Elsevier

NeuroImage

Volume 49, Issue 2, 15 January 2010, Pages 1271-1281
NeuroImage

MP2RAGE, a self bias-field corrected sequence for improved segmentation and T1-mapping at high field

https://doi.org/10.1016/j.neuroimage.2009.10.002Get rights and content

Abstract

The large spatial inhomogeneity in transmit B1 field (B1+) observable in human MR images at high static magnetic fields (B0) severely impairs image quality. To overcome this effect in brain T1-weighted images, the MPRAGE sequence was modified to generate two different images at different inversion times, MP2RAGE. By combining the two images in a novel fashion, it was possible to create T1-weigthed images where the result image was free of proton density contrast, T2⁎ contrast, reception bias field, and, to first order, transmit field inhomogeneity.

MP2RAGE sequence parameters were optimized using Bloch equations to maximize contrast-to-noise ratio per unit of time between brain tissues and minimize the effect of B1+ variations through space. Images of high anatomical quality and excellent brain tissue differentiation suitable for applications such as segmentation and voxel-based morphometry were obtained at 3 and 7 T.

From such T1-weighted images, acquired within 12 min, high-resolution 3D T1 maps were routinely calculated at 7 T with sub-millimeter voxel resolution (0.65–0.85 mm isotropic). T1 maps were validated in phantom experiments. In humans, the T1 values obtained at 7 T were 1.15 ± 0.06 s for white matter (WM) and 1.92 ± 0.16 s for grey matter (GM), in good agreement with literature values obtained at lower spatial resolution. At 3 T, where whole-brain acquisitions with 1 mm isotropic voxels were acquired in 8 min, the T1 values obtained (0.81 ± 0.03 s for WM and 1.35 ± 0.05 for GM) were once again found to be in very good agreement with values in the literature.

Introduction

In the past decade, the magnetization-prepared rapid gradient echo, MPRAGE (Mugler and Brookeman, 1990), sequence has become one of the most commonly used sequences to obtain T1-weighted anatomical images of the human brain, in particular at high magnetic field. MPRAGE images are routinely used as anatomical reference for fMRI or for brain tissue classification in voxel-based morphometry (Ashburner and Friston, 2000). However, at high static magnetic fields (≥ 3 T), the increased inhomogeneity of the transmit B1+ and receive B1 fields creates intensity variations throughout the image (bias field). Bias fields not only render segmentation and quantitative analysis difficult but also severely affect image quality at ultra-high fields (≥ 7 T). The use of adiabatic pulses to perform the inversion in the MPRAGE is only partially able to mitigate the effects of inhomogeneous B1.

A number of strategies have been proposed to minimize or to correct bias fields generated by the inhomogeneity of the B1 fields. Most correction strategies aim at correcting the combined (transmit and receive) bias field via post-processing techniques. This can be done either by low-pass filtering (Cohen et al., 2000, Wald et al., 1995) or by fitting slowly varying functions such as Gaussians or low order polynomials (Styner et al., 2000). The result from these low pass filters or fits is then subtracted from the original image. Such approaches can be performed iteratively together with the segmentation process (Ashburner and Friston, 2005) in which case the quality of the estimated bias field is measured by the resulting image intensity distribution. Although such approaches can improve image quality, they also tend to affect the spatial characteristics of the resulting image (Belaroussi et al., 2006). To correctly deal with B1-induced inhomogeneities, transmit and receive inhomogeneities should be addressed separately and explicitly due to their different nature and implications on the signal intensity and contrast. Although transmit and receive B1 fields can be numerically calculated (Vaughan et al., 2001), such simulations require a priori knowledge of the distribution of brain tissues and their conductivity as well as complete characterization of the used RF coils. Reception B1 inhomogeneities only affect the signal amplitude of the image by a multiplicative factor related to the coil sensitivity, therefore, coil sensitivity maps could be used to correct such reception inhomogeneity (Axel et al., 1987, Roemer et al., 1990). These methods are only valid at low magnetic fields, where the ground truth of “a flat” coil sensitivity is given by the reception with a volume coil. At high fields, the lack of a ground truth makes the task of calculating an absolute coil sensitivity map more intricate. Transmission B1 inhomogeneities are more complex to correct as, for example, the T1 weighting of the sequence is closely linked to the local flip angle implying that the T1 weighting throughout the image will be different. Therefore, although many techniques exist to calculate the transmission B1 field, for example (Stollberger and Wach, 1996), to correct its implication on the image intensity, it is also necessary to have prior knowledge of the contrast dependence on the flip angle (Wang et al., 2005). A recent promising technique is to, instead of correcting a posteriori, effectively reduce the inhomogeneity by performing parallel transmission (Katscher and Bornert, 2006). Due to the complexity and expensiveness of the hardware necessary to perform parallel transmission, the technique is not yet widely available. Another approach to deal with the transmission inhomogeneity is to carefully adjust the amplitude of the various RF pulses used in the sequence so that the resulting contrast is less dependent on the local flip angle accuracy (Thomas et al., 2005, Van de Moortele et al., 2009), which is the approach that will be pursued in this manuscript.

Recently, an extension of the MPRAGE sequence, which performs the acquisition of 2 volumes after each inversion (see Fig. 1), has been proposed (Marques, et al., 2008, Van de Moortele et al., 2008, Van de Moortele et al., 2009). In this article, this sequence will be referred as MP2RAGE (Magnetization Prepared 2 Rapid Acquisition Gradient Echoes).

In a conventional MPRAGE, the signal is not exclusively dependent on T1 contrast but also on M0 (often referred to as proton density) and T2⁎. Both M0 and T2⁎ tend to reduce the available T1 contrast of the MPRAGE image as the values of M0 and T2⁎ of cerebral spinal fluid (CSF), grey matter (GM), and white matter (WM) decrease in this order. On the other hand, if two MPRAGE images are acquired at two different inversion times, but have otherwise the identical sequence parameters, they will be affected in identical manner by B1, M0 and T2⁎ and thus, a combined image by means of a ratio will be independent of B1 as well as M0 and T2⁎ (Van de Moortele et al., 2009).

One way of acquiring bias field-independent images is to perform quantitative imaging, of which T1 mapping is a popular example due to its high tissue contrast. Various T1 mapping techniques have been developed throughout the years. Typically T1 is estimated using variations of either inversion recovery or saturation recovery sequences or based on variable flip angle spoiled gradient echoes (Christensen et al., 1974). The most common variations of the inversion recovery technique are the Look–Locker approaches (Look and Locker, 1970), where multiple RF pulses are applied during the recovery process allowing for a faster measurement of the longitudinal recovery process. This concept has been combined with EPI (Gowland and Mansfield, 1993) or with flash readout (Deichmann et al., 1999). While EPI-based methods offer the quickest method for T1 mapping, they suffer from severe distortions, and the T1 maps in regions with tissue partial volumes will bias the T1 to the tissue with the longest T2⁎ due to the inherently long echo times of the EPI readout. Recently, spoiled gradient echo methodologies to perform T1 mapping have also successfully been brought into clinical practice (Deoni, 2007, Deoni et al., 2005), allowing high-resolution whole-brain T1 maps to be acquired in ∼ 10 min. One of the drawbacks of this technique is that, as it requires various sequences to be applied, the need to co-register the images due to inter-scan motion implies an effective decrease of the spatial resolution.

Since the MP2RAGE resulting image is to a large extent purely T1 weighted, it provides excellent base for fast T1 estimation. The goal of this study was to optimize the parameters that render a bias free T1-weighted whole-brain image that allows calculation of T1 relaxation time at high spatial resolution (< 1 mm3) in ∼ 10 min in one sequence, without the need to co-register images, facilitating online T1 calculations. The proposed method was validated in phantoms by comparison to standard T1 maps (inversion recovery EPI based) and in vivo by comparison to brain tissue T1 values found in literature for both at 3 and 7 T.

Section snippets

Methods

The MP2RAGE sequence shown in Fig. 1, as the conventional MPRAGE, starts with a magnetization preparation performed with an adiabatic inversion and, after a delay TA, a gradient echo block is introduced. This low flip angle gradient echo block has a short repetition time, TR, and is repeated nPE2 times stepping linearly through the whole second-phase encoding direction. The center of k-space in this direction defines the effective inversion time TI1. At the end of the first rapid gradient echo

Results

The optimization of the contrast-to-noise ratio (CNR) of the MP2RAGE image was performed as a function of five variables, rendering its visualization rather complex. In Fig. 2, it is possible to see representations of the dependence of the CNR as a function of two variables at a time, keeping the remaining three variables fixed at the values described on the left column. One conclusion that can be drawn from Fig. 2 is that CNR is a smooth and slow varying function of the inversion times, flip

Discussions and conclusions

In this work, we successfully implemented a MP2RAGE sequence, showed a new way to combine these images that has improved noise propagation characteristics and optimized its contrast for brain tissues at 3 and 7 T. The contrast obtained was independent of B1, T2⁎ and proton density. Furthermore, optimization of the flip angles allowed creating MP2RAGE images that were largely independent from the transmission B1 field, B1+. Therefore, this methodology is of great utility in the context of large

Acknowledgments

This work was supported by Centre d'Imagerie BioMédicale (CIBM) of the UNIL, UNIGE, HUG, CHUV, EPFL, and the Leenaards and Jeantet Foundations.

References (39)

  • van der KouweA.J. et al.

    Brain morphometry with multiecho MPRAGE

    NeuroImage

    (2008)
  • AxelL. et al.

    Intensity correction in surface-coil MR imaging

    AJR Am. J. Roentgenol.

    (1987)
  • ChristensenK.A. et al.

    Optimal determination of relaxation-times of Fourier-transform nuclear magnetic-resonance—determination of spin-lattice relaxation-times in chemically polarized species

    J. Phys. Chem.

    (1974)
  • CohenM.S. et al.

    Rapid and effective correction of RF inhomogeneity for high field magnetic resonance imaging

    Hum. Brain Mapp.

    (2000)
  • DeichmannR. et al.

    Fast T1 mapping on a whole-body scanner

    Magn. Reson. Med.

    (1999)
  • DeoniS.C.

    High-resolution T1 mapping of the brain at 3T with driven equilibrium single pulse observation of T1 with high-speed incorporation of RF field inhomogeneities (DESPOT1-HIFI)

    J. Magn. Reson. Imaging

    (2007)
  • DeoniS.C. et al.

    High-resolution T1 and T2 mapping of the brain in a clinically acceptable time with DESPOT1 and DESPOT2

    Magn. Reson. Med.

    (2005)
  • GowlandP. et al.

    Accurate measurement of T1 in vivo in less than 3 seconds using echo-planar imaging

    Magn. Reson. Med.

    (1993)
  • KatscherU. et al.

    Parallel RF transmission in MRI

    NMR Biomed.

    (2006)
  • Cited by (0)

    View full text