Stephan E. Maier, MD, PhDPeter Bogner, MD, PhDGabor Bajzik, MDHatsuho Mamata, MDYoshiaki Mamata, MDImre Repa, MDFerenc A. Jolesz, MDRobert V. Mulkern, PhD

Index terms:Brain, edema, 10.86Brain, gray matterBrain, white matterBrain neoplasms, MR, 10.121411,

10.12143Magnetic resonance (MR), contrast

enhancement, 10.12143Magnetic resonance (MR), diffusion

study, 10.12144Magnetic resonance (MR), tissue

characterization

Radiology 2001; 219:842–849

Abbreviations:ADC 5 apparent diffusion coefficientA1 5 signal amplitude for ADC1

A2 5 signal amplitude for ADC2

ROI 5 region of interest

1 From the Departments of Radiology,Brigham and Women’s Hospital (S.E.M.,H.M., Y.M., F.A.J.) and Children’s Hospital(R.V.M.), Harvard Medical School, 75Francis St, Boston, MA 02115; and theDiagnostics Center, Pannon University ofAgriculture, Kaposvar, Hungary (P.B.,G.B., I.R.). Received June 27, 2000; revi-sion requested August 7; revision receivedSeptember 6; accepted September 11.S.E.M. supported by grants from the Whit-aker Foundation and the National Insti-tutes of Health (NIH 1R01 NS39335-01A1). P.B. supported by Hungarian Sci-entific Research Fund (OTKA grants F016343, T 034200). Address correspon-dence to S.E.M. (e-mail: stephan@bwh.harvard.edu).© RSNA, 2001

Author contributions:Guarantor of integrity of entire study,S.E.M.; study concepts, S.E.M., R.V.M.;study design, S.E.M., P.B.; definition ofintellectual content and editing, S.E.M.,P.B., R.V.M.; literature research, P.B.;clinical studies, P.B.; data acquisition,P.B., S.E.M.; data analysis, P.B., G.B.,S.E.M., H.M., Y.M.; statistical analysis,S.E.M., P.B., G.B., H.M., Y.M.; manu-script preparation, S.E.M., P.B.; manu-script review, R.V.M., F.A.J., I.R.; manu-script final version approval, S.E.M.,R.V.M.

Normal Brain and BrainTumor: MulticomponentApparent DiffusionCoefficient Line ScanImaging1

Magnetic resonance line scan diffu-sion imaging of the brain, with diffu-sion weighting between 5 and 5,000sec/mm2, was performed in healthysubjects and patients with a 1.5-Tmachine. For each voxel, biexponen-tial signal decay fits produced twoapparent diffusion constants and re-spective signal amplitudes. Imagesbased on these parameters show po-tential for use in the differentiation ofgray and white matter, edema, andtumor.

Tissue characterization is a basic issue inmagnetic resonance (MR) imaging. Infact, the concept of distinguishing nor-mal and tumor tissue with MR imaginggoes back to the observation of Dama-dian (1), who described substantial differ-ences in T1 and T2 between normal andcancerous tissue. Since then, methods toobtain T1- or T2-weighted images haveimproved dramatically, and considerableexperience has been gained with the invivo application of these methods withthe use of contrast agents. Nevertheless,determination of the tumor marginsolely on the basis of contrast enhance-ment on T1- or T2-weighted images isnot successful in every case (2,3).

In several recent publications (4–9),diffusion-weighted imaging has beenproposed as a novel mechanism for pro-ducing contrast in the demarcation ofdifferent cerebral tumors. Apparent diffu-sion coefficient (ADC) maps of brain tu-mors seem also to provide useful infor-mation about structural details of tumors(5,7,8). According to these reports, peri-tumoral edema and solid enhanced, solidnecrotic nonenhanced, and cystic parts

can be recognized on ADC maps. Diffusiontensor imaging adds information about thedirectional dependence of molecular diffu-sion that may also be helpful in the demar-cation of tumor margins (10). Neverthe-less, any of these new diffusion imagingmethods used with contrast material–en-hanced relaxation-weighted imaging failsto be specific enough in every case (5,6).

Routine diffusion imaging of the braingenerally involves the use of b factorswithin the range of 0 to 1,000 sec/mm2.ADC maps are then generated, based onthe assumption that the relationship be-tween the MR signal and b factor ismonoexponential. Recently, however, itwas shown (11) that for rat brain, thesignal decay with b factors in an ex-tended range of up to 10,000 sec/mm2 isbetter described with a biexponentialcurve. Similar findings were made in hu-man brain by using multiple b factors ofup to 6,000 sec/mm2 (12). Both studieslack the anatomic details needed for clin-ical application, since diffusion was mea-sured only within a localized volume (11)or along a column (12).

In the present study, our goal was toobtain diffusion-weighted images of thehuman brain with b factors ranging from5 to 5,000 sec/mm2. Biexponential fitswere applied to the measured signal ofeach voxel. With the four parametersthat describe the biexponential fit, weattempted to characterize normal whiteand gray matter, edematous white mat-ter, and brain tumors.

Materials and Methods

Healthy Subjects and Patients

This study included two healthy sub-jects and 15 patients, including 14 pa-tients with brain tumors (seven men,seven women; age range, 33–79 years;

842

mean age, 56 years) and one patient withsubacute stroke. In the healthy subjects,brain images obtained for localization re-vealed no abnormalities. The patientswith tumor included two with astrocy-toma, eight with glioblastoma, and fourwith metastases. All diagnoses were con-firmed at preoperative biopsy or surgicalresection. All studies in healthy subjectsand patients were performed within theguidelines of the institutional internal re-view board. Informed consent was ob-tained from healthy subjects, patients, ortheir authorized representatives.

MR Diffusion Imaging

Each patient underwent the routineclinical imaging protocol. This protocolincluded T2-weighted (3,000–3,600/80–98, repetition time msec/echo time msec;field of view, 220–240 3 220–240 mm;section thickness, 3–5 mm) and T1-weighted (500–700/14–25; field of view,220–240 3 220–240 mm; section thick-ness, 3–5 mm) spin-echo sequences be-fore and after contrast enhancement(Magnevist [gadopentetate dimeglumine];Berlex Laboratories, Wayne, NJ; 0.2 mLper kilogram of body weight adminis-tered in the cubital vein) to localize thetumor mass and other pathologic details.For that obvious reason, in most casesdiffusion imaging was performed onlyafter the administration of contrast ma-terial. In four patients, the section fordiffusion imaging of tumor tissue was de-termined with nonenhanced images, anddiffusion imaging was performed prior to

the administration of contrast agent.One patient with tumor underwent dif-fusion imaging before and after the ad-ministration of contrast material.

Diffusion-weighted images with a widerange of b factors were obtained with linescan diffusion imaging. Aspects of the MRphysics and the feasibility of this single-shot column-sampling technique havebeen presented (13). Patient studies withline scan diffusion imaging (14,15) havedemonstrated the usefulness of this tech-nique for conventional diffusion imaging.The line scan diffusion sequence was im-plemented by using a 1.5-T whole-bodysystem (Signa Echospeed; GE Medical Sys-tems, Milwaukee, Wis) with version 5.7software. The maximum gradient strengthwas 22 mT/m. The standard birdcage coilwas used, and neither cardiac gating norhead restraints were used.

Images were acquired with a rectangularfield of view of 220 3 165 mm and a ma-trix size of 64 3 48 or 126 3 96 columns.The effective section thickness (13) was setat 6.0–7.3 mm. The receiver bandwidthwas set at 6.25 kHz, which was found to bethe best compromise in view of the de-creased signal-to-noise ratio at higher band-widths and the augmented image distor-tions caused by field inhomogeneities orchemical shifting at lower bandwidths. Six-teen images with linearly increasing diffu-sion weighting between 5 and 5,000 sec/mm2 were acquired.

For patient imaging, diffusion wasmeasured along only a single direction,in a (1,1,1) gradient configuration to

achieve maximal diffusion encodingwith a minimal echo time of 94 msec. A bfactor of 5,000 sec/mm2 was attainedwith trapezoidal gradient pulses of 36.2mT/m amplitude, 40-msec pulse dura-tion (d), and 46 msec between the onsetof the first and second gradient pulses(D). In healthy subjects, data along sixnon-colinear directions were collectedwith the tensor configuration describedby Basser and Pierpaoli (16) by using anecho time of 107 msec. The repetitiontime and effective repetition time (13)were 204 and 2,040–3,600 msec, respec-tively; the total imaging time was 3 min-utes per section (one diffusion directionwith 16 b factors). Shorter repetition andimaging times would have been possiblehad gradient heating not been a concern.

Data Analysis

Data analysis was performed offline atworkstations (Sun Microsystems, Moun-tain View, Calif ) by using MATLAB soft-ware (Math Works, Natick, Mass). A non-linear least-squares Marquardt algorithmwas used for each pixel to fit brain signalintensity decay S with diffusion-weight-ing b to a biexponential function of thefollowing form: S 5 A1 exp(2ADC1b) 1A2 exp(2ADC2b), where ADC1 and ADC2

are the ADCs, with the signal amplitudesfor ADC1 (A1) and ADC2 (A2).

Data points were included in the fitonly if their signal exceeded three timesthe noise baseline, defined by the meansignal intensity in the four corners of theimage. Changes in the calibration of theMR apparatus do not permit a directcomparison among biexponential signalamplitudes measured during different ac-quisitions. We therefore calculated therelative fraction of the biexponential sig-nal amplitude of the slow diffusing com-ponent as follows: A2/(A1 1 A2). Prelimi-nary experiments with a phantom thatsimulates the geometry and load of a hu-man head (head SNR phantom model 46-287900G3; GE Medical Systems) showedthat within a 170-mm diameter the sig-nal variations are 4.4%. We consideredthese variations small enough to permitnormalization of individual tissue values(A1T and A2T) with individual white mat-ter values (A1W and A2W) measured at adifferent location.

Maps of the fit parameters were used forregion-of-interest (ROI) analysis with ded-icated image-analysis software (XPHASE) de-veloped at our institution. With this pro-gram, the contour drawing process can besimultaneously controlled on different im-age backgrounds (eg, T1-weighted, T2-

Figure 1. Transverse line scan diffusion-weighted MR images of a normal brain (3,600/94; 128 396 matrix; field of view, 220 3 165 mm; section thickness, 6 mm). The image on the left is a tracediffusion-weighted image with a typical b factor of 1,080 sec/mm2. The other two images wereobtained with very high diffusion weighting of 5,000 sec/mm2: The image in the middle is tracediffusion weighted, whereas the image on the right is monodirectional diffusion encoded in a(1,1,1) gradient configuration. With very high diffusion weighting, white matter tracts areenhanced, for example, the internal capsule (solid arrowhead). On the monodirectional diffu-sion-encoded image, anisotropic diffusion leads to selective enhancement of white matter tracts,for example, the right optical tract and left half of corpus callosum (open arrowheads), that arenot aligned with the diffusion encoding direction.

Volume 219 z Number 3 Multicomponent Apparent Diffusion Coefficient Line Scan Brain Imaging z 843

weighted, and ADC). ROIs were drawnmanually by two of the authors (P.B., G.B.)in consensus.

The ROI for tumor tissue was definedaccording to contrast enhancement on theconventional T1-weighted images. Sepa-rate ROIs were drawn for cystic parts oftumor lesions. The ROI for peritumoraledema was defined with the use of nonen-hanced T2-weighted images and contrast-enhanced T1-weighted images. Nonpatho-logic periventricular white matter areaswere selected as a healthy control. To re-duce the influence of directional diffusion,the mean value of two independent andlarge white matter ROIs was used. Graymatter values were determined in the cor-tex. In all cases, ROI size and placementwere selected so that bordering tissues ofambiguous origin were excluded as muchas possible. Moreover, to avoid partial vol-ume effects in the ROI analysis of the rela-

tively thick sections used at line scan dif-fusion imaging, considerable effort wasmade to determine that the tissue of inter-est was also present on neighboring sec-tions of the conventional images.

ROI sizes for tumors ranged between 5and 29 pixels (mean, 15.4 pixels or 182mm2). All other ROIs were, on average,larger; that is, there were 27.7 pixels forcysts, 24.6 pixels for peritumoral edema,62.7 pixels for gray matter, and 25.4 pix-els for gray matter. Significant differencesamong mean ROI values were verifiedwith a two-sided Student t test, with Pvalues less than .05 considered to indi-cate a significant difference.

Results

Figure 1 shows a diffusion-weightedimage of the brain of a healthy subject

with a b factor typically used for routineclinical examination, along with a diffu-sion-weighted image acquired with a bfactor of 5,000 sec/mm2, that is, a higher-than-usual b factor. Figure 2 shows mapsof ADC1, ADC2, A1, and A2, computedfrom an image obtained with a widerange of b factors in another healthy sub-ject. On the ADC2 and A2 maps, values inthe ventricles were zero, since the signaldecay in cerebrospinal fluid is monoex-ponential.

Moreover, differences between whiteand gray matter were also evident on themaps of the slow diffusing component.Figure 3 shows typical signal decays inindividual pixels and the respective fitsfor normal white and gray matter, edem-atous white matter, tumor tissue, andcystic fluid, as determined from a studyin a patient with tumor. Only by grosslyexceeding the clinically used b factorrange of 0–1,000 sec/mm2 did the biex-ponential nature of the signal decay forwhite and gray matter, edema, and tu-mor become apparent. If the decay weremonoexponential, the signal decay on alogarithmic plot would follow a straightline. Noise cannot explain the multiex-ponential decay, since, except for that ofcystic fluid, all MR signal amplitudeswere well above the noise threshold up toa b factor of 5,000 sec/mm2.

Attempts to fit brain signal decaycurves, except for those of the cerebrospi-nal fluid and cystic parts of the tumor

Figure 2. Images of the parameters ADC1, ADC2, A1, and A2 gener-ated from biexponential signal analysis. Data were obtained fromtransverse MR images (2,380/107; 64 3 48 matrix; field of view, 220 3165 mm; section thickness, 7.3 mm; b factor, 5–5,000 sec/mm2;tensor configuration [16]) in a healthy subject. The annotation of thegray scale for the A1 and A2 maps is in arbitrary units.

Figure 3. Logarithmic plot of MR signal intensity versus b factor forindividual pixels in areas of normal white matter (WM) and graymatter (GM), edema, tumor, and cystic fluid. Points below the signalthreshold were not included in the fit calculations (see Materials andMethods section). Biexponential fits of the data points are shown fortissues. For the cystic fluid, data points for a b factor of more than1,000 sec/mm2 are below the signal threshold, and the best fit ismonoexponential. The elevated amplitude of the slow-diffusing com-ponent in tumor tissue is well discernible.

844 z Radiology z June 2001 Maier et al

lesions, with monoexponential functionsled to x2 values that were more than anorder of magnitude larger than those ob-tained from biexponential fits. Triexpo-nential fitting functions yielded unstableparameters, with no significant decreasein x2 values compared with those ob-tained from biexponential fits.

Mean ROI values and SDs of the fast-diffusing component (ADC1), the slow-diffusing component (ADC2), and therelative fraction of the slow-diffusingcomponent are summarized in the Table.For each cystic lesion, only one ADCvalue was calculated from the slope ofthe monoexponential line fit to the log-arithm of the MR signal intensity. Themeasured ADC for cystic lesions was, onaverage, comparable with the ADC of wa-ter at 37°C (3.0 mm2/msec [17]). The biex-ponential signal intensity ratio betweendifferent tissues and normal white matterare listed in the Table. Moreover, to pro-vide a quantitative expression of the ap-pearance of the pathologic areas on thediffusion-weighted images, correspond-ing ratios of the MR signal intensitiesat low, high, and very high diffusionweighting are also given in the Table. Thedata in one of the patients with a metas-tasis were not suitable for analysis be-cause of severe motion artifacts. In one ofthe patients with glioblastoma, peritu-moral edema was not depicted in the se-lected section.

The scatterplot in Figure 4 presents therelative A2 fraction versus ADC1 andADC2 data of individual ROI measure-ments. Evidently, both the A2 fraction

and the ADC1 value alone permit a clearseparation between normal white matterand edematous white matter. None ofthese parameters by itself is sufficient to

allow separation of tumor from brain tis-sues. Together, the A2 fraction, ADC1,and ADC2 permit the separation betweentumor and other brain tissues in about

Mean Biexponential Diffusion Parameters, Biexponential Signal AmplitudeRatios, and Signal Intensity Ratios

A: Biexponential Diffusion Parameters

Tissue ADC1 (mm2/msec) ADC2 (mm2/msec) A2/(A1 1 A2)*

White matter (n 5 15) 1.25 (0.08) 0.19 (0.02) 0.31 (0.04)Gray matter (n 5 15) 1.71 (0.12) 0.37 (0.04) 0.42 (0.05)Edema (n 5 13) 1.80 (0.14)† 0.20 (0.05)† 0.10 (0.02)‡

Tumor (n 5 14) 1.75 (0.20) 0.23 (0.07) 0.24 (0.10)Cyst (n 5 5) 2.75 (0.19) Not applicable Not applicableStroke (n 5 1) 0.84 0.14 0.34

B: Biexponential Signal Amplitude Ratios between Tissue and Normal White Matter

Tissue A1T/A1W§ A2T/A2W

§

Gray matter (n 5 15) 0.96 (0.10) 1.55 (0.21)Edema (n 5 13) 2.78 (0.26)\ 0.73 (0.25)‡

Tumor (n 5 14) 2.18 (0.57) 1.53 (0.68)Cyst (n 5 4) 4.45 (0.59) Not applicable

C: Signal Intensity Ratios between Tissue and Normal White Matter

Tissue b 5 5 sec/mm2b 5 1,004sec/mm2

b 5 5,000sec/mm2

Gray matter (n 5 15) 1.12 (0.06) 0.98 (0.08) 0.71 (0.23)Edema (n 5 13) 1.95 (0.56)† 1.04 (0.17)# 0.64 (0.20)#

Tumor (n 5 14) 1.87 (0.37) 1.29 (0.37) 1.29 (0.95)Cyst (n 5 4) 2.79 (0.63) 0.43 (0.14) Not applicable

Note.—Data in parentheses are the SDs.* A is signal amplitude.† Difference between this value and that of tumor was not significant; P . .05.‡ Difference between this value and that of tumor was significant; P , .001.§ A is signal amplitude, T is tissue, and W is white matter.\ Difference between this value and that of tumor was significant; P , .01.# Difference between this value and that of tumor was significant; P , .05.

Figure 4. Scatterplots of the relative biexponential signal amplitude fraction A2/(A1 1 A2) versus (a) ADC1 and (b) ADC2 ROI values (in squaremicrometers per millisecond). Numbers in parentheses indicate the number of the figure in which images of the particular case are presented.Differentiation of astrocytoma (A) and metastatic tumors (M) on the basis of these two diffusion parameters seems difficult, although more samplesare needed to fully address this issue. GM 5 gray matter, WM 5 white matter.

Volume 219 z Number 3 Multicomponent Apparent Diffusion Coefficient Line Scan Brain Imaging z 845

two-thirds of patients. However, on threeof the 14 images in patients with tumor,tumor values were not different fromedema or gray matter values.

Despite limited resolution and a rela-tively poor signal-to-noise ratio, the cal-culated maps showed a genuine separa-tion of tumor in most cases; that is, thetumor area was clearly separate from theperitumoral edema and normal whitematter. The cystic components of tumorswere easily recognized; however, no at-tempt was made to distinguish necroticand solid parts or other minute details oftumor types.

Figure 5 shows images of a metastasis

of a breast adenocarcinoma in the leftoccipital lobe. On the T2-weighted im-age, the tumor appears to have a signalintensity that is slightly higher than thatof normal white matter. The lesion is sur-rounded by edema that extends to thethalamus and the dorsal limb of the ex-ternal capsule. The A1 map, which repre-sents the biexponential signal amplitudeof the fast-diffusing component of tissuewater, clearly delineates the edema with-out tumor. The tumor is seen separatelyon the A2 map that represents the biex-ponential signal amplitude of the slow-diffusing component.

The images in Figure 6 are from a patient

with an anaplastic astrocytoma in the sple-nium of the corpus callosum. The exami-nation was performed as a control exami-nation in an ongoing radiation therapyprogram. The diffuse high signal intensityin the periventricular white matter on theT2-weighted image may, therefore, be aside effect of this therapy. A congruent dif-fuse high-signal-intensity pattern is seenon the A1 map. Moreover, similar to thecase in Figure 5, the tumor lesion, asdefined by the contrast-enhanced T1-weighted image, is not part of the high-signal-intensity area seen on the A1 map.The A2 map, however, provides good de-piction of the tumor, with high signal in-

Figure 5. Transverse MR images of a breast adenocarcinoma metas-tasis in the left occipital lobe. The extent of edema is seen on theT2-weighted (T2W) spin-echo image (3,000/80; field of view, 240 3240 mm; section thickness, 3 mm). The tumor lesion (arrowhead) isenhanced on the T1-weighted (T1W) contrast-enhanced spin-echoimage (700/14; field of view, 240 3 240 mm; section thickness, 3mm). On the T2-weighted image, the same area appears slightlybrighter (38% higher signal intensity) than the normal white matteron the opposite side. The A1 map that was calculated from diffusion-weighted images with a wide b factor range obtained after the admin-istration of contrast material (2,040/94; 64 3 48 matrix; field of view,220 3 165 mm; section thickness, 7.3 mm; b factor, 5–5,000 sec/mm2)also shows the edema. The A2 map depicts the tumor lesion, with asize and location comparable to those on the contrast-enhancedT1-weighted image.

Figure 6. Transverse MR images of an anaplastic astrocytoma local-ized in the splenium of the corpus callosum. The extent of edema isseen on the T2-weighted (T2W) spin-echo image (3,000/98; field ofview, 220 3 220 mm; section thickness, 5 mm) and on the A1 mapthat was calculated from images obtained with a wide b factor rangeprior to the administration of contrast material (2,040/94; 64 3 48matrix; field of view, 220 3 165 mm; section thickness, 7.3 mm; bfactor, 5–5,000 sec/mm2). The border (open arrowhead) of the tumorlesion is enhanced on the T1-weighted (T1W) contrast-enhancedspin-echo image (600/25; field of view, 220 3 220 mm; sectionthickness, 5 mm). The solid part of the lesion (solid arrowhead) showsno enhancement on the T1-weighted image but appears bright on theA2 map.

846 z Radiology z June 2001 Maier et al

tensity and excellent contrast between tu-mor and surrounding white matter.

Figure 7 shows a case of postoperativeglioblastoma. The lesion in the right fron-tal brain includes a cyst and a contrast-enhanced solid part dorsal to the cyst. Sub-stantial peritumoral edema is present,reaching as far as the central sulcus. Notethat unlike the cases in Figures 5 and 6,tumor tissue is also enhanced on the T2-weighted image and the A1 map. The im-ages of another case of astrocytoma shownin Figure 8 demonstrate that normal diffu-sion weighting will not cause enhance-ment in solid tumor structures. However,the elevated biexponential signal ampli-

tude of the slow-diffusing component pro-duces selective signal enhancement in ar-eas of tumor tissue on the image with veryhigh diffusion weighting.

Discussion

It has been demonstrated that diffu-sion-weighted imaging with a b factorrange wider than that typically used canbe performed in healthy subjects and pa-tients. The line scan diffusion imagingtechnique appears to be suitable for theacquisition of images free of motion arti-

facts, even with very high b factors of upto 5,000 sec/mm2. The fact that imagingwith very high diffusion weighting failedin only one of 15 patients is evidence of aremarkable degree of robustness. Single-shot diffusion-weighted echo-planar im-aging (18), which was not available withour MR system at the time of this study,would have been equally insensitive tomotion. Moreover, with echo-planar im-aging, data from more than one sectioncould have been obtained without in-creasing the imaging time.

On the other hand, in areas near large

Figure 7. Transverse MR images of a frontal glioblastoma with apostoperative cyst. The edema (solid arrowheads) and cyst (C) areenhanced on the T2-weighted (T2W) spin-echo image (3,600/98; fieldof view, 220 3 220 mm; section thickness, 5 mm) and the A1 map.The T1-weighted (T1W) contrast-enhanced spin-echo image (600/25;field of view, 220 3 220 mm; section thickness, 5 mm) predominantlyshows the margins (open arrowheads) and not the solid part (solidarrowhead) of the tumor. However, the A2 map that was calculatedfrom images obtained with a wide b factor range (2,040/94; 64 3 48matrix, field of view, 220 3 165 mm; section thickness, 7.3 mm; bfactor, 5–5,000 sec/mm2) shows the solid part of the tumor. The areaof edema appears remarkably dark on the A2 map.

Figure 8. Transverse MR images of an astrocytoma with edema. Theline scan images (2,040/94; 64 3 48 matrix; field of view, 220 3 165mm; and section thickness, 7.3 mm) were obtained prior to theadministration of contrast material. Edema and tumor are enhancedon the T2-weighted (T2W) line scan spin-echo image (b factor, 5sec/mm2). The ventricular system appears to be compressed on oneside and enlarged on the other. The tumor lesion margins (openarrowheads on upper right image) are depicted on the T1-weighted(T1W) contrast-enhanced spin-echo image (500/25; field of view,220 3 220 mm; section thickness, 5 mm). On the diffusion-weighted(DW) image obtained with a conventional b factor of 1,004 sec/mm2

(high diffusion weighting), no signal intensity enhancement is seenin the areas of edema and tumor, whereas on the image obtained witha b factor of 5,000 sec/mm2 (very high diffusion weighting), thetumor (solid arrowheads) is enhanced. Also enhanced are white mat-ter tracts (open arrowheads on lower right image) that are not alignedwith the diffusion encoding direction (compare with Fig 1).

Volume 219 z Number 3 Multicomponent Apparent Diffusion Coefficient Line Scan Brain Imaging z 847

bone structures, the sensitivity of single-shot echo-planar imaging to susceptibil-ity variations and chemical shifting can,with inadequate shimming, result inghosting artifacts, image distortions, orcomplete signal loss, whereas images ob-tained with line scan diffusion imagingdo not show such artifacts (14,15). Inaddition, eddy currents that are causedby the application of the diffusion en-coding gradients can generate distortionartifacts (19). One would expect thatthese distortion artifacts increase withhigher b factors. With line scan diffusionimaging, however, no distortions wereobserved, even with the highest b factors.

These findings, in agreement withthose of earlier reports (11,12), revealedthat biexponential fits better describe dif-fusion-related signal loss in the brain.The deviation from a purely monoexpo-nential model that becomes evident asthe diffusion encoding is extended be-yond the normal range should not beconfounded with effects that arise frommicrocirculation (20,21). Signal loss dueto microcirculation is observed only withb factors of less than 300 sec/mm2 andtherefore cannot account for the devia-tions measured in various tissues. A re-analysis of the data without use of thelowest b factor (ie, only 15 b factors be-tween 338 and 5,000 sec/mm2) revealedthe same biexponential behavior. Thenormal human white matter parametersobserved in the current study are in goodagreement with those of an earlier study(12) in which a considerably higher num-ber of b factors over a slightly larger rangewere used. Limitations in the interpreta-tion of the slow ADC component as in-tracellular water and the fast ADC com-ponent as extracellular water, including adiscrepancy between the ADC compo-nent volume fractions and reported val-ues of intracellular and extracellular wa-ter volume fractions in brain, werediscussed by Niendorf et al (11).

The novelty of this study, besides theacquisition of these fit parameters in im-age formats, is the observation of biexpo-nential diffusion in pathologic brain tis-sues. Moreover, the four parameters thatdescribe the biexponential fit seem topermit the distinction among the varioustissues. In normal cortical gray mattercompared with normal white matter, allparameters except the biexponential sig-nal amplitude A2 were strongly elevated.The low spatial resolution of the col-lected data may have resulted in a con-tamination of the measured signal withcerebrospinal fluid signal. This signalcontamination would primarily affect

the measurement of the fast-diffusingcomponent, since the monoexponentialdiffusion constant of cerebrospinal fluidis relatively high. In subacutely infarctedtissue, a lower ADC for the fast-diffusingcomponent (Table) is expected, since it isknown from conventional diffusion im-aging that diffusion is reduced (22). Inthe present study, the ADC of the slow-diffusing component was also reduced.This observation, however, is not wellcorroborated, since only one patient withstroke participated in the study.

Characteristic changes in biexponen-tial diffusion parameters, which remainto be explained, were observed in peritu-moral edema and tumor tissue; com-pared with the value in white matter, theADC1 value increased by almost 50%.The A1 value increased by 178% in edemaand 118% in tumor tissue. While therewere only minimal changes in the ADCvalue of the slow-diffusing component,the biexponential signal amplitude A2

decreased in peritumoral edematouswhite matter, whereas it increased in tu-mor tissue. We do not have an explana-tion for the contrary behavior of thebiexponential signal amplitude of theslow-diffusing component in the two tis-sues, since the exact nature of the bi-exponential signal attenuation is notknown. The amplitude of each diffusionconstant is influenced by the spin den-sity and relaxation time of each observedcomponent. Since spin density and therelaxation times of the components arenot necessarily the same, imaging proto-cols with different settings of the repeti-tion and/or echo time may yield differentamplitude values.

In another study (23), however, wefound no statistically significant differ-ences in T1 relaxation. It can be specu-lated that the slow-diffusing componentis determined by the concentration ofwater-binding macromolecules, cellularsize, and tissue architecture (tortuosity)(24–28), that is, factors that are indeeddifferent among the investigated tissues.Thus ADC1, ADC2, and the biexponentialsignal amplitudes A1 and A2 permit theseparation of the various tissues.

Changes in tumor tissue appear to bemore variable, which most likely reflectsthe fact that tumors do not represent asingle type of tissue. From the limitednumber of different tumor types studiedwith this technique to date, correlationsbetween histologic type and biexponentialdiffusion parameters cannot be estab-lished. Cystic regions are easily distin-guished by their monoexponential diffu-sion and high ADC.

Another limitation of the currentstudy is the monodirectional diffusionencoding used for patient imaging. Aspointed out in the Data Analysis section,to reduce the potential error in whitematter tracts where restricted diffusion ismost likely to be present, the mean valueof two independent and large white mat-ter ROIs was used. For each case (eg, Fig6), we also verified that the observed tu-mor enhancement was not due to re-stricted diffusion in a white matter tract.

Diffusion-based tissue differentiationdoes not depend on the use of contrastagents, since the diffusion-weighted im-ages are only minimally T1-weighted (ef-fective repetition time, 2,040 msec) (29).The tumor-tissue contrast generated onmulticomponent ADC maps is differentfrom the enhancement produced byparamagnetic contrast agents. The con-trast agents in use are not specific to tu-mor tissue, but rather, they enhance ar-eas where the blood-brain barrier hasbecome permeable due to neoplasticgrowth, surgical procedures, or radiationtherapy (2,3). Moreover, not all tumortypes enhance with the use of contrastagents. On multicomponent ADC maps,on the other hand, as demonstrated inFigures 6 and 7, enhancement occurs inthe solid part of the tumor.

Diffusion-weighted imaging has beenpreviously evaluated (5–7) in the charac-terization of brain tumors and associatedpathologic structures. With this study,we demonstrated that only limited infor-mation is gained from diffusion imagingin the normal b factor range (Fig 3). Withvery high b factors, however, the signalin tumor tissue is elevated in comparisonwith that of surrounding edema and nor-mal white matter; the elevated signal isdue to the higher biexponential signalamplitude of the slow-diffusing compo-nent in tumor tissue. The data in theTable and the diffusion-weighted imagesin Figure 8 reiterate this observation:With low diffusion weighting, peritu-moral edema and tumor tissue are en-hanced; at high diffusion weighting(around 1,000 sec/mm2), the contrastamong the different tissues is minimal;and at very high diffusion weighting, tu-mor tissue is selectively enhanced.

From the equation in the Materials andMethods section, it follows that for bequal to 0, the measured signal equalsA1 plus A2. At very high diffusionweighting, the signal is dominated bythe second diffusion component, that is,S ' A2 exp(2ADC2b). Consequently, forwhite matter, edema, and tumor, whereADC2 is similar (Fig 4b, Table), images

848 z Radiology z June 2001 Maier et al

obtained with very high diffusion imag-ing may be considered A2-weighted.Thus, for routine diagnostic imaging, itmay be sufficient to acquire only one dif-fusion-weighted image with very highdiffusion weighting. Signal averaging byusing different diffusion encoding direc-tions could then be performed to over-come a poor signal-to-noise ratio andlimited spatial resolution. Trace diffusionweighting would be necessary to elimi-nate selective enhancement of whitematter tracts (Fig 8), which could be mis-interpreted as lesions.

In summary, we investigated a methodfor tissue characterization with multiex-ponential analysis of the diffusion-atten-uated signal. We examined brain tumors,and our results indicated that the calcu-lated images of slow- and fast-diffusingcomponents helped distinguish edemaand tumor tissue in a number of cases.Provided that more experience is gainedwith different tumor types, the tech-nique could potentially be used withoutparamagnetic contrast agents for tumorlocalization. Clearly, further experimentsare needed to reveal the biophysical basisof the described phenomenon and to op-timize the method for routine clinical di-agnosis.

References1. Damadian R. Tumor detection by nuclear

magnetic resonance. Science 1971; 171:1151–1153.

2. Hasso AN, Kortman KE, Bradley WG. Su-pratentorial neoplasms. In: Stark DD,Bradley WG Jr, eds. Magnetic resonanceimaging. 2nd ed. Vol 1. St Louis, Mo:Mosby–Year Book, 1992; 770–817.

3. Hicks RJ. Supratentorial brain tumors. In:Edelman RR, Hesselink JR, Zlatkin MB,eds. Clinical magnetic resonance imag-ing. 2nd ed. Vol 1. Philadelphia, Pa: Saun-ders, 1996; 533–556.

4. Eis M, Els T, Hoehn-Berlage M, HossmanKA. Quantitative diffusion MR imaging ofcerebral tumor and edema. Acta Neuro-chir Suppl (Wien) 1994; 60:344–346.

5. Krabbe K, Gideon P, Wagn P, Hansen U,Thomsen C, Madsen F. MR diffusion im-aging of human intracranial tumours.Neuroradiology 1997; 39:483–489.

6. Le Bihan D, Douek P, Argyropoulou M,Turner R, Patronas N, Fulham M. Diffu-sion and perfusion magnetic resonanceimaging in brain tumors. Top Magn Re-son Imaging 1993; 5:25–31.

7. Tien RD, Felsberg GJ, Friedman H, BrownM, Macfall J. MR imaging of high-gradegliomas: value of diffusion-weightedechoplanar pulse sequences. AJR Am JRoentgenol 1994; 162:671–677.

8. Tsuruda JS, Chew WM, Moseley ME, Nor-man D. Diffusion-weighted MR imagingof extraaxial tumors. Magn Reson Med1991; 19:316–320.

9. Yanaka K, Shirai S, Kimura H, Kamezaki T,Matsumura A, Nose T. Clinical applica-tion of diffusion-weighted magnetic reso-nance imaging to intracranial disorders.Neurol Med Chir 1995; 35:648–654.

10. Brunberg JA, Chenevert TL, McKeever PE,et al. In vivo MR determination of waterdiffusion coefficients and diffusion an-isotropy: correlation with structural alter-ation in gliomas of the cerebral hemi-spheres. Am J Neuroradiol 1995; 16:361–371.

11. Niendorf T, Dijkhuizen RM, Norris DG,van Lookeren Campagne M, Nicolay K.Biexponential diffusion attenuation invarious states of brain tissue: implicationsfor diffusion-weighted imaging. MagnReson Med 1996; 36:847–857.

12. Mulkern RV, Gudbjartsson H, Westin CF,et al. Multi-component apparent diffu-sion coefficients in human brain. NMRBiomed 1999; 12:51–62.

13. Gudbjartsson H, Maier SE, Mulkern RV,Morocz IA, Patz S, Jolesz FA. Line scandiffusion imaging. Magn Reson Med1996; 36:509–519.

14. Maier SE, Gudbjartsson H, Patz S, et al.Line scan diffusion imaging: characteriza-tion in healthy subjects and stroke pa-tients. AJR Am J Roentgenol 1998; 17:85–93.

15. Robertson RL, Maier SE, Robson CD,Mulkern RV, Karas PM, Barnes PD. MRline scan diffusion imaging of the brainin children. AJNR Am J Neuroradiol 1999;20:419–425.

16. Basser PJ, Pierpaoli C. A simplifiedmethod to measure the diffusion tensorfrom seven MR images. Magn Reson Med1998; 39:928–934.

17. Le Bihan D. Molecular diffusion nuclearmagnetic resonance imaging. Magn Re-son Q 1991; 7:1–30.

18. Turner R, Le Bihan D, Maier J, Vavrek R,Hedges LK, Pekar J. Echo-planar imagingof intravoxel incoherent motions. Radiol-ogy 1990; 177:407–414.

19. Haselgrove JC, Moore JR. Correction for

distortion of echo-planar images used tocalculate the apparent diffusion coeffi-cient. Magn Reson Med 1996; 36:966–964.

20. Le Bihan D, Breton E, Lallemand D,Aubin ML, Vignaud J, Laval-Jeantet M.Separation of diffusion and perfusion inintravoxel incoherent motion MR imag-ing. Radiology 1988; 168:497–505.

21. Conturo TE, McKinstry RC, Aronovitz JA,Neil JJ. Diffusion MRI: precision, accuracyand flow effects. NMR Biomed 1995;8:307–332.

22. Warach S, Chien D, Li W, Ronthal M,Edelman RR. Fast magnetic resonance dif-fusion-weighted imaging of acute stroke.Neurology 1992; 42:1717–1723.

23. Mulkern RV, Zengingonul HP, RobertsonRL, et al. Multi-component apparent dif-fusion coefficients in human brain: rela-tionship to spin-lattice relaxation. MagnReson Med 2000; 44:292–300.

24. Hajnal JV, Doran M, Hall AS, et al. MRimaging of anisotropically restricted dif-fusion of water in the nervous system:technical, anatomic, and pathologic con-siderations. J Comput Assist Tomogr1991; 15:1–18.

25. Neeman M, Jarret KA, Sillerud LO, FreyerJP. Self-diffusion of water in multicellularspheroids measured by magnetic reso-nance microimaging. Cancer Res 1991;51:4072–4079.

26. Pilatus U, Shim H, Artemov D, Davis D,van Zijl PCM, Glickson JD. Intracellularvolume and apparent diffusion constantsof perfused cancer cell cultures, as mea-sured by NMR. Magn Reson Med 1997;37:825–832.

27. Sevick RJ, Kanda F, Mintorovitch J, et al.Cytotoxic brain edema: assessment withdiffusion-weighted MR imaging. Radiol-ogy 1992; 185:687–690.

28. van Zijl PCM, Moonen CTW, Faustino P,Pekar J, Kaplan O, Cohen JS. Completeseparation of intracellular and extracellu-lar information in NMR spectra of per-fused cells by diffusion-weighted spec-troscopy. Proc Natl Acad Sci USA 1991;88:3228–3232.

29. Morriss C, Haselgrove J. Intravenous gad-olinium does not alter the apparent dif-fusion coefficient: a study in enhancingbrain tumors (abstr). In: Proceedings ofthe Eighth Meeting of the InternationalSociety for Magnetic Resonance in Medi-cine. Berkeley, Calif: International Soci-ety for Magnetic Resonance in Medicine,2000; 1103.

Volume 219 z Number 3 Multicomponent Apparent Diffusion Coefficient Line Scan Brain Imaging z 849