direction was not technically feasible due to TI constraints.On the other hand, removing partial Fourier in the time-efficient protocol allowed a narrower point spread function, resulting in a higher effective spatial resolution and subsequent improved delineation of small anatomical details, while maintaining a desired T 1 -weighted contrast and high CNR WM/GM (Figure 3 and Table 4).A homogeneous signal within the brain was observed, notably in the temporal lobes and the cerebellum (Figure S4).
As a proof of concept, 1 healthy volunteer was scanned with a whole-brain (0.45 × 0.45 × 0.45 mm 3 ) MP2RAGE protocol, with an acquisition time of 19 min 56 s using the proposed time-efficient parameterization.No impactful motion artifacts were reported.Ultrahigh spatial resolution allowed the visualization of very subtle anatomical structures such as Virchow-Robin spaces, cerebellar cortex gyration, and claustrum (Figure 4).

T A B L E 4
Contrast-to-noise ratio measurements on healthy volunteers and multiple sclerosis subjects.

F I G U R E 4
Orthogonal views of whole-brain (0.45 mm) 3 acquisition with acquired T 1 map (A) and synthetic UNI contrast (B) obtained from the proposed MP2RAGE time-efficient parameterization.High spatial resolution can be well perceived by zooming in (images at the bottom), allowing us to visualize with precision fine Virchow spaces (small arrows), the details of gyration of the cerebellar cortex (full arrowheads), and the claustrum (full arrows).Acquisition time was 19 min 56 s with a GRAPPA factor of 2.

F I G U R E 5
Comparison of time-efficient protocols with synthetic imaging and conventional MP2RAGE for multiple sclerosis (MS) ([juxta]cortical lesions delineation).Examples of (0.67 mm) 3  Finally, highly accelerated GRAPPA conventional protocols exhibited aliasing artifacts, for an acquisition time similar to the proposed optimized parameterizations, highlighting that this reduction of acquisition time is fully independent of parallel-imaging acceleration methods (Figure S3).All sequences were effectively run under the protected-mode specific absorption rate limits of the MR system.

F I G U R E 6
Examples of on-demand synthetic T 1 -weighted contrasts on a patient with multiple sclerosis (MS).Synthetic UNI contrast (sUNI), fluid and white-matter suppression-high contrast opposite (sFLAWS HCO ), and fluid and white-matter suppression-minimum intensity contrast (sFLAWS MIN ) contrasts were obtained from a (0.67 mm) 3 time-efficient MP2RAGE on a MS subject.A (0.80 mm) 3 T 2 -weighted double-inversion-recovery (DIR) image is also provided on the right-hand side.The top row focuses on a leukocortical temporal MS lesion.Both white matter (WM; long arrow) and gray matter (GM; short arrow) components of the lesion are clearly visible on sUNI and sFLAWS HCO contrasts.Arrowheads on sFLAWS MIN show the hyperintense "halo" around the leukocortical lesion, already reported in the literature.Contrary to sFLAWS MIN , DIR shows an entirely hyperintense leukocortical lesion (big arrow).The bottom row focuses on a small right thalamic MS lesion.Although highly visible on sUNI and sFLAWS HCO , this lesion is extremely hard to distinguish in sFLAWS MIN and DIR.

In vivo scans on MS patients
Six MS patients (3 males/3 females, mean age of 40.0 ± 7 [33-52] years, mean EDSS of 5.3 ± 9.8 [2-6], mean disease duration of 12.6 ± 8.6 [4.0-24.0]years) were prospectively scanned with both conventional and time-efficient MP2RAGE protocols.The choice of sequence parameters for the time-efficient protocol led to a 15.2% acquisition time savings, associated with a 41% reduction in voxel size.
Visual analysis showed a high similarity of lesion-to-tissue apparent contrasts between the two protocols (conventional MP2RAGE UNI and time-efficient with sUNI; Figure 5).This analysis was supported by CNR measurements: CNR differences below 15% were reported between the two protocols (Table 4), with a remarkable similarity for the CNR between (juxta)cortical MS lesion and CSF (2.4 vs. 2.4).Interestingly, the gain in spatial resolution led to improved delineation of small MS lesions close to the cortex: (juxta)cortical lesions were either as clear or better delineated, especially in their intracortical component with the time-efficient protocol (Figure 5).
Finally, Figure 6 illustrates synthetic contrasts that were generated online, with a focus on a typical leukocortical MS lesion.The clear lesion delineation with sUNI and sFLAWS HCO can be noticed, with high visibility of the WM but also GM involvement.Synthetic FLAWS MIN contrast provides the same peripheral hyperintense "halo 31 " surrounding the MS lesions, as described in typical FLAWS MIN imaging.

DISCUSSION AND CONCLUSION
Since its inception in 2009, 1 the MP2RAGE sequence has been of great interest for the UHF MR community, thanks to its ability to provide uniform T 1 -weighted contrast.Over the years, numerous parameterizations, including FLAWS, have been proposed 17 and optimized, [18][19][20]25 to offer more clinically relevant contrasts from this sequence. In his work, T 1 -based synthetic imaging framework 26 was used to achieve this "single acquisition-multiple contrasts" objective, while also fundamentally changing the way the MP2RAGE sequence is parameterized.By decoupling the final contrasts from the sequence parameter choice, a time-efficient parameterization (i.e., with minimal idle time) could be set to obtain sizeable reduction of the acquisition time or alternatively spatial resolution increase for a given time budget.The hereby proposed approach has the advantage of "unifying" MP2RAGE imaging, without facing the inevitable tradeoffs encountered by other optimizations.For instance, to generate clinically relevant contrasts, FLAWS imaging requires TI 2 to be very short and close to TI 1 .20 As a consequence, there is not enough time to fit the RAGE trains between the two TIs for high-spatial-resolution FLAWS acquisitions (unless it is segmented, leading to even longer acquisition times). Th limitation holds true even for the recently described FLAWS/MP2RAGE parameterizations, 25 which would also become unfeasible at a given high spatial resolution (i.e., < 0.65 mm 3 isotropic).With the proposed approach, this upper limit is removed, as a whole-brain (0.45 mm) 3 imaging was successfully performed in 20 min (Figure 4).Achieving this level of spatial resolution was until now primarily possible by performing several slabs (e.g., six slabs of 10 min each leading to a 60-min whole-brain acquisition with a [0.35 mm] 3 resolution 32 ).Note that using higher-performance gradients would naturally push this high spatial resolution limit further for all parameterizations, with a retained advantage for the time-efficient, free-TI choice approach.Finally, although increased flexibility in the choice of MP2RAGE parameter is possible with this approach, this choice is still fully relevant to determine the signal-to-noise level of acquired images, which the subsequent T 1 estimation relies on.This consideration is discussed as a limitation later in the manuscript.
The efficiency gain of the proposed approach has been demonstrated at various spatial resolutions, allowing sizeable reduction of acquisition time or reinvestment of the time for removal of partial Fourier to obtain a finer point spread function in the case of the (0.65 mm) 3 protocol (Figure 3).As a side demonstration, it was shown that the decrease in acquisition time provided by this time-efficient approach is independent from parallel-imaging techniques and could not be readily gained by increasing the GRAPPA factor to high values (Figure S3).Nevertheless, because this "time efficient" MP2RAGE parameterization comes without compromising image quality, it could likely be combined with other advanced reconstruction techniques, such as compressed sensing 33 or deep learning, 34 to undersample the k-space and further reduce acquisition times.This is an opportunity for future work, such as using the SPARKLING 35 technique.
During the optimization process, the CNR WM/GM metric on UNI images was considered to be the most relevant, as it directly translates how accurate the MP2RAGE acquisition is to sample T 1 values in this WM-GM range.However, maximizing CNR WM/GM to an extreme level is not recommended, as it would automatically minimize CNR GM/CSF .A minimal contrast should be ensured to enable proper visualization of GM close to CSF.This tradeoff has so far been heuristically found (Figure S1).Another instrumental condition is the bijective mathematical relationship between T 1 values and the acquired UNI signal (a non-monotonous relationship, or even a first-derivative tending to infinite, would be extremely detrimental to precision in the corresponding T 1 range).For this reason, the FA combination showing the highest CNR WM/GM was not always chosen (see Figure S1).Interestingly, the measured CNR WM/GM in synthetic UNI contrasts was comparable to the conventional UNI contrasts, inspired by Marques et al. 1 Several considerations could be derived from these experimental measurements.First, while the choice of TI might be suboptimal for CNR, this is mitigated to a certain extent by the optimization of the FA; additionally, this showed that synthetic UNI retains a similar CNR level, presumably because the T 1 estimation is robust even with time-efficient parameterization, limiting its susceptibility to noise.In any case, this is in line with the original objective of this work, to provide the canonical UNI contrast, which is confirmed by subjective visualization of images (sUNI and UNI are extremely similar; see Figures 3 and 5).
For imaging evaluations on MS patients, we focused on lesions involving the cortex (referred to here as [juxta]cortical lesions), because these lesions are still a major challenge for clinical MR practice 36 but are of key significance for the diagnosis 37 and prognosis 38 of the disease.Preliminary results from the prospective exams on pathological brains showed no clinical information loss and strong potential, as higher-spatial-resolution acquisitions can be carried for the same time budget.This opens the way to more accessible use of high-spatial-resolution 3D T 1 -weighted imaging, not solely in the research context but also for clinical examinations, with the aim of providing notable comfort to the radiologist to evaluate the lesion load.Additionally, because there is only an abacus between T 1 map and synthetic contrasts, all synthetic imaging was reconstructed instantly at the MR system console using the manufacturer's framework.This is an important asset for radiologist acceptance.
In this work, multiple T 1 -based synthetic contrasts were generated, each of which elaborated to match all MP2RAGE-derived contrasts described in the literature for the past decade.To do this, MP2RAGE sequence parameters were heuristically adjusted by looking at the brain-tissue signal as a function of T 1 .All synthetic contrasts were calculated based on the acquired T 1 map but highlighted different brain tissues and could be used for a given clinical indication (e.g., the "WM-nulled MPRAGE" optimized for intrathalamic nuclei delineation 39 ).More generally, nearly infinite T 1 -based contrast possibilities could be directly defined from any mathematical formula based on T 1 .Assessing the relative clinical added value of each contrast with regard to the others is beyond the scope of this study, as it largely depends on the studied pathology.For MS imaging of small (juxta)cortical lesions, however, the radiological analysis conducted in this study concluded that the "canonical" high WM/GM contrast UNI (or sUNI) was very well adapted, as previously described in recent literature. 5,8,9Nevertheless, dedicated studies comparing all different contrasts would certainly be of great interest, to determine more precisely their respective added value in the various clinical contexts.In particular, our results confirm the previously described 31 perilesional halo in FLAWS MIN contrast around MS lesions, a potentially confusing artifact (Figure 6).This finding highlights the significant difference in MS lesion contrast between FLAWS MIN (T 1 -weighted) and DIR (T 2 -weighted).
In this work, parallel transmission combined with universal pulses 28 was used to homogenize the RF inversion and excitation in the whole brain.With this technique, spatial B + 1 inhomogeneities that would locally affect MP2RAGE signal are largely mitigated (see Figure S4 for a focus on the cerebellum and temporal lobes).However, because of their calibration-free nature, UP introduces both local and global bias in T 1 estimation, which eventually affects the contrast between tissues in the framework of the presented time-efficient approach.Based on the reported performance of UP on a database, 40 a CNR WM/GM decrease of up to 20% compared with expected values could be anticipated in some regions of the brain (see Figure S2).These relatively small imperfections could be further minimized by using standardized UPs. 40Another benefit of pTx it that it simplifies the EPG optimization, although a simple loop over FA values 25 could be added for MR systems without pTx.In practice, optimizing a time-efficient parametrization for single-channel MR systems would require fine-tuning FA values to mitigate the T 1 estimation bias caused by large B + 1 inhomogeneities, presumably at the cost of reduced global CNR.An illustration of this heuristic approach for a single-channel 7T system is provided in Figure S1C,D (see blue vs. red curves).
The framework proposed in this work could be readily translatable to 3T imaging by adjusting T 1 values to the ones of brain tissues and CSF at this magnetic field during the FA simulation and synthetic contrast generation steps.][43] Finally, it is important to emphasize that imaging with contrast agent injection was not considered in the design of this study, as T 1 shortening induced by gadolinium would dramatically affect the sequence optimization.MP2RAGE shares the same characteristics as MPRAGE (including flow sensitivity and high tissue contrast), making it, to date, not recommended for postinjection brain imaging 44 compared with 3D T 1 -weighted turbo spin echo.
Considering that all synthetic contrasts were computed from a T 1 map, this formalism is not limited to MP2RAGE and is in theory compatible with all T 1 mapping techniques, including double-angle techniques such as variable flip angle. 45A dedicated work to study the efficiency of T 1 mapping per unit of time could be imagined to determine what the optimal acquisition scheme would be.Nevertheless, this study purposely focused on the MP2RAGE sequence because of its recognized added value in the current recommendations of medical societies (notably regarding epilepsy 46 and MS 36 ).As a side note, although EPG was used in this study to optimize the MP2RAGE parametrization, a pure spoiled longitudinal magnetization model could have been used as an alternative.
Because T 1 mapping is used to generate synthetic contrasts, it is relevant to study how shortening MP2RAGE acquisitions affects T 1 mapping accuracy and precision.For example, reducing TIs and TR is likely unfavorable for the mapping of long T 1 , such as in the CSF (although measuring T 1 is already quite challenging in the CSF due to flow).Only small differences have been observed between conventional and time-efficient T 1 maps; however, the results presented in this study pertain only to the range of described spatial resolutions and tissue T 1 values.Caution should be exercised in the case of low-resolution, time-efficient protocols, which lead to very short TIs and result in possibly inaccurate T 1 estimation.A dedicated study with T 1 phantom experiments and possibly multicentric data would be needed to quantitatively evaluate how T 1 accuracy and precision are affected.
Exploiting the T 1 map provided by a time-efficient parametrization alone is therefore not recommended at this point.Nevertheless, because final synthetic contrasts are T 1 -weighted, the T 1 bias will likely have limited effect on image quality and interpretation (see Figure S2) except for very specific contrast (sEDGE), which relies on canceling the signal at tissue interface, where moderate T 1 variations could cascade in significant local contrast changes.
Another source of bias in T 1 mapping is the magnetization-transfer effect, 47 which was not considered in pTx pulse design.In some images, and particularly at 1 × 1 × 1 mm 3 resolution (Figure 3), a slight right-left contrast asymmetry with localized shading was sometimes observed, with both MP2RAGE parameterizations.It has been recently shown that homogenizing B 1rms in the RF pulse design is critical to mitigate the magnetization-transfer effect in the brain.UP considering B 1rms homogenization 48 in addition to FA could be designed in the future.
The relatively small MS patient cohort and the difference in spatial resolution (0.65 × 0.65 × 0.65 vs. 0.80 × 0.80 × 0.80 mm 3 ) were not readily adapted to a strict double-blinded radiological analysis comparing time-efficient and conventional MP2RAGE.Another time-efficient acquisition at the same spatial resolution would have been necessary for such a comparison.Nevertheless, based on the results obtained on healthy volunteers, a high-resolution acquisition was chosen to foster lesion evaluation comfort for the radiologist.Finally, the fact that the synthetic contrast is fixed beforehand is both a strength (for consistency) and a weakness (no flexibility).The ability to "slide" a synthetic contrast would be an interesting future development.
In conclusion, a time-efficient MP2RAGE parameterization combined with synthetic T 1 -based imaging could pave the way for fast, high-resolution, multicontrast T 1 -weighted imaging, in clinical routine use.parallel-imaging artifacts are well visible in the whole brain (arrows).For comparison, time-efficient MP2RAGE with synthetic UNI contrasts is provided ([1 mm] 3 isotropic resolution, TA = 5 min 06 s [C] and [0.80 mm] 3 isotropic resolution, TA = 6 min 32 s [D]).Figure S4.Orthogonal views of (0.65 mm) 3 synthetic and conventional UNI images on a healthy volunteer.Synthetic UNI is obtained from a time-efficient MP2RAGE acquisition; UNI is obtained from a conventional MP2RAGE acquisition (see Table 2 for sequence parameters).This figure highlights the efficiency of parallel transmission (pTx) to limit B + 1 bias notably in the cerebellum and temporal lobes.
synthetic UNI contrast (sUNI) imaging obtained from the proposed MP2RAGE parameterization (A) and (0.80 mm) 3 UNI contrast obtained from conventional MP2RAGE (B) on four different MS subjects with comparable time-budget acquisitions.Acquisition times were 9 min 50 (A) and 11 min 36 (B).For each subject, zoomed-in (juxta)cortical lesions are shown underneath.Long arrows indicate better delineation of a millimetric intracortical lesion with the time-efficient protocol.Short arrows and arrowheads indicate better delineation of the intracortical component of two leuko-cortical MS lesions, which would have been classified as "juxtacortical" with conventional MPR2AGE.The circle highlights a small leukocortical lesion, which is hard to see with the conventional protocol.

65 mm) 3b MP2RAGE parameterization Conventional Time-efficient Conventional Time-efficient Conventional Time-efficient
Abbreviations: CNR, contrast-to-noise ratio; GM, gray matter; sUNI, synthetic UNI contrast; WM, white matter.a Measured on 2 subjects.b Measured on 3 subjects.c Measured on 5 patients, corresponding to a total of 16 (juxta)cortical multiple sclerosis lesions (no [juxta]cortical lesions for 1 patient).