Cerebrospinal fluid–suppressed T₂‐weighted MR imaging at 7 T for human brain

Abstract

Purpose

T₂‐weighted lesional imaging is most commonly performed using inversion recovery turbo spin echoes. At 7 T, however, this acquisition is limited for specific absorption rate and resolution. This work describes and implements a method to generate CSF‐suppressed T₂‐weighted imaging.

Methods

The strategy uses a driven equilibrium spin‐echo preparation within an inversion recovery with multiple 3D gradient‐echo imaging blocks. Images are combined using the self‐normalization approach, which achieves CSF suppression through optimized timing of individual blocks and minimizes sources of variation due to coil receptivity, T₂ ^*, and proton density. Simulations of the magnetization‐prepared fluid‐attenuated inversion recovery gradient‐echo (MPFLAGRE) method over T₁ and T₂ relaxation values are performed, and in vivo demonstrations using an 8 $\times$ 2 transceiver array in healthy controls are shown.

Results

The specific absorption rate of the calculated MPFLAGRE sequence is 11.1 ± 0.5 W (n = 5 volunteers), which is 74 ± 2% of the US Food and Drug Administration guidelines. This method acquires both contrasts for CSF suppression with detection of long T₂ components and T₂‐weighted imaging in a single acquisition. In healthy controls, the former contrast generates increased signal in the cortical rim and ependyma. A comparison is shown with a conventional 3D SPACE fluid‐attenuated inversion recovery acquisition, and sensitivity to pathology is demonstrated in an epilepsy patient.

Conclusion

As applied with the 8 $\times$ 2 transceiver, the MPFLAGRE sequence generates both whole‐brain contrast suitable for lesional and T₂‐weighted imaging at 7 T in fewer than 10 minutes within the US Food and Drug Administration's specific absorption rate guidelines.

1. INTRODUCTION

The increase in SNR at 7 T relative to 3 T has been used by numerous groups for higher‐accuracy physiological and metabolic MRI.1, 2, 3 However, although high‐resolution T₁‐weighted structural imaging has been applied quickly,4, 5 T₂‐weighted and fluid attenuated inversion recovery (FLAIR) imaging at 7 T has been less implemented—most likely because of the increased power deposition and poor transmit homogeneity due to the need for multiple spin echoes. Parallel transmit arrays have mitigated inhomogeneity problems; for example, RF shimming and a dual‐row transceiver array can achieve 15% SD B₁ ⁺ over the human brain (Supporting Information Figures S1 and S2).6 However, given the squared dependence of voltage with frequency at 7 T, 7 turbo spin‐echo acquisitions or the use of adiabatic pulses still result in high specific absorption rate (SAR), such that T₂‐weighted imaging covering the entire brain in acceptable scan times remains challenging. The variable flip angle turbo spin‐echo sequence provides a reasonable solution to control power deposition; however, it is sensitive to B₁ ⁺ inhomogeneity,8, 9, 10, 11 with several groups proposing design of parallel transmit pulses as a function of k‐space position to generate more homogeneous excitation and inversion profiles. With this approach for turbo spin echo, however, achieving consistent contrast and motion insensitivity can be problematic by virtue of the dependence on accurate B₁ ⁺ maps and applied B₁ variation between k‐points.8, 12 Alternatively, to altogether avoid multiple refocusing pulses, a driven equilibrium spin‐echo strategy to longitudinally encode T₂ contrast (i.e., T₂ preparation) has been useful.13, 14, 15 To reduce the sensitivity of the acquisition to B₁ ⁺ inhomogeneity at 7 T, Dyvorne and Balchandani16 implemented the T₂ preparation with an adiabatic spin‐echo module in a multislab approach to generate excellent T₂‐weighted whole‐brain coverage at 0.8‐mm³ isotropic resolution (approximate 5.5‐minute acquisition). The FLAIR imaging at 7 T, however, remains challenging.

To suppress the CSF signal, the now commonly used calculated MP2RAGE images from Marques17 provides an insightful T₁‐based approach. This self‐correcting normalization combines 2 images acquired at different delays after an inversion to give a calculated signal within the range of [$-$0.5, 0.5]. With a reconstruction in the complex domain (Eq. (1)),17 the calculated signal retains sensitivity to the inversion recovery, while eliminating B₀‐based T₂ ^* phase effects and correcting for common factors of M₀ and B₁ ^‐. Inspection of this reconstruction shows that the largest signal (+0.5) is seen when the intensities between the 2 images are equal, and the smallest signal ($-$0.5) when the intensities are equal and inverted in sign. For CSF, the calculated signal returns in the negative range ($-$0.5 to 0), whereas the white matter (WM) and gray matter (GM) return in the positive range (0 to +0.5). In the MP2RAGE acquisition, the sign and timings of the individual imaging readouts are optimized to generate a suppressed CSF intensity of $-$0.5 and high contrast between WM and GM. The equation of self‐normalization reconstruction is

$$R=real\left(\frac{S_1^{\ast }{S}_2}{{\left|S_1\right|}^2+{\left|S_2\right|}^2}\right)$$ (1)

where S₁ and S₂ are the GRE signal at inversion recovery delays TI1 and TI2, respectively; and the “*” operator is a complex conjugator.

As discussed by O'Brien,18 the MP2RAGE reconstruction provides a much flatter T₁‐weighted image (than the non‐reference‐corrected MPRAGE), which enables high‐consistency gray‐white‐CSF segmentation. More importantly, however, it is recognized that the self‐correcting strategy can be applied to other acquisitions that acquire multiple readout blocks with a common preparation sequence.

In this report we describe the incorporation of a longitudinal T₂ preparation module with a multiblock inversion recovery 3D acquisition to achieve a CSF‐suppressed T₂‐weighted image for 7T use in the detection of brain pathology. Our preliminary data demonstrate that the proposed sequence MPFLAGRE (magnetization‐prepared fluid attenuated gradient echo) achieves this goal. The sequence uses an inversion recovery with a T₂ spin‐echo preparation and multiple short gradient‐echo imaging blocks, with the self‐correcting normalization17 generating the calculated MPFLAGRE image. Cerebrospinal fluid suppression is achieved through appropriate timing of TI for T₁ weighting. In this report, we show Bloch simulations to examine the sequence's dependence on T₁ and T₂ and demonstrate its performance with a transceiver array at 7 T. Because of the longitudinal T₂ preparation that is performed with conventional adiabatic refocusing pulses, this sequence efficiently generates whole‐brain T₁ and T₂‐weighted coverage in single sequence that is well within SAR guidelines.

2. THEORY

2.1. Two‐block sequence (MPFLAGRE‐2)

To generate T₂ FLAIR contrast, our method introduces T₂ weighting into the T₁‐weighted MP2RAGE sequence using a longitudinal T₂ encoding module performed after the initial inversion (Figure 1). The T₂ weighting is performed with a nonselective spin‐echo module (90_x+‐180_y ‐180_y ‐90_x‐ of duration TE) and is followed by multiple short 3D gradient‐echo (readout) blocks. The MPFLAGRE‐2 uses 2 blocks, with the first block performed immediately after the spin echo, block S₁. Another signal block, S₂, is performed at another timepoint in the T₁ recovery, which with increasing delay from the S₁ block, reflects primarily T₁ weighting. Equations (2) and (3) give the expressions for the signal for the 2‐block sequence:

$$S_1\propto M_0{e}^{{-te}_g/T_2^{\ast }}{B}_1^{-}sin{\alpha }_1\left[\left(\frac{e^{\frac{-te}{T2}+\frac{-ti1}{T1}}\ast {\epsilon M}_{\mathit{ss}}}{M0}+\left(1-e^{\frac{-ti1}{T1}}\right)\right)\ast A_1^{\mathit{Ncenter}}+R_1^{\mathit{Ncenter}}\right]$$ (2)

Figure 1.

Pulse sequence showing the overall sequence (A), T₂ spin‐echo module (B), and readout block (C). The MPFLAGRE (magnetization‐prepared fluid‐attenuated inversion recovery gradient‐echo) sequence is able to run with 2, 3, or 4 spoiled gradient blocks. In (C), the readout block is shown either as a centric out (S₁) or a linearly (S₂) encoded coverage

$$S_2\propto M_0{e^{{-te}_g/T_2^{\ast }}B}_1^{-}sin{\alpha }_2\left\{\begin{array}{c}\left[\left(1-e^{\frac{-ti2}{T1}}+e^{\frac{-ti2}{T1}}{R}_1^N\right)+e^{\frac{-te}{T2}+\frac{-ti2}{\mathit{Ti}}}\ast A_1^N\left(1-e^{\frac{-ti1}{T1}}-\frac{M_{\mathit{ss}}}{M0}\epsilon \ast e^{\frac{-ti1}{T1}}\right)\right]\\ \ast A_2^{\mathit{Ncenter}}+R_2^{\mathit{Ncenter}}\end{array}\right\}$$ (3)

where $R_{1or2}^{\mathit{Npts}}=\left(1-e^{\frac{-tr}{T1}}\right)\frac{1-{\left({cos({\alpha }_{1or2})e}^{-trT1}\right)}^{\mathit{Npts}}}{1-{cos({\alpha }_{1or2})e}^{-trT1}}$; and $A_{1or2}^{\mathit{Npts}}={\left(cos\left({\alpha }_{1or2}\right)e^{-trT1}\right)}^{\mathit{Npts}}$,

where tr is the echo spacing; ε is efficiency of inversion; N is the resolution; ti1 is the delay for block 1; ti2 is the succeeding delay for block 2; te_g is the TE for a single gradient echo; M_ss is the steady‐state magnetization at the start of each inversion recovery; and M₀ is the total available magnetization signal. Inspection of Eqs. (2) and (3) shows that, as expected, S₁ is weighted strongly by T₂ (center term with $e^{-te/T2}$; with a very short ti1, there is minimal T₁ recovery). Whereas S₂ demonstrates the T₁ recovery due to ti2 (center term $1-e^{-ti2/T1}$), the T₂ effect dissipates with increasing ti2. To combine these images, we use the self‐correcting normalization, $R_{1/2}$ as described by Marques,17 with the inclusion of sign inversion on the T₂‐weighted image S₁, to generate the desired tissue orientation (Eq. (3)) as follows:

$$R_{1/2}=real\left(\frac{{-S}_1^{\ast }{S}_2}{{\left|S_1\right|}^2+{\left|S_2\right|}^2}\right).$$ (4)

The normalization from Eq. (4) for the 2 signals gives a high 0.5 value when the 2 input intensities, S₁ and S₂, are equal and of opposite sign, and a lesser value when the intensities are unequal. Thus, CSF suppression can be achieved through appropriate timing for the S₂ block to give a CSF signal intensity that is different from the S₁ block. Figure 2 shows Bloch simulations of the 2‐block sequence performed over the T₁, T₂ parameter space. For comparison, simulation of the MP2RAGE is shown in Figure 2A using timings of ti1/ti2/TR = 0.9/2.6/5 seconds and tip angles of 5° and 9°. As expected from the MP2RAGE, the base images (S₁ and S₂) and the calculated intensity R_1/2 (from Eq. (1)) is primarily dependent on the T₁ with minimal T₂ dependence (a small dependence results from the finite duration of the adiabatic inversion pulse).

Figure 2.

Bloch simulations of a 2‐block acquisition, S₁ and S₂. A, Using MP2RAGE parameters (i.e., without a T₂ spin‐echo preparation). B, MPFLAGRE‐2 sequence with T₂ preparation applied early in the inversion recovery. In both (A) and (B), the time course of Iz amplitudes is shown (5 tissue components: blue, CSF; black, gray matter [GM]; red, white matter [WM] with inversion, pathologic T₂). Note that the time axis is not linear, with each unit representing a simulation step. The surface plots of the signal amplitudes are shown over the [T₁, T₂] parameter space for S₁, S₂, and the calculated signal R_1/2 (using Eqs. (1) and (4)). The MP2RAGE sequence (A) shows only the T₁ dependence, whereas the MPFLAGRE‐2 sequence (B) shows the effect of the T₂ preparation and T₁. The calculated R_1/2 shows the CSF suppression and signal detection over the pathologic range of T₂. The black dots on the surface plots indicate the 5 tissue components. The T₁ and T₂ values for CSF are assigned at 4.3 seconds, 0.9 seconds; GM at 2.0 seconds, 60 ms; and WM at 1.2 seconds, 60 ms19, 20, 21

In comparison, Figure 2B shows simulations of the MPFLAGRE‐2 sequence, including the T₂ preparation module applied early in the T₁ recovery and using a parameter set of ti1/ti2/TR/TE = 0.1/1.6/5/0.085 seconds and tip angles of 5° and 9°. The calculated image (Eq. (4)) shows that the CSF signal (T₁, T₂) taken at [4.3, 0.9] seconds19 is partially suppressed with $R_{1/2}^{\mathit{CSF}}$ ~‐0.18, which is substantially lower than normal $R_{1/2}^{\mathit{GM}}$ and $R_{1/2}^{\mathit{WM}}$R_WM. (The T₁, T₂ values for tissue components used in these simulations are CSF = 4.3 seconds, 0.9 seconds; GM = 2.0 seconds, 60 ms; WM = 1.2 seconds, 60 ms.^19‐21) It should be stated that the strategy for MPFLAGRE can also be applied with the spin‐echo application applied late (rather than early) in the T₁ recovery. Although there are differences between the early or late T₂ preparation (Supporting Information Figure S3), the strategy remains similar (i.e., to optimize the timings of the T₂ and T₁ weighted blocks used with the self‐normalization to achieve CSF suppression due to T₁ differences while maintaining T₂ sensitivity). This report focuses on the T₂ weighting applied early in the T₁ recovery.

2.2. Three‐block and 4‐block sequences (MPFLAGRE‐3 and MPFLAGRE‐4)

The use of multiple readout blocks of acquisition in combination with both T₁ and T₂ preparation in the MPFLAGRE sequence provides additional flexibility. Two aspects of the multiple‐block acquisition are considered. First, although the MPFLAGRE‐2 sequence is able to suppress CSF, Figure 2B shows that there is relatively limited dynamic range for the calculated R_1/2 image. To increase this sensitivity with T₂, we recognize that the T₂ preparation induces a change in the longitudinal magnetization and thus the rate of T₁ recovery. Given the typically slow T₁ recovery, the T₂‐dependent effects on amplitude and T₁ recovery rate thus transiently persist after the spin echo. To maximize the effect of the spin echo, we can then acquire an additional signal block (Figures 1 and 3A) and sum the 2 succeeding image blocks, S₁ and S₂. Analytically, this can be expressed as a sum of Eqs. (2) and (3), with simplifications including the small values for ti2, ti1, identical tip angles α for S₁ and S₂ (see Eqs. (2) and (3) for other terms, omitting the B₁ ^‐, tip angles, density, and T₂ ^* factors), and extracting the TE dependent terms:

$$S_1+S_2\propto -e^{-te/T2}{\frac{\epsilon \ast M_{\mathit{ss}}}{M0}{e}^{-ti1/T1}A}^{N/2}\ast \left(1+e^{-ti2/T1}{A}^N\right)+f\left(M_{\mathit{ss}},A^N,R^N\right).$$ (5)

Figure 3.

Bloch simulations of a 4‐block acquisition. A, Time course of Iz amplitudes (5 tissue components: blue, CSF; black, GM; red, WM; GM and WM with long T₂ values for pathology). The [T1, T2] values for CSF are assigned at 4.3 seconds, 0.9 seconds; GM at 2.0 seconds, 60 ms; and WM at 1.2 seconds, 60 ms. B, Signal block 1 intensity is strongly T₂ weighted, shown over the T₁, T₂ parameter space. C, Signal block 2 intensity. D, Summed signal blocks 1+2 show the combined T₁ and T₂ sensitivity. E,F, Signal blocks 3 and 4 are strongly influenced by T₁. All signal intensities are plotted on a scale of [−0.1, 0.1]

With this sign inversion to maintain the desired tissue sensitivity, a third (delayed) block S₃ is then used as the reference image for the self‐correcting normalization according to Eq. (6):

$$R_{(1+2)/3}=real\left(\frac{-{\left(S_1+S_2\right)}^{\ast }{S}_3}{{\left|S_1+S_2\right|}^2+{\left|S_3\right|}^2}\right).$$ (6)

As seen from Eq. (5) and simulated in Figures 3B‐D, the summed block S₁₊₂ increases the weighting of the $-e^{-te/T2}$ term from 1 to $1+e^{-ti2/T1}{A}^N$. Consistent with this, simulation of the summed block strategy shows that the increase in dynamic range for the R_(1+2)/3 image arises from a drop in the calculated signal for normal tissue, whereas the pathologic long T₂ values “saturate” at 0.5 (Figure 4). Thus, the persistence of the spin‐echo effect is a function of T₁ (WM returns more rapidly than GM) (i.e., R_(1+2)/3 depends on the T₁, as $R_{(1+2)/3}^{GM,normal}$ is larger [brighter] than $R_{(1+2)/3}^{WM,normal}$). With the adjacent block sum of S₁ and S₂, given the same numerical T₂ range is present between normal and pathology for WM and GM (approximately 60 ms to 160 ms), the absolute increase in calculated intensity between normal to pathologic T₂ values is similar between WM and GM: $R_{(1+2)/3}^{GM,normal}-R_{(1+2)/3}^{GM,pathol}$ and $R_{(1+2)/3}^{WM,normal}-R_{(1+2)/3}^{WM,pathol}$.

Figure 4.

Bloch simulations for the MPFLAGRE‐4 sequence, adding a delayed S₄ block. See Figure 3A for the time course. Behavior of 5 tissue components identified CSF (blue), GM (black), and WM (red); and GM and WM with long T₂ values for pathology. Calculated R intensities are shown: R_1/3 and R_(1+2)/3 intensities as a function of T₁, T₂ space (A) and plots of signal amplitude from the 5 tissue points (far right). B, The R_1/4 and R_(1+2)/4 intensities. The R_1/3, R_(1+2)/3 amplitudes show CSF suppression and increased amplitudes over long T₂ components. The R_1/4 amplitude shows predominantly T₂ dependence. The R_(1+2)/4 intensity, similar to R_(1+2)/3, shows a larger dynamic range for T₂ dependence

The second aspect of multiple‐block acquisitions arises from the timing and use of the delayed reference image (e.g., through the addition of a fourth imaging block S₄) (Figure 4). Combining the S₁ and S₄ images using Eq. (4) generates a calculated R_1/4 that, by virtue of their common T₁ factors, largely eliminates T₁ dependence to generate a solely T₂‐weighted image (which can be estimated from the simple 2‐block analysis in Eqs. (2) and (3) under the conditions of long ti2 and short ti1). For completeness, we also show the calculated R_(1+2)/4; similar to R_(1+2)/3, this shows an enhanced dynamic range over T₂ but with mild T₁ sensitivity. Figure 4 shows the resulting simulation, including the calculated R_1/3 and R_(1+2)/3 (CSF‐suppressed, T₂‐weighted) and R_1/4 and R_(1+2)/4 (T₂‐weighted) images.

2.3. B₁ ⁺ dependence

Although the self‐correcting normalization eliminates B₁ ^‐ variation, variation in B₁ ⁺ is not wholly corrected. Figure 5 considers the B₁ ⁺ dependence of the sequence when multiple acquisition blocks are used to calculate the signal. The transceiver array exhibits 15% SD over the brain when combined with RF shimming (Supporting Information).6 Simulation results from a MPFLAGRE‐4 sequence are shown over the range of T₂ values, using B₁ ⁺ values at ± 15% and ±30% of the optimum B₁ ⁺ value used (750 Hz). The CSF shows the greatest effect with B₁ ⁺ variation, although the R^CSF intensity remains at or below the normal‐tissue WM intensity R^WM. Normal GM and WM show increased R^GM and R^WM, increasing by less than or equal to 25%. However, it should be noted that in the range of normal T₂ values for GM and WM (40 ms‐70 ms^20,21), the calculated intensity is steeply rising with T₂, and that within a 15% erroneous B₁ ⁺, a less than 10‐ms T₂ increase will give the same R^GM intensity (i.e., the same signal intensity is equivalent to less than a 10‐ms T₂ rise).

Figure 5.

B₁ ⁺ influence on MPFLAGRE‐4 as a function T₂. A fixed T₁ value is used for GM (2.0 seconds, black) and WM (1.2 seconds, red). The CSF is a single value plotted (4.3 seconds T₁, 0.9 seconds T₂, blue). The range of B₁ ⁺ is 0.70 to 1.30, optimum at 1.0. A, Calculated intensity R_1/3. B, Calculated intensity R_(1+2)/3. Consistent with the surface plots in Figure 4, the dynamic range of R_(1+2)/3 is larger than R_1/2 by more than 50%. Based on the transceiver range of ±15% (dashed lines), the B₁ ⁺ sensitivity of GM and WM is within approximately 25% over a T₂ range of 60‐160 ms

2.4. Pathologic T₂ values

As the goal of the FLAIR sequence is to detect tissues with long (and likely pathologic) T₂ values, we need to have the values of pathologic relaxation at 7 T to properly optimize the parameters of the MPFLAGRE sequence. Although such values have not been widely reported, a 3T estimate is available,22 which found a less than 20% increase in hippocampal T₂ in epilepsy versus control. We estimate that to leave adequate “head room” for pathology, the calculated image values for normal GM and WM tissue need to be less than 0.30, with CSF intensities at or below the normal tissue values; these values were achieved in these optimizations.

3. METHODS

We used a Siemens whole‐body 7T Magnetom 8‐channel multiple transmit system with body gradient coil and an 8 $x$ 2 transceiver array for all acquisitions. All studies were approved by the institutional review board. The transceiver array was driven in coil pairs using 8 one‐to‐two splitters such that coils at equivalent azimuthal positions from the 2 rows are driven by the same RF transmit channel, with independent reception from all 16 channels. A fixed phase shift between the 2 rows is used to ensure constructive addition of the RF across rows and to maximize spatial coverage. B₁ shimming was performed in all subjects, requiring about 3.5 minutes (including 2.5 minutes of B₁ mapping acquisitions). In control subjects, a mean B₁ ⁺ of 17.6 uT with a SD of 10.4 ± 1.8% over the brain using a maximum voltage for every RF channel of less than 170 V (n = 8 subjects6; Supporting Information Figures S1 and S2). As a result, conventional pulses are used, including hyperbolic secant pulses with mu = 10 and 4 time constants. With a single T₂ preparation in the MPFLAGRE‐4 acquisition and 2π refocusing pulses, the acquisition has a global SAR of 74 ± 2% of the US Food and Drug Administration's guidelines of 3.2 W/kg (using 5 kg for adult head mass), as determined from the Siemens calibrated directional coupler measurements. A total of 5 control subjects were studied (Table 1), with mean height and weight of 62.1 ± 11 kg and 168.1 ± 8.1 cm (3 males, 2 females). A GRAPPA factor of 3 was used, achieving a 0.7 $\times$ 0.7 $\times$ 1.2 mm (nominal volume 0.6 mm³) resolution, with an acquisition time of 9.5 minutes. The phase encoding for the imaging blocks were either linear or center out, with the latter used to maintain maximal T₂ weighting at the center of k‐space. All timings are thus reported as the duration between the center of the inversion recovery to the k‐space center acquisition.

Table 1.

Signal‐to‐noise ratio, contrast‐to‐noise ratio, and SAR for the MPFLAGRE sequence

	SNR			CNR		SAR (FDA max. 15 W)
	GM	WM	CSF	GM‐CSF	WM‐CSF
3D SPACE	8.6	12.0	5.6	0.20	4.21	11.6 W (77%)
3D MPFLAGRE‐4 R1/3	13.7	30.4	7.0	6.17	8.00	10.7 W (70%)
3D MPFLAGRE‐4 R1/3, N=5	17.0 ± 2.3	35.2 ± 6.1	5.4 ± 18	6.63 ± 1.0	6.94 ± 1.0	11.1 ± 0.5 W (74%)

Abbreviations: CNR, contrast‐to‐noise ratio; FDA, US Food and Drug Administration.

4. RESULTS

4.1. MPFLAGRE‐2

Figure 6 shows the data acquired from a healthy control using the MPFLAGRE‐2 sequence. For comparison, Figure 6A,B shows the data without and with the T₂ preparation, respectively. The timings in Figure 6A are set to create MP2RAGE‐like contrast (ti1/ti2/TR = 0.9/2.67/5 seconds, tip angle = 5° and 9°), and as expected, the acquired and calculated images are all strongly T₁ weighted. Figure 6B,C shows the MPFLAGRE‐2 data, acquired with ti1/ti2/TR/TE = 0.14/1.65/5/0.100 seconds using pulse angles of 5° and 9°. Figure 6C shows the base images with T₂ contrast dominating the S₁ image. In S₁, the WM‐GM contrast is small; this is due to the dynamic range of the signal intensity governed by the high proton density CSF signal and the small WM‐GM variation in T₂ at 7 T. Consistent with simulation, there is little T₂ contrast in the CSF‐suppressed R_1/2 image (Figure 6B).

Figure 6.

Images from a control acquired without a spin‐echo preparation with parameters selected to create (A) an MP2RAGE‐like image with CSF nulling and bright WM and (B) the MPFLAGRE‐2 R_1/2 image, acquired with a spin‐echo TE of 100ms. The individual signal blocks S₁ and S₂ are also shown in (B) for the MPFLAGRE contrast, with image S₁ being heavily T₂ weighted. Image S₂ acquired later is heavily T₁ weighted. For the MPFLAGRE image in (B), the R_1/2 image shows CSF suppression with relatively weak discrimination between WM and GM, as seen from simulation

4.2. MPFLAGRE‐4

As the MPFLAGRE‐3 acquisition is very similar to that of MPFLAGRE‐4, Figure 7 shows the data from the MPFLAGRE‐4 sequence. Shown are both the CSF‐suppressed (Figure 7A) and T₂‐weighted (Figure 7B) calculated images, over a range of [$-$0.5, 0.5]. The T₂‐weighted images R_1/4 and R_(1+2)/4 again show the limited dynamic range of T₂ variation at 7 T between WM and GM. The difference between the calculated R_1/3 and R_(1+2)/3 images (showing the increase in dynamic range in total signal) is consistent with the simulation, and is a result of decreased calculated signal in normal WM and GM in R_(1+2)/3. With this effect being different between WM and GM, the R_(1+2)/3 images show slightly more GM‐WM contrast in comparison with R_1/3. Although this initial report is not intended to evaluate pathology, Figure 8 shows the performance of the MPFLAGRE‐4 (same acquisition parameters as in control from Figure 7) in 2 patients (epilepsy, left neocortical temporal lobe, anaplastic astrocytoma). In comparison to a clinical 3T FLAIR (Figure 8C), the epilepsy patient shows bright signal in the lateral cortex and hippocampus consistent with clinical data. The tumor patient shows that the bright MPFLAGRE signal in the temporal lobe extends into the posterior thalamus. The 3T studies were performed on a GE Signal Discovery MR750, with an inversion recovery for the epilepsy T₁ image; T₁ optimized fast spin echo for the tumor T₁ image; and both FLAIRs acquired with 2D fast spin echoes (TR/TI/TE = 11.6 seconds/2.5 seconds/154 ms, resolution = 0.43 $\times$ 0.43 $\times$ 3.3 mm³ [epilepsy] and TR = 8.7 seconds/2.2 seconds/150 ms, resolution = 0.63 $\times$ 0.63 $\times$ 5 mm³ [tumor]).

Figure 7.

The MPFLAGRE‐4 data from healthy control. A, Suppressed CSF, showing the 3D view of R_1/3 and R_(1+2)/3 images. B, T₂‐weighted image showing a 3D view of R_1/4, and calculated R_(1+2)/4 image. All gray scaling is shown at [−0.5, 0.5]

Figure 8.

Data from an epilepsy (A) and brain tumor patient (B). For both, the coregistered 7T MP2RAGE, MPFLAGRE‐4 (S_(1+2)/3), 3T clinical T₁, and FLAIR images are shown. For the epilepsy patient, increased signal intensity is identified in the lateral left temporal lobe (thick arrows, axial and coronal) and is consistent with clinical data. Note from the coronal image that there is also increased signal in the left hippocampus (thin arrow, coronal), which is characteristic of the local network involvement in temporal lobe epilepsy. The 3T FLAIR from this patient, acquired within 3 months of the 7T images, was interpreted as negative. For the tumor patient, all 7T and 3T images were acquired within 1 week. The arrows identify the increased MPFLAGRE signal in the left temporal lobe extending into the posterior thalamus

4.3. Comparison with 3D SPACE and contrast‐to‐noise ratio

For whole‐brain coverage, the MPFLAGRE sequence is compared with the 3D variable flip angle SPACE acquisition. As reported by Visser,10 the SPACE acquisition benefits substantially from 2D acceleration (2.5 $\times$ 2.5) but still requires a long TR. Our implementation used a GRAPPA factor of 4; however, a TR of 10 seconds was needed to reduce SAR to within the US Food and Drug Administration's guidelines, resulting in an acquisition of 19.7 minutes with TR/TI/TE of 10.1 seconds/2.35 seconds/200 ms. Figure 9 shows the mildly brighter GM, characteristic of the residual T₁ weighting. However, this contrast is B₁ ⁺‐sensitive, and the consistency of the image intensity is relatively poor, reflecting the variation in B₁ ^‐ (and B₁ ⁺). Matched in resolution (0.7 $\times$ 0.7 $\times$ 1.2 mm³) with the MPFLAGRE, Table 1 provides the contrast‐to‐noise ratio (CNR) and SNR between the SPACE and R_1/3 images for a single volunteer (single session). For region of interest measurements, contiguous 4 $\times$ 4 $\times$ 1 pixel blocks were taken from the centrum semi‐ovale, thalamus, and posterior ventricle for WM, GM, and CSF. The CNR and SNR were calculated in 5 healthy volunteers (Table 1) using the following equation:

$$CNR=\frac{\left(R_a-R_b\right)}{\sqrt{\left({\sigma }_a^2+{\sigma }_b^2\right)/2}},SNR=\frac{R_a}{{\sigma }_a}.$$ (7)

Figure 9.

Comparison of 3D SPACE (top) with R_1/3 (middle) and R_(1+2)/3 (bottom) MPFLAGRE‐4 single volunteer in a single session

5. DISCUSSION

5.1. T₁ and T₂ dependence of the MPFLAGRE signal

As shown in simulation and implementation, the MPFLAGRE sequence is able to generate T₂‐weighted images with controlled T₁ effects. The data show that the T₂ preparation is effective and the self‐correcting normalization with multiple acquisition blocks enables modulation of the T₁ relaxation effects. The T₁ and T₂ behavior is better understood by recognizing the balance between the T₂‐weighted S₁ and reference (non‐T₂‐weighted) images used in the normalization. In the case in which the common T₁ relaxation factor contributes to both S₁ and the reference image (e.g., S₄ in the MPFLAGRE‐4), the normalization results in negligible T₁ dependence regardless of the T₂ value (Figure 4D).

However, if the reference image S₄ is acquired with some residual T₂ (in addition to T₁) weighting, the normalization cannot cancel the T₁ weighting over all T₂ values. Tissues with very long T₂ will be transparent to the spin‐echo weighting, and generate a T₁‐weighted image. Tissues with extremely short T₂ will approach 0.0 for all T₁ values. Signal from tissues with intermediate relaxation values will vary depending on the specific acquisition parameters and the relaxation values, and thus is a target for optimization. As discussed, suppressing CSF is achieved by optimizing the timings of the 2 blocks (either S₁ and S₂ in MPFLAGRE‐2 or S₁ and S₃ in MPFLAGRE‐3 and MPFLAGRE‐4) to acquire substantially dissimilar signal intensities, ideally generating a signal in the 0 to $-$0.5 range. With lesional imaging, the goal is to generate significant calculated R_T2w,Ref difference between normal and longer (pathologic) T₂ values. Applying the T₂ weighting before the T₁ recovery null, we recognize that pathologic T₂ values will result in “less T₁ later shift” than tissues with normal T₂ values. Thus, to generate an increase in calculated signal from tissue with pathology, the reference image needs to be acquired after the null, sign inverted and larger in absolute amplitude compared with normal T₂ values (i.e., as exemplified with the MPFLAGRE‐2 timing).

The limited dynamic range of the 2‐block acquisition arises from the amplitudes of signal in the T₂‐weighted and reference images. With the MPFLAGRE‐3 and MPFLAGRE‐4, we expand this range and take advantage of the temporal persistence of the T₂‐dependent T₁ later shift effect through the summation of the 2 adjacent blocks. This later shift effect is larger for shorter T₂ components (i.e., normal T₂) rather than pathologic T₂ values (Figure 3). Thus, the sign‐sensitive summation of 2 adjacent blocks after the spin echo will result in an apparent decrease in amplitude for normal T₂ components and give decreased R_(1+2)/3, whereas the pathologic long T₂ components effectively saturate with R_(1+2)/3 at 0.5. With the shorter T₁ of normal WM, the R^WM falls more than R^GM. However, because of the similar T₂ values of WM and GM at 7 T, the absolute increases in $R_{(1+2)/3}^{GM,normal}-R_{(1+2)/3}^{GM,pathol}$ and $R_{(1+2)/3}^{WM,normal}-R_{(1+2)/3}^{WM,pathol}$ are similar for both tissues; this effect is also seen with the simpler MPFLAGRE‐2 sequence.

5.2. Specific absorption rate and CNR of the MPFLAGRE sequence

From a SAR perspective, the use of a single nonselective adiabatic spin echo allows the MPFLAGRE sequence to function with high efficiency. With the MPFLAGRE‐4 sequence, over n = 5 adult control volunteers, the 6‐minute average SAR was maintained at 74% ± 2% of the US Food and Drug Administration's guidelines. The CNR between WM‐GM is approximately 1.0 (data not shown), which is consistent with the small differences in their T₂ at 7 T. For GM‐CSF the CNR is 6.63 ± 1.0, and for WM‐CSF the CNR is 6.94 ± 1.0, which again is consistent with the similar intensities between WM and GM. In spite of the combination of multiple images used in this approach, which may otherwise be expected to increase variability, the coefficient of variance(s) in the MPFLAGRE data are comparatively low at less than 15%, reflecting their nonrandom source of variability.17 As a result, the SNR and CNR are higher with the MPFLAGRE sequence in comparison with the variable flip angle acquisition.

Inspection of Figures 5 and 6 show that there is enhanced signal intensity at the cortical rim and ependyma. This is consistent with the report of von Veluw,23 who concluded that increased T₂ contributes to their enhanced FLAIR signal intensity. Although quantitative T₂ data do not exist on these tissue components, a long T₂ for the cortical rim is not surprising, given that it contains the comparatively poorly vascularized molecular layer I of the neocortex. Similarly, the high signal intensity from the ependyma is likely related to its function as the neuroepithelial layer around the ventricle that contributes to the production and regulation of CSF.

5.3. Caveats with the MPFLAGRE sequence

The primary challenge with this sequence is its B₁ ⁺ sensitivity, as shown in Figure 9. A 15% increase or decrease in B₁ ⁺ results in increased intensity R_1/3 by less than 20% for WM and less than 28% for GM, although this effect decreases at longer T₂. However, this effect arises from the steep rise of R over the normal range of T₂ values. From a viewpoint of a given R_1/3 intensity, a 15% B₁ ⁺ variation results in a small decline in apparent T₂ by about 10 ms for GM (about 5 ms for white matter). With 30% B₁ ⁺ inhomogeneity, these effects increase substantially, with the T₂ sensitivity curves broadly shifting to lower T₂ values (i.e., there can be erroneously bright signal [> 0.30] for relatively normal T₂ values). In practice, our experience with the transceiver with the adiabatic refocusing pulses has not been shown to be a major problem for this over the entire head. Nonetheless, this sensitivity can be improved with incorporation of an adiabatic excitation spin‐echo module, such as that used by Dyvorne and Balchandani,16 although at added SAR cost.

With the CSF suppression generated from the normalization strategy performed between the multiple acquisition blocks, acquisition of differing resolution images but at the same delay times can change the resulting contrast. However, for the given parameters there is excellent consistency between different subjects, as indicated in Table 1 and the patient data. Nonetheless, this effect can be minimized with use of additional acceleration methods in 2 or 3 dimensions. In addition, magnetization‐transfer effects, which are not modeled here, can differentially affect the signal between the S_i,j blocks and thus contribute to the contrast.24 Notably, with magnetization‐transfer effects typically decreasing the signal intensity in high‐density (normal) tissue, this would result in an enhancement of the apparent MPFLAGRE sensitivity to pathology (the pathologic longer T₂ would exhibit a lesser magnetization‐transfer effect). Finally, the need for relatively long TR still limits the entire acquisition at approximately 10 minutes. However, with generation of both the T₂ and CSF‐suppressed T₂‐weighted images in a single matched image, the MPFLAGRE remains an efficient acquisition.

6. CONCLUSIONS

In summary, we have developed and demonstrated the MPFLAGRE sequence as a flexible acquisition that builds on the self‐correcting normalization strategy17 to generate T₂ and CSF‐suppressed T₂ weighted contrast at 7 T. This is performed by incorporating a longitudinal spin‐echo weighting within the context of an inversion recovery. With appropriate timing and combination of multiple imaging blocks during the inversion recovery, it is possible to control the T₁ dependence and specifically suppress CSF. Not surprisingly, the T₂‐weighted image, even when corrected for M₀, B₁ ^‐ and T₂ ^*, displays relatively little contrast between WM and GM, reflecting the small to minimal difference in T₂ at 7 T. As applied with a transceiver array, the sequence functions well within SAR guidelines. Overall, as the CSF‐suppressed T₂‐weighted contrast is designed for sensitivity to pathology, additional work with patients will be necessary to further optimize the parameters at 7 T.

Supporting information

FIGURE S1 Two configurations are shown for driving the 8 $\times$ 2 transceiver from 8 independent RF channels. In configuration A, a single RF channel drives 2 adjacent coils within a row. In configuration B, a single RF channel drives 2 longitudinally adjacent coils

FIGURE S2 Experimental data showing performance of the 2 configurations. For each configuration, the MP2RAGE and B₁ maps are shown. In configuration B, 2 RF distributions can be defined to excite the “homogeneous” volume and a “ring” volume

FIGURE S3 Bloch simulations of a late T₂ preparation 2‐block acquisition, S₁ and S₂. A, The time course of Iz amplitudes is shown (5 tissue components: blue, CSF; black, GM; red, WM with inversion; pathologic T₂). Note that the time axis is not linear, with each unit representing a simulation step. B, The surface plots of the signal amplitudes are shown over the [T₁, T₂] parameter space for S₁, S₂, and the calculated signal R_1/2 (using Eq. (1)). The calculated R_1/2 shows the CSF suppression and signal enhancement over the pathologic range of T₂. The black dots on the surface plots indicate the 5 tissue components: T₁, T₂ values for CSF are assigned at 4.3 seconds, 0.9 seconds; GM at 2.0 seconds, 60 ms; and WM at 1.2 seconds, 60 ms

Pan JW, Moon CH, Hetherington HP. Cerebrospinal fluid–suppressed T₂‐weighted MR imaging at 7 T for human brain. Magn Reson Med. 2019;81:2924–2936. 10.1002/mrm.27598