A comparison of magnetization prepared 3D gradientecho (MP-RAGE) sequences for imaging of intracranial lesions

Magnetic

Resonance

Imaging, Vol. 14, No. 3, pp. 329-335, 1996 Copyright 0 1996 Elsevier Science Inc. Printed in the USA. All rights reserved 0730-725X/96 $15.00 + .oO

0730-725X( 95) 020950

ELSEVIER

0 Technical Note A COMPARISON OF MAGNETIZATION PREPARED 3D GRADIENTECHO (MP-RAGE) SEQUENCES FOR IMAGING OF INTRACRANIAL LESIONS STEFAN BLUML,~ LOTHAR R. SCHAD, * JOHANN SCHARF,$ FREDERIK~ENZ,~ MICHAEL v. &VOPP,* AND WALTER J. LORENZ* *Department of Radiology, German Cancer Research Center, Heidelberg, Germany, $Department of Radiodiagnostic and SDepartment of Clinical Radiology, University of Heidelberg, Heidelberg, Germany In a pilot study including 64 patients with different types of brain tumors we investigated four types of MP-RAGE sequences. The sequences dller in the length of the recovery period and the data acquisition mode (sequential vs. centric phase-encoding). The sequence with sequential encoding and a short recovery period provided images that reached the quality and reliability of spin-echo images. The other MP-RAGE sequences failed in providing equivalent information. In particular, a considerable number of small lesions identified in spin-echo images were not detected in MP-RAGE images. The impact of the evolving magnetization on the point spread function was analyzed by performing simulation calculations. It was found that lesions with short T1 times are rendered with low spatial resolution when sequence parameters are not set appropriately. The low overall quality of images obtained by sequences applying centric encoding may be explained by eddy current effects as reported in other recently published studies. Keywords: lesions.

Magnetic

resonance imaging; Magnetization

prepared

3D gradient-echo

techniques;

Brain

Therefore, both image contrast and blurring effects in MP-RAGE imaging are functions of various parameters. The image contrast is approximately given by

INTRODUCTION Highly T, weighted ( Tlw) images can be generated with use of the 3D magnetization-prepared rapid gradient-echo (MP-RAGE) sequence.’ MP-RAGE imaging employs three phases: a preparation period, a data acquisition phase, and a recovery period before the next preparation (Fig. 1) . An additional phase-encoding gradient in the slice selection direction is applied for 3D Fourier transform imaging. This set of phase-encoding is nested inside the phase-encoding for in-plane spatial information. Figure 1 shows a diagram of the MP-RAGE sequence. Fundamental to this technique is the acquisition of k-space data during an evolving magnetization. Sequence parameters are preparation flip angle p, excitation flip angle LY,the repetition time TR of gradient-echo data acquisition segments, the inversion time TI1, and the recovery delay time T12.

tions those methods for sequence optimization

l/7/95; ACCEPTED lllfY95. Address correspondence to PD Dr. L.R. S&ad, Forschungsschwerpunkt Radiologie, Deutsches Krebsforschungszen

trnm, Postfach 101949, D-69009 Heidelberg, Germany. tPresent address: Huntington Medical Research Institute, Magnetic Resonance Spectroscopy Unit, Pasadena CA.

the signal intensities when central spectral frequencies were measured. The amount of blurring decreases with a more gradual approach of the magnetization to steady-state due to a more regular point-spread function (PSF) . Methods for deriving optimum parameters for this sequence type are discussed in the literature.2’3 Mugler et al.’ describe an algorithm for deriving the flip angle series, which results in arbitrarily chosen k-space filters. A variable flip angle 3D MP-RAGE sequence with a volume acquisition for brain imaging, promising optimum white-gray matter contrast, is presented. Epstein et a1.3 gives a more general framework of simulated annealing to the specific case of optimizing the white-gray matter contrast. For clinical applica-

RECEIVED

329

suffer

Magnetic Resonance Imaging l Volume 14, Number 3, 1996

330

(a>

A

(b)

cc>

Recovery period

Preparation

+

Inner (slice)phaseencodingloop

Outer (in-plane)phaseencodingloop

Fig. 1. Diagram of the MP-RAGE sequence. The outer (in-plane) phase-encoding loop consists of a preparation period (a), an inner (slice direction) phase-encoding loop (b), and a recovery period (c).

most from the amount of computation to calculate parameters, and from the fact that the physical characteristics of pathological tissue is often unknown prior to an examination. Further, changing sequence parameters from exam to exam results in confusing contrasts that might render the reading of the images difficult and is not acceptable in clinical routine. Therefore we restricted our investigations to four MP-RAGE sequence types, which can be clearly separated by the data acquisition mode and the recovery time T12, as described in the methods section. In these sequences a constant flip angle for the readout pulses was used. Until now, several clinical studies were undertaken with MP-RAGE sequences of this type.4-10 Brant-Zawadzki et aL4 and Wenz et aL6 reported that the MP-RAGE sequence may provide an alternative to T,-weighted spin-echo (SE) brain imaging, but however not all lesions detected with spin-echo were seen in MP-RAGE images. This clearly limits the practical application of this sequence for diagnostic purposes, because not the white-gray matter contrast but high sensitivity in lesion detection is the most important point. A lack of sensitivity would require an additional SE examination and the MP-RAGE measurement can be omitted. This study was performed because specific clinical applications, like high precision radiotherapy treatment planing of brain tumors based on MR data” require a large number of high-quality images for optimum target definition. This can be achieved with use of SE multislice sequences. But to gain high resolution, highquality images and for scanning the whole brain, at least two measurements must be performed with gap shift. The large number of slices requires a fast repetition of 90 and 180 radiofrequency (RF) pulses, resulting in a high specific absorption rate ( SAR) , which

often exceeds the allowed limits. The use of a 3D gradient-echo sequence would simplify the measurement protocol, improve the spatial resolution, and solve the problem of high SAR values. A further improvement of diagnostic value offers the possibility of image reformatting to different orientations for better visualization of lesions and their vascular connections. MATERIALS

AND

METHODS

Imaging Techniques All measurements were performed on a MAGNETOM SP 1.5 T (Siemens, Erlangen, Germany) superconducting whole body imager. Two dimensional T,weighted SE imaging was performed with repetitions times TR = 600-700 ms, echo times TE = lo-15 ms, a 2-6 mm slice thickness, field of view FOV = 260-339 mm, and a matrix size of 256 X 256 pixels. The number of excitations (NEX) was equal to two in all cases, yielding measurement times of 5-6 min. MP-RAGE measurements were performed using flip angles cu of lo-14”, TR = 10 ms, TE = 4 ms. One hundred twenty-eight partitions, 256 x 256 image matrices were measured with a slab thickness of 256 mm and FOVs ranging from 256 to 339 mm. After each phase-encoding scan, the remaining transverse magnetization was dephased by a spoiler gradient. In addition, the phase angle of the small angle excitation pulse was incremented for each phase-encoding scan or randomized with use of an RF spoiling kit, which is provided to switch the phase of the MR frequency between the phase encoding scans ( 12). For preparation, inversion pulses of /? = 180” were used. We investigated four types of MP-RAGE sequences that differ in the length of the recovery period T12 and in the data acquisition mode. The first type, designated

MP-RAGE

sequences 0 S. BLUML ET AL.

331

Table 1. Designation of MP-RAGE sequences investigated in this study Phase encoding

Designation

TI* h-4

Standard

1000 Sequential

Inversion-contrast

1000

Centric-3DTPLmin 3DTFLmin

Acquisition time (T-4 bN

Centric

10-12 10-12

5 Centric 5 Sequential

6-8 6-8

The sequences differ by the Tlr time and the k-space acquisition order of the inner phase encoding loop. Because of hardware restrictions, the minimum T12 delay was 5 ms. The total acquisition time (TA) for 128 3D-partitions (256 X 256 matrix) also depends on the TI,-delay (TI, typically 200-400 ms), which results in measurement times as given in the table.

as “standard” sequence, was the originally implemented sequence on our scanner employing sequential encoding and a long recovery period T12 of 1000 ms. In the second sequence type the recovery period TIa was set to a minimum value of 5 ms. This sequence with sequential phase-encoding but minimum recovery time TIa is called “3DTFLmin.’ ’ While both sequences use the standard sequential phase encoding, the Trcontrast can be significantly improved by means of centric phase encoding order 10~13~14 in the inner (slice encoding) loop. Applying a long recovery period T12, magnitude images show a typical inversion contrast. A sequence showing this contrast is designated as “inversion” (contrast), while a reordered MP-RAGE with minimum TI*-delay is designated as “centric3DTFLmin.” The designations of the different sequences are arranged in Table 1. Measurement Protocol

Transversal SE and MP-RAGE data sets were obtained from 64 patients with primary (23 patients) or secondary (41 patients) brain tumors during the workup for high precision, stereotactic radiotherapy (Table 2). A sagittal scout (FLASH 2D) was followed by the transversal standard SE measurement. Two measurements with gap shifts were performed to cover the whole volume of interest. In a random order after the injection of Gd-DTPA (0.1 mmol/kg b.w., 2-min delay) either the SE measurements were repeated followed by an MP-RAGE experiment, or MP-RAGE images were generated followed by the SE measurements. The examinations were performed with a homemade linear polarized head coil that fits closely to stereotactic localization system. Using this localization system a stereotactic coordinate system could be derived by a linear fit from the distances of reference points in each slice, which are directly correlated to the absolute z-position.“s’5

RESULTS The MP-RAGE images obtained from patients with primary tumors (23 patients) were of diagnostic value in all of the cases studied. Due to the short echo time of TE = 4 ms no substantial magnetic susceptibility effects at brain-bone-air interfaces were notified. However, we found that the capability of MP-RAGE sequences to depict small lesions, in particular small metastases, differed significantly between the investigated sequences. In 3 out of 11 cases not all metastases seen in SE images were seen in MP-RAGE images obtained by the “standard” sequence. The “centric3DTFLmin” sequence failed in three out of four cases. The “inversion” MP-RAGE was tested in two cases where all metastases were detected. In the images obtained from the “3DTFLmin” sequence (24 patients with metastases) two lesions more than in the SE images were found. These lesions were detected during rereading the SE images after having them localized in MP-RAGE images. In general the MP-RAGE images appeared to be less sensitive to flow artifacts due to pulsation in the sag&al sinus or in the carotid arteries, which plays an important role in the posterior cranial fossa. SE (TR/TE = 690/ 15 ms) and MP-RAGE images from patients with multiple brain metastases are shown in Fig. 2. SE (a) and corresponding “standard” MPRAGE TIr/a = 200/ 12 (b) image: The arrows mark lesions visible in the SE image but not in the “standard” MP-RAGE image. SE (c ) and corresponding ‘ ‘inversion’ ’ MP-RAGE TI,/a = 350/13 (d) image. SE (e) and corresponding “centric-3DTFLmin” MPRAGE TIr /(Y = 300/ 13 (f ) image. SE (g) and corresponding “3DTFLmin” TII/n = 275113 (h and i): The arrow in Fig. 2h marks a small lesion which was not detected in a first reading of the SE images. DISCUSSION We investigated the recently developed 3D gradient-echo sequence MP-RAGE for ‘its possible clinical Table 2. A total of 64 patients with different types of primary or secondary brain tumors were examined

Standard

Centric 3DTFL

3DTFL

Inversion

min

rnin

Meningioma Metastases Glioma

1 11 4

0 2 4

1 4 2

6 24 5

Total

16

6

7

35

Magnetic Resonance Imaging 0 Volume 14, Number 3, 1996

332

CD)

(El

Fig. 2. SE (TR/TE=690/ 15) and MP-RAGE images of patients with brain metastases. SE (a) and corresponding “standard” MP-RAGE T&/a = 200/12 (b) image: The arrows mark lesions visible in the SE image but not in the “standard” MPRAGE image. SE (c) and “inversion” MP-RAGE TIl/a = 350/13 (d) image. SE (e) and corresponding “centric-3DTFLmin” MP-RAGE TI,tcv = 300113 (f) image. SE (g) and corresponding “3DTFLmin” TI,Ia = 275/13 (h and i): The arrow in Fig. 2h marks a small lesion that was not detected in a first reading of the SE images. After having detected this lesion in the MP-RAGE image the lesion was also detected in the SE image. Two successive MP-RAGE images are displayed here due to the lower slice thickness of 2 mm in comparison to the 4 mm of the SE image.

MP-RAGE sequences 0

S. BLUML

333

ETAL.

.25 -j------

r,

‘I I -1.0

I 1 ( ’

_._._.-

T,

=

1000

mm.

T,

*

_____-

T,

=

WOO

ms.

T,

-

)

0

I

I

1000

500

100 1400

m.

.05

ms

1

1500

2000

Distance in pixel

Time (ms)

r .25 .20 E VI a

-

T,

-

200

ma.

T,

-

60

ms

T, T,

= =

600 lODo

ms. m,,

T, T,

= =

60

m.

-.-.-.-

100

ms

----_-

T,

=

2000

ms.

T,

=

1400

ma

. .

-0.5

I

-1.0

I

0

I

.._

I

500

1000

I

I

I

1500

2000

2500

.15 .lO .05

Time (ms)

Distance in pixel

.25 .20 .15-

-

T,

-

200

ms.

T,

-

60

.10 -

mm

.--. . 1, = (100 ,J,,, T, a 20 ma -.-.-.T, - 1000 ms, T, = 100 ms ______ T, * 2000 ms, T, = 1400 ms -1.0

I 0

I 200

I 400

I 800

I 800

I 1000

I 1200

.05 I

I 1400

-10

-5

0

5

10

Distance in pixel

Time (ms)

(C) 1.0

’

1

Tll

i

Tl2=

I - 200 ma

I

I m-

12'

6 ms

.25 0.5 -I

0 q 9”

I/

,____.__.......

. ...“‘...

. ..__._.... . ________

, ./I*__. .;__.’ ____.-.-.- ___._____.-.-.-.-. ._.._____---- ______-----. /I 0.0 ! t .y-

_ _- - - ---J _

F-k

.20 .15

IS

.lO .05 .oo -10

-5

0

5

Distance in pixel

Time (ms)

CD)

10

334

MagneticResonance Imaging0 Volume14,Number3. 1996

use for imaging of intracranial lesions. A straightforward application of 3D gradient-echo imaging with improved spatial resolution is radiotherapy treatment planning where a large number of images are required for optimum target volume definition. MP-RAGE imaging differs from conventional imaging by a dynamic evolving magnetization during the readout period. Sequence parameters may be set appropriately to optain optimum gray-white matter contrast as reported.2X3To perform such calculations the physical tissuecharacteristics such as the T1, T2 relaxation times and the spin density must be known. But these parameters are often not available for pathologic tissue. Further, online changing of sequenceparameters may render the reading of the images more difficult. In this pilot study we focused our investigations on four MP-RAGE sequencetypes, which were compared with conventional SE imaging. The MP-RAGE sequences differ by the data acquisition mode and the recovery period T12 (Table 1). The patients were scheduled for radiotherapy treatment and examined in a stereotactic localization system minimizing any patient movement during or between measurements. The most striking finding is that MP-RAGE imaging fails completely in a considerable number of casesin detection of small lesions (see results section). This problem occurred only when the “standard” or “centric-3DTFLmin” sequencewas used. In fact, for ethical reasons,the use of these two MP-RAGE sequences in patients’ exams was stopped after this observation was confirmed. While one may speculate that eddy current effects are responsible for the low quality of the “centric3DTFLmin” images,* this does not explain our fmdings for the “standard” sequence. Systematically reduced signal enhancements within the lesions due to a lower contrast agent concentration (as discussedby Muggler and Brookman16) can be excluded as an explanation, becausethe exams were performed in different successionfrom patient to patient. T2 shortening effects, as mentioned by Brant-Zawadzki et a1.4should influence all used MP-RAGE sequencesand are therefore unlikely to cause the observed problems in two out of four sequences,each using the same TE.

In simulation calculations we investigated the effects of the evolving magnetization on the PSFs for the sequencetypes used. For this purpose a simulation program basedon the partition model ” similar to those of Crawly et al.‘* and Sekihara” was used. The program allows a complete simulation of the developing magnetization at any voxel position for the MP-RAGE sequences(cf. Fig. 3 ) . From the transverse magnetization immediately after each excitation pulse the PSF was computed by taking the phase-encoding order into account. We used T, and T2 times that are typically for white matter (T, = 600 ms, T2 = 80 ms), gray matter (T, = 1000 ms, T2 = 100 ms), and cerebrospinal fluid (T, = 2000 ms, T2 = 1400 ms). A fourth simulation with a short T, time of T, = 300 ms was added because of the Ti-shortening effect of contrast agents after accumulation in the lesion. The spin density was assumedto be equal to one. It does not effect the time course of the magnetization nor the shape of the PSF. Simulation of the “standard” sequenceshows considerable blurring for all simulated tissue types (cf. Fig. 3a). In comparison, the PSF of the compartment with a short T, time of T, = 300 ms is significantly improved for the other sequences(cf. Fig. 3b, 3c, and 3d). We speculate therefore that lesions with short T1 relaxation times are rendered with a lower spatial resolution using the “standard” sequence. It may be of interest for further investigations whether sequence optimization must be performed in a way that compartments producing the highest MR signals do have a uniform magnetization during data acquisition (and therefore optimum PSFs) . The main disadvantageof the “inversion” sequence is its sensitivity to eddy currents producing low quality images and the unusual “inversion contrast.” Using an improved imaging sequenceexecuting the nonphaseencoding gradient pulses several times before the acquisition, the overall image quality may be improved.* But images not showing the usual T1 contrast are not well accepted in clinical routine. A solution to get images “looking like T,-weighted,” would be to calculate imagesfrom the real part of the MR signal. However, this makes an additional amount of computation necessary. On the other hand, this sequenceoffers a striking high

Fig. 3. The evolving longitudinal magnetizationof compartmentswith T, and T2 timesas given in the legendis plotted vs. the measurement time. From the transversemagnetizationimmediatelyafter the excitation pulsesthe PSF was computedby taking centric and sequentialphase-encoding into account for eachMP-RAGE sequencetype. The simulationsshow that in caseof the “standard” sequenceblurring effects for all tissuecompartmentsare expected (a). The simulationsof the other sequences,“inversion” (b), “centric-3DTFLmin” (c), and “3DTFLmin” (d) show that tissueswith short T, relaxation times are imagedwith high signaland high spatialresolution,and thus a minimumamountof blurring.

MP-RAGE sequences0

contrast between contrast agent accumulating lesions and noncontrast agent accumulating surrounding edema (cf. Fig. 2d and Ref. 19). This may be of value for an optimum target definition in treatment planning. In comparison to MP-RAGE imaging SE sequences produce optimum PSFs for all compartments because the magnetization is kept in a steady state. Disadvantages arise from the fact that the achievable spatial resolution, in particular the slice thickness, is lower than in 3D gradient-echo sequences. Therefore partial volume effects may lower the sensitivity for lesion detection as demonstrated in one patient (cf. Fig. 2g, 2h, and 2i). Well known other advantages of the MP-RAGE sequence are lower SAR values and a better slice profile. Additional slice orientations can be reconstructed from a 3D data set improving the diagnostic value. Due to shorter TEs and a saturation effect of flowing blood in the whole excitation volume, MP-RAGE sequences are less sensitive to pulsation artifacts. Our investigations suggest that MP-RAGE sequences may provide images reaching the quality of SE images when appropriate sequence parameters are set. workwassupported by theRolandEmstStiftungTS 076/2.001/94( S.B.). The authorsaregratefulto L.J. Haseler for helpin reviewingthemanuscript andusefuldiscussions. Acknowledgments-This

REFERENCES 1. Mugler, J.P.; Brookman,J.R. Three-dimensional magnetization preparedrapid gradient-echo(3D MP RAGE). Magn. Reson.Med. 15:152-157; 1990. 2. Mugler, J.P; Epstein, F.H.; Brookeman,J.R. Shaping the signalresponseduring the approachto steady state in three-dimensionalmagnetization-prepared rapid gradient-echo imaging using variable flip angles.Magn. Reson.Med. 28:165-185; 1992. 3. Epstein,F.H.; Mugler, J.P.; Brookeman,J.R. Optimization of parametervalues for complex pulse sequences by simulatedannealing:Application to 3D MP-RAGE imagingof the brain. Magn. Reson.Med. 31:164- 177; 1994. 4. Brant-Zawadzki, M.; Gillan, G.D.; Nitz, W.R. MP RAGE: A three-dimensional, T,-weighed,gradient-echo sequence-Initial experiencein the brain. Radiology 182:769-775; 1992. 5. Shah, M.; Ross, J.S.; Tkach, J.; Medic, M.T. 3D MPRAGE evaluation of the internal auditory canals.J. Comput. Assist.Tomogr. 17:442-445; 1993. 6. Wenz, F.; He/?, T.; Knopp, M.V.; Weisser,G.; Bli.iml,

S. BLUML ET AL.

335

S.; Schad, L.R.; Hawighorst, H.; van Kaick, G. 3D MPRAGE evaluation of lesionsin the posteriorcranial fossa.Magn. Reson.Imaging 12:553-558; 1994. 7. Shah,M.; Ross,J.C.; Vandyck, C.; Rudick, R.A.; Goodkin, D.E.; Obuchowski, N.; Medic, M.T. Volume T,weighted gradient-echoMRI in multiple sclerosispatients. J. Comput. Assist. Tomogr. 16:731-736; 1993. 8. Ross,J.S.; Ruggieri, P.M.; Tkach, J.A.; Masaryk, T.J.; Paranandi,L.; Dillinger, J.J.; Medic, M.T. Gd-DTPAenhanced3D MR imagingof cervical degenerativedisk disease:Initial experience.AJNR 13:127-136; 1992. 9. de Lange,E.E.; Mugler, J.P.; Bertolina, J.A.; Gay, S.B.; Janus,C.L.; Brookeman,J.R. Magnetization prepared rapid gradient-echo(MP-RAGE) MR imaging of the liver: Comparison with spin-echo imaging. Magn. Reson.Imaging 9:469-476; 1991. 10. Foo, T.K.F.; Sawyer, A.M.; Faulkner,W.H.; Mills, D.G. Inversion in the steadystate:Contrastoptimization and reducedimagingtime with fastthree-dimensional inversion-recovery-preparedGRE pulse sequences.Radiology 191:85-90; 1994. 11. Schad,L.R.; Bliiml, S.; Hawighorst,H.; Wenz, F.; Lorenz, W.J. Radiosurgicaltreatment planning of brain metastases basedon a fast, three-dimensionalMR imaging technique.Magn. Reson.Imaging 12:811- 819; 1994. 12. Zur, Y.; Wood, M.L.; Neuringer,L.J. Spoiling of transverse magnetizationin steady-statesequences.Magn. Reson.Med. 21:251-263; 1991. 13. Chien,D.; Edelman,R.R. Ultrafast imagingusinggradient-ethos.Magn. Reson.Quart. 7:31-56; 1991. 14. Holsinger,A.E.; Riederer,S.J.The importanceof centric phase-encodingorder in snapshot imaging. Magn. Reson.Med. 16:481-488; 1990. 15. Schad,L.R.; Ehricke, H.-H.; Wowra, B.; Layer, G.; Engenhart,R.; Kauczor, H.U.; Zabel, H.-J.; Brix, G.; Lorenz, W.J. Correction of spatialdistortion in magnetic resonanceangiographyfor radiosurgicaltreatmentplanning of cerebral arteriovenousmalformations.Magn. Reson.Imaging 10:609-621; 1992. 16. Mugler, J.P.; Brookman, J.R. Analysis of GD-DTPA enhancementin T,-weighted 3D MP RAGE sequences. JMRI 2:79; 1992. 17. Kaiser,R.; Bartholdi, E.; Ernst,R.R. Diffusion andfieldgradienteffectsin NMR fourier spectroscopy.J. Chem. Phys. 60:2966-2979; 1974. 18. Crawly, A.P.; Wood, M.L.; Henkelman,R.M. Elimination of transversecoherencesin FLASH MRI. Magn. Reson.Med. 8:248-260; 1988. 19. Sekihara,K. Steady-statemagnetizationsin rapid NMR imaging using small flip anglesand short repetition times.IEEE Trans.Med. ImagingMI-6:157-164; 1987.