GB2356459A - MRI using modified FLAIR sequence - Google Patents

Abstract

MRI apparatus using a modified Fluid-Attenuated Inversion Recovery (FLAIR) sequence to reduce cerebro spinal fluid (CSF) or other fluid flow artefacts whilst maintaining good contrast in each imaging slice. A non-selective inversion pulse is applied to a plurality of slices, rf resonance in MR active nuclei is excited in each of those slices individually, and the slice excitation order is cycled so that the time from the non-selective inversion pulse to any particular phase-encode gradient of all the slices is the same. The time of the zero phase-encode gradient preferably corresponds to that for zero signal from cerebro spinal or other moving fluids.

Description

2356459 MAGNETIC RESONANCE IMAGING APPARATUS This invention relates to

magnetic resonance imaging apparatus.

The invention especially relates to apparatus for producing images which suppress fluid, particularly apparatus using inversion recovery sequences.

One such sequence is Fluid-Attenuated Inversion Recovery (FLAIR).

FLAIR provides a highly sensitive means of detecting lesions of the brain, particularly in the periventicular and cortical regions. It provides heavily T2 weighted images of the brain or spinal cord with low signal from cerebrospinal fluid (CSF) by operating at an inversion time (TI) which nulls the magnetisation in CSF. However flow can lead to poor CSF suppression, particularly in the posterior fossa, around the foramen of Munro and in the spinal canal. This problem can be avoided by the use of a non-selective inversion pulse, which inverts all the CSF, so that flow ceases to have an effect on signal intensity. A disadvantage in implementing this concept is that, the sequential excitation of the subsequent slices results in an inversion time (TI) that increases from slice to slice. This leads to variable contrast from one slice to the next with incomplete suppression of CSF in many slices.

Recently a method has been presented (D G Norris "Low Power Multi-slice MDEFT Imaging" in Proceedings, ESNIRMB page 108, 1999) of reducing RF dose for MDEFT imaging at high field by using a single non-selective inversion pulse followed by a rapid

2 stream of field echo slice excitations during each repeat time (TR). This method was subject to variable contrast due to the variable TI per slice, and this was dealt with by cycling the slice excitation order and associating each TI with a region of k-space rather than a particular slice.

In accordance with the present invention, slice order cycling is applied to the FLAIR sequence.

This allows uniform contrast to be attained combined the desirable flow insensitive properties of non-selective inversion pulse. The cycling provides a mechanism which can ensure that data for which CSF is at its null point is placed at the centre of k-space in each slice.

The sequence has interesting properties because of the very different state of magnetisation recovery of the brain or spinal cord, and CSF. This has consequences for the point spread functions (PSFs) of tissue and CSF, for both conventional and Fast Spin Echo (FSE) variants of this sequence.

Ways of carrying out the invention will now be described in greater detail, by way of example, with reference to the accompanying drawings, in which:

Figure I is a diagram of magnetisation recovery for brain tissue and CSF following a nonselective inversion pulse showing a conventional timing of the slice excitations in a five slice acquisition; 3 Figure 2 is a diagram of magnetisation recovery for brain tissue and CSF following a nonselective inversion pulse showing the timing of slice excitations in a five slice acquisition in accordance with the invention; Figure 3 shows an alternative timing of the slice excitations in a five slice acquisition in accordance with the invention; Figure 4 shows point spread functions for grey matter/white matter (top) and CSF (bottom) plotted on the same scale-, 10 Figure 5 shows curves of brain tissue (top) and CSF (bottom) plotted against kP assuming 165 msec per slice excitation and nine slices; Figure 6 shows the resulting PSF for brain tissue; 15 Figure 7 shows the resulting PSF for CSF; Figures 8 to 10 shows curves corresponding to Figures 5 to 7 for a fast spin echo sequence with ETL = 4; 20 Figures I I to 13 show curves corresponding to Figures 5 to 7 for a fast spin echo sequence with ETL = 4 and the T I order of alternate echoes swapped; and Figure 14 to 16 are curves corresponding to Figures 5 to 7 for a fast spin echo sequence 4 with ETL = 4 with TI order chosen to minimise modulation.

The basic sequence (which may be termed Reordered Inversion Prepared (RIP) FLAIR sequence) consists of a multi-slice spin echo acquisition with a non- selective inversion recovery (IR) pulse at the start of each TR, Within each TR each slice was excited once.

At the next TR the slice firing order was incremented so that what was previously the last slice was excited first followed by the rest of the slices in the previous order. The phase encoding structure was similar in concept to Fast Spin Echo. Each phase encode is associated with a slice firing time, such that the lowest strip of k- space comes from the first excitation, the next part from the second and so on. The number of lines in each strip is the total number of phase lines divided by the number of slice. This results in a smooth continuous T I filter across k-space with data from the centre of k-sp ace collected at the null point for CSF time (TI) for each slice.

Figure I illustrates a conventional FLAIR sequence in which each slice has a dfferent TI and CSF is only nulled in slice 3 in this case.

Figure 2 illustrates a RIP FLAIR in accordance with the invention with slice re-ordering in successive TRs so that each slice contains data from each TI. TI is the delay from inversion pulse to individual slice excitation.

In Figure 2 the slice firing order is incremented to excite the last slice to be previously excited, and so on. In Figure 3, an alternative slice firing order is shown in which the slice firing order is incremented to excite the second slice to be previously excited, and so on, Use of spin echoes fired in rapid succession with different phase encoding presents similar potential problems with stimulated echoes as are encountered with Fast Spin Echoes. In this case rather than using re-winders, the sequence structure requires the phase encode gradient to come after the refocusing pulse. Placement of the phase gradient prior to the refocusing pulse produces a stimulated echo offset from the centre of k- space by one phase encode group.

In one example, the sequence was implemented on a LOT Picker (Highland Heights, Cleveland, OH) HPQ Plus scanner. Its properties were tested using phantoms and on normal volunteers, who gave informed consent. Brain images were obtained with a bird cage coil and C-spine images were collected using a quadrature C-spine coil. Raw data was taken off line for reordering and reconstruction. The following parameters were used with the test sequence and a conventional FLAIR with non-selective inversion and a conventional fully interleaved FLAIR sequence: TE 30 msec, TR 6000 msec, FOV 35cm, slice thickness 4mm, I signal average, 128x256 resolution, TI 2100 rnsec, 9 slices.

The point spread functions (PSFs) for brain tissue (T1 = 400msec) and CSF (TI 4500msec) were simulated using IDL (Research Systems, Inc., Boulder, Colarado).

The new sequence produced CSF nulled images though out the brain and in the spinal canal. It was clearly superior to the fully interleaved IR. variant in regions where CSF flow is significant, such as in the C-spine. The SNR of the test sequence was XO/o inferior than the conventional nonselective variant of FLAIR for the optimal slice, but compensated for this by producing uniform contrast in all slices.

6 The PSFs for tissue and CSF are shown in Figure 4. The Tissue PSF is very similar to that for a standard conventional spin echo, whereas the PSF for CSF is suppressed by a factor of 8 and lacks a central high signal peak.

The variant of FLAIR test here provides appears to provide a practical solution to the problem examinations where CSF flow adversely affect the performance of conventional FLAIR sequences. A disadvantage of the current approach is that it is relatively slow, since no use has been mad of Fast Spin Echo methods. Combination of both multiple TIs and Tes in the same data set presents a challenge similar to the problems encountered with GRASE, where T2 and T2 are mixed.

In another example, the PSFs for brain tissue (T,=450msec) and CSF (TI=45OOmsec) were simulated for RIEP FLAIR sequences using IDL (Research Systems, Inc. Boulder, Colorado).

This entailed modelling signal intensity (S) as a function of phase encode position (kP) in k-space for a delta function signal source of normalised intensity. For a standard spin echo sequence there is a direct mapping of TI to kP, so that S(kP) = S(TI,TR;T,) = I - 2exp(-TI/Tl) + exp(-TR/T,) for an inversion recovery sequence, where T, is the longitudinal relaxation time of the tissue/fluid of interest. In our model TI varied from 1000 to 3 000 msec and TR was set to 6000 msec.

In the FSE sequence, there are variations of both TI and TE associated with kP. To achieve the required image properties the TE values of the multiple echoes must be smoothly distributed across kP with the desired effective echo time at the centre of k-space and the 7 CSF nulling value of TI must be associated with kP = 0. These requirements lead to a kspace tiling scheme with sequential variation of TI across blocks at constant TE. Various ways in which this can be achieved were explored. Because of the very long T2 of CSF, T2 decay was excluded from these initial simulations.

The sequence and associated reconstruction software were implemented using a spin echo structure on a LOT HPQ Plus scanner, Marconi Medical Systems (Highland Heights, Cleveland, OH). Sequence properties were tested using phantoms and normal volunteers. Brain images were obtained with a quadrature birdcage coil and C-spine images were collected using a quadrature C-spine coil. The following parameters were used for comparing REP FLAIR, with a conventional FLAIR sequence with a non-selective inversion and a conventional fully interleaved FLAIR sequence with slice selective inversions: TE 30 or 150 msec, TR 6000 msec, FOV 35cm, slice thickness 4mm, I signal average, 128x256 resolution, TI 2100 msec for CSF null, 9 slices.

The magnitude PSFs for brain and CSF are shown in Figures 5 to 7. Figure 5 shows curves of brain tissue (top) and CSF (bottom) plotted against kp assuming 165 msec per slice excitation and nine slices. The resulting PSFs for brain tissue and CSF are shown in Figures 6 and 7, respectively. The vertical scale in Figure 7 is magnified by a factor of 10.

The brain PSF is very similar to that for a standard conventional spin echo, whereas the PSF for CSF is suppressed by a factor of 8 and lacks a central high signal peak.

Figures 8-16 show results for FSE RIP FLAIR sequences with echo train length (ETL) of four. These require k-space tilings of four sets of fines with different TE, distributed with 8 the block acquired at the desired effective echo centred on kP = 0 and the other echoes distributed in symmetrical blocks on either side. Neglecting T2 effects, discontinuities are introduced as a result of T I recovery for brain (top) and CSF (bottom) as shown in Figure 8 for simple tiling in sequentially ordered blocks of constant TE. These discontinuities slightly increase the side bands on the brain PSF (Figure 9), but are reflected most dramatically in the magnitude PSF for CSF, Figure 10. The CSF signal is still suppressed, but undesirable non-local signal properties are introduced. The vertical scale in Figure 10 is magnified by a factor of 10.

In order to reduce the ringing in the PSF for CSF, two methods of smoothing out the discontinuities were investigated. The first case is shown in Figures I I - 13, in which the TI order in the first, third, fifth and seventh phase encode blocks have been reversed. This removed the discontinuities and increased the period of the k-space oscillations resulting in a PSF that decays at a more rapid rate. The asymmetric T, filter response for brain, with half of k-space homogeneous, might be exploited in half Fourier acquisitions for further reduction in imaging time. Figure I I shows T I filtered curves for brain tissue (top) and CSF (bottom) plotted in constant TE blocks as function of kP, and corresponding PSFs for Figure 12 brain tissue and Figure 13 CSF. The vertical scale in Figure 13 is magnified by a factor of 10.

In the second case we not only reverse the order of selected regions of kspace but also interchange regions in such a fashion to achieve a smooth curve with even lower periodicity. This is shown in Figures 14 to 16. The magnitude PSF for the CSF shows slightly higher intensity centrally, but side lobes of very low intensity. This would 9 correspond to slightly less complete CSF signal suppression, but the signal would be correctly localised. Figures 14 to 16 show simulated PSFs for a FSE sequence with ETL = 4 with TI order chosen to minimise modulation. Figure 14 shows T I filtered curves for brain tissue (top) and CSF (bottom) plotted in constant TE blocks as function of kP, and corresponding PSFs are shown for (Figure 15) brain tissue and (Figure 16) CSF. The vertical scale in (c) is magnified by a factor of 10.

The sequence as implemented produced CSF nulled images throughout the brain and the spinal canal. It was clearly superior to the fully interleaved IR variant in regions where CSF flow is significant, such as in the C-spine. The SNR of the test sequence was 12% inferior to the conventional non-selective variant of FLAIR for the optimal slice, but compensated for this by producing uniform contrast in all slices. We have imaged the brain and spine in 3 subjects with good results in all cases. The sequence has been released for clinical testing.

RIP FLAIR appears to provide a practical solution to examinations where CSF flow adversely affects the performance of conventional FLAIR sequences. A disadvantage of the current implementation is that, like conventional FLAIR, it is relatively slow. Using a faster acquisition technique, such as FSE could reduce imaging time. Simulations of the PSFs for brain and CSF show that FSE implementations are likely to be well behaved. A critical consideration is likely to be the choice of ETL and number of slices. Increasing the ETL increases the time between slice excitations within a TR. For a given number of slices, this will increase the time between the first and last excitation, so increasing the range of TI values that must be accommodated. A practical choice for heavily T, weighted FLAIR sequences may be to use the maximum ETL consistent with the last echo having the desired effective echo time for the image. The interaction of the T, and T2 filtering across k-space is implied by this approach.