1H-spectroscopic imaging using a modified Dixon method

Magnetic Resonance Printed in the USA.

Imaging, All rights

Vol. 6, pp. 617-622, reserved.

1988 Copyright

0730-725X/88 S3.W + .OLl 0 1988 Pergamon Press plc

0 Original Contribution

‘H-SPECTROSCOPIC GUNNAR

BRIX,

IMAGING LOTHAR

USING A MODIFIED R. SCHAD,

From the Institute of Radiology and Pathophysiology,

DIXON

AND WALTER

METHOD

J. LORENZ

German Cancer Research Center, Heidelberg, FRG

Inhomogeneities of the static magnetic field and the different susceptibilities of the various types of tissue are a serious problem for all imaging methods of spectral separation of fat and water. In the Dixon method this problem is solved by using the absolute values of the image signals for the separation. In image regions where the fat signal is greater than the water signal, however, this results in an incorrect assignment of the computed solutions. A modified Dixon method was developed to easily carry out the spectral separation completely over the entire image by interactively building up a phase correction matrix after the data acquisition. The spectral delineation of the fat tissue finds an interesting application in the treatment planning with fast neutrons in accounting for the increase in dose. Keywords: Spectroscopic

imaging; Dixon method; Neutron treatment planning.

lead to field inhomogeneities.* In light of this, a modified Dixon method was developed which enables us to easily carry out the spectral separation over the entire image.

INTRODUCTION

Hydrogen nuclei imageable by MR almost exclusively occur in the water and in the CH2-groups of the aliphatic acids. Due to the different chemical environment of the hydrogen nuclei in these two groups, the proton resonance in the MR spectrum splits into a water and a fat peak, whose resonance frequencies differ by 3.5 ppm. In the past, several methods have been developed

MATERIALS

AND

METHODS

The study was performed using a MAGNETOM 1.5 T (Siemens, Erlangen, F.R.G.) superconducting whole body imager. The MR computer (VAX1 l/750, DEC, Maynard, MA, U.S.A.) was connected directly (DEC-NET, DEC, link) to a central computer (VAX1 l/780, DEC), where the fat-water images were computed. The correction was carried out interactively on a RAMTEK 2020 (Ramtek, Santa Clara, CA, U.S.A.) color graphic system (1024 x 1280 x 8 Bit). According to the Dixon method our sequence records in-phase and opposed images linewise alternat ing with a single scan to minimize motion artifacts. The signals S1 (x) and S,(x) of the measured complex images can be written for a pixel at position x:

which make use of this chemical shift for spectroscopic imaging. l-5,7 In the Dixon method’ an in-phase image and an opposed image is acquired. After the data acquisition, separate water and fat images can be calculated on the base of these two images. In image regions, where the fat signal is greater than the water signal, however, this method results in an incorrect assignment of the computed solutions, that is, the fat signal is taken for the water signal and vice versa. Other methods, such as CHESS3 and STEAMCHESS’ use selective 90”-pulses, which only affect the hydrogen nuclei of one chemical group. This, however, requires that the inhomogeneity of the static magnetic field over the imaged region is smaller than the chemical shift. For high field whole body scanners, however, this condition is extremely difficult to realize. In addition, local tissue susceptibilities also

s1 (x) = F(x) * (&v(x) Sz(x)

Received 1l/7/87; ACCEPTED 3/24/M. The authors would like to thank Dr. Dieter Schlaps for helpful suggestions and Dr. Karl-Heinz Hijver for discussions on the field of neutron treatment planning.

=

r(x)

*P(X)

+ S/(X))

* (&v(x)

- Sf(XN

(1) (2)

Address correspondence and reprint requests to Gunnar Brix, Institut fiir Radiologie und Pathophysiologie, Deutsches Krebsforschungszentrum, POB 101949, D-6900 Heidelberg, FRG. 617

618

Magnetic Resonance Imaging 0 Volume 6, Number 6, 1988

with r (x) a space and sequence dependent phase factor, +(x) and S,(x) are the fat and water signals and p(x) is an additional phase factor originating from the delayed acquisition of the opposed image. The phase factor r(x) describes RF phase errors, gradient effects and B0 inhomogeneities which are the same in both images, where p(x) reflects mainly B0 inhomogeneities appearing only in the opposed image. Note that the phase effects are not necessarily the same for fat and water components since they come from different locations. With B0 = 1.5 T and a sufficiently large read gradient of about 1.5 mT/m the fat component is shifted by two pixels for a 256 x 256 matrix with respect to the water component. For the fat/water separation, however, the same phase factors can be used in a good approximation for both components The unknown phase factor p(x) can be eliminated by taking the absolute values of the measured image signals according to: IS* (x)1 = &v(x) + $(x)

(3)

I&(x)1 = *(&v(x)

(4)

- Sf(X)) .

The ambiguity of the sign in Eq. (4) leads to two solutions for the fat and water signals. In the original Dixon method the solution corresponding to the positive sign was used. This, of course, requires that the water signal is always larger than the fat signal. If this condition is not met the fat signal is taken for water signal and vice versa. The actual problem of the Dixon method therefore consists in deciding which of the two solutions in a certain volume element is the correct one. In order to obtain an unambiguous criterion for this decision an iterative and interactive method has been used in the following way: 1. considering a pixel at position x0 where S, is greater than S, (i.e. muscle, brain tissue) the phase factors I’ (x0) and p(xo) can be evaluated according to Eqs. (1) and 2. Next, for all other pixels of the image the complex quantity R(x) is computed by: R(x) =

S2(X)/(r(X)

= exp(Wx)).

‘P(XO)) (&(x)

- $(x1)

,

(5)

with r(x) = Si(x)/(Si(x)l and exp(i+(x)) an unknown phase difference between pixel x, and pixel x, caused by B. inhomogeneities and differences in the susceptibilities. If the real part of R(x) is greater than zero we conclude that S, is greater than S,. Using

these conditions “fat” and “water” images were computed in the first step of the iterative procedure. 2. with respect to phase continuity arguments the assumption of I ‘P( x)1 < 7r/2 holds for pixels near by position x0. In this case “fat” and “water” signals of the calculated images are interpreted correctly (Fig. 1B). At pixels with I@(x)1 > 7r/2 the fat signal is taken for water signal and vice versa and we get wrong solutions in the calculated “fat” and “water” images (Fig. 1C). But it is exactly this extensive and therefore easily visible exchange, which makes it possible to do a phase correction interactively. A region of interest (ROI) is placed into the displayed image, which amply surrounds the area of the wrong solution. For all pixels inside the ROI the complex quantity R(x) is being rotated by + ?r/2 which lead to correct solution for pixels with the wrong and right solution (Figs. Id and le). In general, surrounding the areas with the wrong solution amply (for example in a distance of 1 cm) means that the field inhomogeneity inside this surrounding strip may not be larger than about 1.5 ppm. However, this condition is so weak that it practically is no restriction. If necessary the described procedure can be repeated. RESULTS

Fat/ Water Separation As an example, Fig. 2 shows the original images of the pelvis of a patient with hairy cell leukemia. Figs. 2A and 2B show the in-phase and the opposed images. In Fig. 3 the right halves of “fat” and “water” images are displayed simultaneously in order to achieve the same gray value distribution for comparison. One ROI necessary for the phase correction is outlined. The resultant fat and the water image for the entire image are shown in Fig. 4. Neutron Therapy Planning The quantitative knowledge of the hydrogen concentration in the tissue is an indispensable prerequisite for the individual planning of a therapy with fast neutrons, since the recoil nuclei of the H(n,n)H reaction supply the main share of the energy deposited in the tissue. Unfortunately, the quantitative determination of the proton density from MR tomograms is not possible for two reasons: (a) inhomogeneities in the illumination of the tomograms cannot be corrected to the necessary extent by phantom images, since the different filling factors of patient and phantom strongly influence the resonance behavior, (b) only part of the protons of the tissue contributes to the image signal, since the share of protons with short T2 times (T2 << TE) has relaxed at the time of data acquisition.

‘H-Spectroscopic

PIXEL

‘Ti

619

imaging 0 GUNNARBFCIXET AL.

PIXEL

-?

((2

0)

03

Fig. 1. (A) In order to demonstrate the effect of the phase correction, consider two pixels inside a marked-off ROI, one of which lies in the area of the correct solution (pixel x), and the other lying in the area of the exchanged solution (pixel y). In each case let the water signal be greater than the fat signal. (B) and (C) The quantity R is shown in the complex plane (u + iv) in case of 3 < s/2 and @ > n/2. (D) and (E) show the situation after rotation about an angle of -r/2 leading to correct solutions in both cases.

Magnetic Resonance Imaging 0 Volume 6, Number 6, 1988

(A)

(B)

Fig. 2. (A) In-phase and (B) opposed spin echo images (TR/TE 1000/22 ms) of a patient with hairy cell leukemia. The welldefined delineation of the fat tissue against the muscle tissue by low signal structures in the opposed image results from the

fact that in these locations the fat signal is equal the water signal.

ing field of 5 x 5 cm2 for a left thigh. Figure 5A shows the treatment plan without inhomogeneity correction. Figure 5B shows the corresponding plan, which has been computed on the base of the inhomogeneity matrix. Note that, in contrast to Fig. 5A, in Fig. 5B the 90% isodose completely lies inside the fat tissue. Since in both plans the maximum dose is normalized to lOO%, the increase in dose in the fat tissue therefore leads to a shift of the isodoses in direction of the beam penetration point. DISCUSSION

Fig. 3. The right halves of the “fat/water” images are displayed simultaneously in order to achieve the same gray value distribution for comparison. One ROI necessary for the phase correction is outlined.

On the base of the fat/water images it is possible, however, to construct an inhomogeneity matrix, which delineates the fat tissue and the bone regions against the normal tissue. Therefore the local increase in dose in the fat tissue (due to the higher proton density) can be taken into account at least globally in the planning stage.6 To demonstrate the importance of the inhomogeneity correction, we computed two plans for a stand-

The inhomogeneities of the static magnetic field and the different susceptibilities of the various tissue types are a serious problem for the spectral separation of fat and water in the MR tomogram. To eliminate these disturbing effects, we use-just as in the original Dixon method-the absolute values of the image signals for the calculation of the solutions. A modified Dixon method, however, enables us to decide which of the two possible solutions must be taken in a certain volume element. If needed, we interactively build up a phase correction matrix on the display, with which the spectral separation of the two components can be done completely. An exchange of fat and water signal is easily seen in the calculated images and the interactive correction method can also be used in the presence of large gaps (i.e. nasal cavity, bone regions). Up to now, the phase correction algorithm was applied successfully to more than 30 patients. Since the disturbing field inhomogeneities can be eliminated, the spectral separation is done equally well

‘H-Spectroscopic

imaging 0 GUNNARBRIXET AL.

621

C-V Fig. 4. (A) Water

W and (B) fat image of the pelvic after complete

correction.

@I

(4

Fig. 5. Neutron treatment plan of a left thigh for a standing field of 5 x 5 cm* calculated (A) without inhomogeneity correction and (B) with inhomogeneity correction. Note that the increase in dose in the fat tissue consequently leads to a dislocation of the isodoses in direction of the beam penetration point (in (B) the 90% isodose lies completely inside the fat tissue).

for all pixels. This is an important advantage over the frequency selective CHESS methods, which frequently lead to an incomplete fat-water separation especially in the fringe area of the images. Furthermore, no time consuming shimming on the individual patient is necessary as for the CHESS methods and the actual measurement is as fast and as easy as it is in conventional imaging. As has been shown, the spectral separation of fat and normal tissue offers the possibility of a purely MR-based

neutron

therapy

planning.

Thereby

it is

possible to make use of the superior soft tissue contrast of the MR images for the planning of a therapy with fast neutrons.

REFERENCES 1. Buxton, R.B.; Wismer, G.L.; Brady, T.J.; Rosen, B.R. Quantitative proton chemical-shift imaging. Mugn. Reson. Med. 3:881-900; 1986. 2. Dixon, W.T. Simple proton spectroscopic imaging. Radiology 153:189-194; 1984.

Magnetic Resonance Imaging 0 Volume 6, Number 6, 1988

622

J.; Haase, A.; Hanicke, W.; Matthaei, D.; Bomsdorf, H.; Helzel, T. Chemical shift selective (CHESS) MR imaging using a whole-body magnet. Radiology 156:

3. Frahm,

441-444; 1985. 4. Haacke, E.M.; Patrick,

J.L.; Lenz, G.W.; Parrish, T. The separation of water and lipid components in the presence of field inhomogeneities. Rev. Magn. Reson. Med. 1(2):123-154; 1986. 5. Haase, A.; Frahm, J. Multiple chemical-shift-selective NMR imaging using stimulated echoes. J. Magn. Res. 64:94-102; 1985. 6. Hover, K.H.; Lorenz, W.J.; Scharfenberg, H.; Schlegel,

W. Treatment planning at the Heidelberg neutron therapy facility. G. Burger, ed., Treatment planning for external beam therapy with neutrons. Miinchen, Wien, Baltimore; Urban und Schwarzenberg; Supplements to “Strahlentherapie”, 1981; 77: 116-122. 7. Yeung, H.N.; Kormos, D.W. Separation of true fat and water images by correcting magnetic field inhomogeneity in situ. Radiology 159:783-786; 1986. 8. Young, I.R.; Khenia, S.; Thomas, D.G.T.; Davis, C.H.; Gadian, D.G.; Cox, I.J.; Ross, B.D.; Bydder, G.M. Clinical magnetic susceptibility mapping of the brain. J. Comput. Assist. Tomogr. 11(1):2-6; 1987.