Incidental magnetization transfer contrast in standard multislice imaging

Magnetic

Resonance

Ima&$ng.

Vol.

8, pp.

417-422,

0730-725X/90

1990

$3.00

+

.OO

Copyright 0 1990PergamonPress plc

Printed in theUSA. Allrightsreserved.

0 Original Contribution

INCIDENTAL MAGNETIZATION TRANSFER CONTRAST IN STANDARD MULTISLICE IMAGING W. THOMAS DIXON, HANS ENGELS, MAURICIO CASTILLO, AND MAZIAR SARDASHTI The Frederik Philips Magnetic Resonance Research Center of Department of Radiology, Emory University, Atlanta, Georgia 30322, USA Radio frequency magnetic fields from 5 to 100 kHz off resonance reduce longitudinal magnetization of water in some tissues without affecting the magnetization of simple solutions (Wolff, S.D.; Balaban, R.S. Magnetization transfer contrast (MTC) and tissue water proton relaxation in vivo. Magn. Reson. Med. 10:135-144; 1989). We demonstrate that off resonant fields used in day to day multislice imaging have a significant effect. Extra slices reduced brain (gray and white matter) and muscle intensities 10010-20% relative to fat, water, and paramagnetic solutions, a larger effect than produced by slice overlap. Implications for image contrast and quantitation are discussed. Keywords:

Magnetization

transfer contrast.

single off resonant frequency while several are used in multislice imaging. Is MTC significant in patient imaging, and if so, is it helpful or detrimental? One MTC experiment from ref 1 produced roughly a 50% effect after 4 set of irradiation 5 kHz off resonance with 300 Hz of RF in the rotating frame. No effort was made in ref. 1 to determine the duration of off resonant RF power needed to achieve the MTC effect. Resonant frequencies of 85 and 200 MHz were used with the same result. How do these conditions compare with clinical conditions? The 4 set of off resonant irradiation used in the experiments is longer than that seen in clinical images, nevertheless for two reasons this does not rule out significant MTC in clinical work. First, the fractional signal loss caused by MTC may be roughly constant as magnetization builds up. Second, in many clinical sequences (not T, -weighted) steady state magnetization does build up before observation, either because of long TR or low flip angle. (This takes less time in clinical imagers than in higher frequency research instruments.3,4) The Wolff and Balaban experiments were done

INTRODUCTION

Recently Wolff and Balaban discovered’ that an RF field up to about 20 kHz off resonance decreases the water proton longitudinal magnetization considerably in some substances. Simple dilute solutions of paramagnetic salts are not affected but tissues are to varying degrees. Only the water resonance (not fat) was affected in the tissues examined. Since the effect is stronger in some tissues than in others, it is possible this will prove a useful source of contrast in medical images. Wolff and Balaban call this effect magnetization transfer contrast or MTC. Images demonstrating this type of contrast have been presented. ‘J Although the idea of MTC is new, the effect is not. Wolff and Balaban irradiated tissues with off resonant power while the spin system was progressing to equilibrium or at least to steady state. This is exactly what happens during ordinary multislice imaging. Multislice imaging differs from these recent experiments only in the RF strength, duty cycle, and frequency offset used, and in that the experiments were clean, using a RECENED

10/28/89;

Address correspondence to Dr. W. Thomas Dixon, Department of Radiology, Emory University, Room 419 Woodruff Memorial Building, Atlanta, GA 30322, USA.

ACCEPTED 2/10/90.

Acknowledgments-The authors thank Drs. R. Balaban, R. Bryant, R. Henkelman, and D. Kelley for helpful conversations. 417

418

Magnetic Resonance Imaging 0 Volume 8, Number 4, 1990

with frequency offsets of at least 5 kHz. The effect is stronger at smaller offsets. Many multislice images are separated by 1 or 2 kHz per slice, so most of the incidental MTC effect in clinical images will likely come from offsets less than 5 kHz, making clinical images more sensitive than the original experiments, An Hi field of 300 Hz was used in the original experiments. That is about as strong as the peak Hi used in a clinical imager. The low RF duty cycle used in clinical imaging (a few percent) should counteract the effect of the small frequency offsets used clinically. Not being certain of the mechanism of MTC, it is hard to extrapolate from ref. 1 to clinical results. Should the effect scale with the average RF power or with the average Hi? This is equivalent to asking whether we are irradiating a homogeneously broad line or burning a hole in an inhomogeneous line with slow exchange between the bulk water and the hole in the resonance line. With duty cycle making MTC stronger in the original research, frequency offsets making it stronger in clinical imaging, and with both the original experiments and many clinical protocols allowing steady state to develop before observation, an experiment was needed to determine whether MTC affects day to day clinical imaging. MATERIALS

were monitored on an oscilloscope, the amplitude could be seen to vary from slice to slice within a single scan. This variation was probably l%-2%. Phantoms were included in the image with the subject. Tap water and a 4-mM solution of copper sulfate in distilled water were included in all images; 0.5 mM manganese (II) chloride in distilled water was included as well in the leg images. A gap between the subject and the phantoms can be seen in the images. This was filled with gauze sponges or a folded towel in an effort to minimize temperature changes and consequent signal changes from the phantoms during an examination. One centimeter thick, parasagittal slices of the head of one of the authors were taken just off midline. Slices were spaced on 1.8-cm centers, which with a gradient of 1.04 kHz/cm gave a frequency offset of 1.87 kHz/slice. TR values of 0.85 and 1.50 set were used for 2 sets of scans, allowing a maximum of 9 slices in one set, 17 in the other. The field of view was 300 mm. Transverse slices were taken above the knee of a different author. For these images, slices were 6-mm thick on 12-mm centers with a gradient of 1.56 kHz/cm leading to the same frequency spacing used in the head, 1.87 kHz/slice. The field of view was 320 mm. A 1 second TR allowed 11 slices.

AND METHODS RESULTS

A Philips Gyroscan using level 5.6 HP software and operating at 1.5 Tesla (64 MHz) was used. Linearly polarized, transmit, receive coils were used for both body and head imaging. Absolute value reconstruction was used and each image was separately scaled by the reconstruction software. Data were oversampled in the read direction allowing the ends to be discarded leaving images 256 pixels square with ends free of wrap around and audio filter fall off. Double-echo exams, 90-180-collect-180-collect, were used with echo times of 20 and 50 msec. In the work we present, scans were taken in groups of 4, varying the number of slices observed. A single slice scan was followed first by the maximum odd number of slices allowed by the chosen TR, then by a 3 slice scan and finally the single slice scan was repeated. The slices were excited in order of position, not in the default order of odd numbered slices before even. Slices were excited 80 msec apart when more than 1 slice was taken in a scan. Signal averaging was not used. Gradient and RF pulses were nominally identical in amplitude and profile for all slices within a single scan and for all scans within a set. When the RF pulses

One of the head images with regions of interest (ROI) marked in the splenium of the corpus callosum (mainly white matter) and in the thalamus (mainly gray matter), is presented in Fig. 1. ROIs were also examined in the phantom tubes and in marrow and subcutaneous fat. Table 1 contains averages and standard deviations of pixel values in ROIL Only the first echo of the center slice (or only slice) of each scan was evaluated. Data from Table 1 were resealed. All ROI pixel averages in each image were divided by the value for the copper solution appearing in that image. The copper value is therefore 1 in all images and the value is less for all other materials, representing roughly the fraction of the equilibrium proton magnetization. This value is a function of both the material and the number of slices in the scan. From these resealed numbers, the effect of the number of slices was isolated. For a given material (e.g., tap water) and set of scans (e.g., TR = 0.85), the polarization values for each scan (e.g., the 3-slice scan) were divided by the corresponding value for a single slice image, the average of the first and last scan

Incidental magnetization transfer contrast in standard multislice imaging l W.T.

DIXON ET AL.

419

Fig. 1. Typical images. (A) head. One of the TR = 1.5 set images is shown with regions of interest in the thalamus (1) and in the splenium of the corpus callosum (2) marked. The darker phantom is tap water, the brighter one contains copper sulfate. (B) legs. TR = 1 sec. Regions of interest are CuS04 (l), MnC12 (2) tap water (3), marrow fat (4), subcutaneous fat (5-7), musdes @-lo), numbered as in Table lc.

being used for this divisor.

Figure

2 presents

natural

logarithms of these quotients as “slice effect.” In the leg study, regions were taken in all three solutions, bone marrow fat, subcutaneous fat, and in muscles of both legs. Data were handled in exactly the same manner as with the head. An image appears in

Fig. 1, numerical and graphical results are in Table 1 and Fig. 2. We found definite signal losses which grew with the number of slices in white matter, gray matter, and muscles. An effect in the head noted incidentally is that mul-

Magnetic Resonance Imaging 0 Volume 8, Number 4, 1990

420

Table 1. Average pixel values Number of slices/ average pixel value

a

b

c

Region #

1

9

3

1

1914 747 374 1294 860 850

1928 740 379 1337 732 776

1925 737 373 1325 842 841

1923 736 388 1303 879 883

1

17

3

1

1921 1043 500 1296 1017 1078

1925 1014 457 1303 884 951

1930 1010 478 1424 978 1042

1915 1004 557 1298 1047 1125

1

11

3

1

1916 1121 692 1546 1543 1437 1412 560 617 634

1944 1112 704 1510 1512 1382 1402 478 520 550

1929 1112 696 1516 1515 1406 1416 524 570 591

1911 1120 665 1536 1528 1434 1436 544 609 631

-

8 9 10

Substance

SD

4mMCu tap water muscle subcut. fat white gray

44 34 22 23 11 32

4mMCu tap water muscle subcut. fat white gray

47 38 25 14 20 10

4mMCu 0.5 mM Mn tap water marrow subcut. fat subcut. fat subcut. fat muscle muscle muscle

49 33 32 50 47 80 112 35 33 63

Average pixel values. (a) head, TR = 0.85 set; (b) head, TR = 1.50 set; (c) leg, TR = 1 sec. Region numbers correspond to Fig. l(c). Standard deviations are given for single slice repeat scans (column 4) only.

tiple

slices

darken

the

sagittal

sinus

and

reduce

artifacts. DISCUSSION M

=

~~

e

-TE/T2 [

1 _

e

--TR/V ]

This simple equation for tissue proton magnetization has long been used to explain contrast and to solve for the tissue parameters TI and T2. While this equation

may give excellent results in spectrometers, it does not represent image behavior very well,5 even for the comparatively simple case of single slice imaging. Multiple slice imaging is troubled by the effects of slice overlap and, for some substances, the effect discovered by Wolff and Balaban. Careful design of selective pulses brings slice profiles closer to the ideal rectangle, or at least to the ideal of nonoverlapping profiles. Neither ideal is possible and overlaps are sig-

nificant6*7- wings or side lobes of one slice can decrease magnetization in nearby, nominally nonoverlapping slices. Our interest lies in MTC rather than overlap effect. First we had to determine the effect of multiple slices on image intensity, then we had to separate MTC from slice overlap effects. Multiple slice effects were measured by comparing multislice to single slice images. Logarithms were used in the hope of getting additive results. Since different scaling factors had been used to reconstruct our original images, we normalized all intensities in every image to the intensity of the copper sulfate phantom in that image. This phantom was assumed not to show MTC and to produce a reasonably pulse sequence independent signal. We used a three-pronged approach to showing MTC over and above the overlap effect. First we used gaps (80%-100% of slice thickness) between slices, assuming that the overlap effect falls off faster with distance than does MTC. Second we compare an image with no neighbors, an image with only one neighbor on each side, and an image with many neighbors on each side. If our assumption that overlap effects fall off rapidly is correct, any differences between the image with many neighbors on each side and the image with only one on each side are due to MTC. Overlap may affect only the immediate neighbors. In order to make a proper comparison it was necessary to excite the slices in order of position, not odd-numbered slices before even, the default order of our instrument and many others. With the linear order, the time between observation of the center slice and excitation of its immediate neighbors is the same whether three or more slices are examined. With the odd then even order this timing depends on the number of slices imaged. Third we included both tap water and copper sulfate solution as reference materials. The tissues we examined have Tl values between those of the copper sulfate solution and water. This is very important because simple overlap effects vary with interslice time and T,. There is no overlap effect if T, is much shorter than the interslice time, therefore the overlap effects are greatest in the substance with the longest T,. The tissues we examined have no more overlap effect than does the tap water-and we did not see any. Perhaps the non-MTC effects depend on the signal intensity of a substance. Our use of both bright (copper) and dim (tap water) reference materials guard our conclusions against this possibility. Another scientific issue must be considered before attributing the effect shown in Fig. 2 to MTC. Cross relaxation between different proton spectral lines

Incidental magnetization transfer contrast in standard multislice imaging 0 W.T. DIXONETAL.

0.05 1

421

-

4mM Cu

-

tap water

-

OSmM Mn

+

muscle

-0.00

-0.05

a-. sutxlJt. fat

-0.10

Q

marrow

-o- white * -0.20 1

9

3 A

Fig. 2. Slice effect versus number

1

1

17

3 B

1

1

I

I

1

11

3

1

gray

C

of slices. (A) head, TR = 0.85 set; (B) head, TR = 1.5 set; (C) legs, TR = 1 sec.

could also complicate contrast in images.’ Examining magnetization recovery of methylene groups in dog bile following nonselective inversion, Gore et al. found a 25% overshoot 1 set after inversion. Different results follow selective inversion of either the water or methylene line. Similar but weaker effects might be seen 1 set after saturation (rather than inversion) as in our experiments. Image signal changes could result from both selective and nonselective saturation (or inversion depending on the pulse sequence). In our images the slices were about 1 kHz thick, the water and fat lines about 200 Hz apart, so about 80% of a slice underwent nonselective saturation while 20% saw a selective saturation. With slices close to each other, manipulation of water spins in one slice could affect fat in an adjacent slice and vice versa. We doubt this was significant in our results because our interslice frequency gap was well over the water fat chemical shift difference. The effect of ref. 8 may give us fat intensities which do not follow Eq. (1). However, because of our slice gap, cross relaxation will not be affected by the number of slices imaged. Therefore, cross relaxation does not influence Fig. 2 (interaction of cross relaxation and multiple slices is not ruled out for clinical images made with reduced gaps). Technical details as well as scientific arguments affect the interpretation of our results. In the head, our “slice effect” was always obvious for brain and always increased with increasing number of slices. Results in

other materials were inconsistent. This appeared to be size related with consistent results obtained in large uniform regions of the image (brain) but not in small areas. Image data were analyzed 2 ways. Table 1 data were taken from ROIs with identical pixel addresses. In an attempt to follow any shifts between images in a set, we also examined ROIs set by observing landmarks. Consistency for small regions was poor using either method. This led us to examine the legs, to use larger phantoms, and to include a manganese phantom (which has a short r,). We are confident in the consistent effects seen in the leg. Less consistent data from smaller muscles and fat regions in the head are included in the table but not in the graph. We have no explanation other than MTC for the lo%-20% signal losses seen repeatedly in brain and muscle. We found no obvious difference between the strength of the MTC effect in white matter, gray matter, or muscle. Under common imaging protocols which examine the maximum allowed number of slices, MTC is present but not strongly TR or slice number dependent. Given MTC effects of over 10% in multislice imaging, is MTC good or bad for images? In cases where stronger MTC and faster longitudinal relaxation go together, MTC wiil oppose T,-induced contrast. This appears to be the case in the kidney, where the cortex has greater MTC and faster relaxation rates than the medulla.’ Contrast should decrease as more slices are examined at constant TR.

422

Magnetic Resonance Imaging 0 Volume 8, Number 4, 1990;

MTC will confound efforts to choose between short TR, short TE or inversion recovery sequences based on Eq. (1). Long TR, long TE images can be made with one echo or with a short TE echo in addition to the long TE, heavily &weighted echo. Introduction of the short TE echo may increase contrast in the long TE image if faster transverse relaxation is accompanied by stronger MTC. It appears that prediction of image contrast requires knowledge of tissue T, , T2 and MTC at a minimum. The lo%-20% MTC effect seen here may have a significant effect in the choice of 3D versus 2D multislice techniques because off resonant pulses are used in 2D but not in 3D imaging. Assuming perfectly rectangular slice profiles and identical voxel matrices with a small enough number of slices for multislice imaging in 1 pass-and ignoring MTC-imaging time and signal to noise ratio should be identical in 2D and 3D imaging. MTC lowers the signal (which may increase or decrease contrast) in the 2D images. The most important conclusion regarding MTC may be that quantitative measurement of tissue TI or fat fraction from multislice images is inadvisable. Our results show fat fraction measurements for muscle are certainly inaccurate and the same is likely to be true of most other tissues. If multislice Tl measurements of muscle or brain give correct answers, it is pure coincidence. Multislice Tz measurements may not be affected, but nothing said here should be construed to mean that such T2 measurements are valid.

REFERENCES 1. Wolff, S.D.; Balaban, R.S. Magnetization transfer contrast (MTC) and tissue water proton relaxation in vivo. Magn. Reson. Med. 10:135-144; 1989. 2. Kelley, D.A.C.; Graham, R.; Kappler, F.; Kowalyshyn, M.; Keller, S.; Brown, T.R. Magnetization transfer contrast imaging at 2.3T. Sot. Magn. Reson. Med. Abstract Book works in progress 1989; p. 1036. 3. Koenig, S.H.; Brown, R.D. III; Adams, D.;‘Emerson, D.; Harrison, C.G. Magnetic field dependence of l/T, of protons in tissue. Investig. Radiol. 19:76-81; 1984. P.A.; Foster, T.H.; Argersinger, R.E.; 4. Bottomley, Pfeifer, L.M. A review of normal tissues hydrogen NMR relaxation times and relaxation mechanisms from l-100 MHz: Dependence on tissue type, NMR frequency, temperature, species, excision, and age. Med. Phys. 11:425428; 1984. 5. Rosen, B.R.; Pykett, I.L.; Brady, T.J. Spin lattice relaxation time measurements in two-dimensional nuclear magnetic resonance imaging: Corrections for plane selection and pulse sequence. J. Comput. Assist. Tomogr. 8: 195-199; 1984. 6. Kucharczyk, W.; Crawley, A.P.; Kelly, W.M.; Henkelman, R.M. Effect of multislice interference on image contrast in T2- and Ti-weighted MR images. Am. J. Neuroroentgen. 9~443-451; 1988. 7. Schwaighofer, B.W.; Yu, K.K.; Mattrey, R.F. Diagnostic significance of interslice gap and imaging volume in body mr imaging. Am, J. Roentgen. 153:629-632; 1989. 8. Gore, J.C.; Brown, M.S.; Armitage, I.M. An analysis of magnetic cross-relaxation between water and methylene protons in a model system. Magn. Reson. Med. 9:333342; 1989.