Gerraty, Ting Li and Stephen M. DavisQing Yang, Brian M. Tress, P. Alan Barber, Patricia M. Desmond, David G. Darby, Richard P.

StrokeSerial Study of Apparent Diffusion Coefficient and Anisotropy in Patients With Acute

Print ISSN: 0039-2499. Online ISSN: 1524-4628 Copyright © 1999 American Heart Association, Inc. All rights reserved.

is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231Stroke doi: 10.1161/01.STR.30.11.2382

1999;30:2382-2390Stroke. 

http://stroke.ahajournals.org/content/30/11/2382World Wide Web at:

The online version of this article, along with updated information and services, is located on the

  http://stroke.ahajournals.org//subscriptions/

is online at: Stroke Information about subscribing to Subscriptions: 

http://www.lww.com/reprints Information about reprints can be found online at: Reprints:

  document. Permissions and Rights Question and Answer process is available in the

Request Permissions in the middle column of the Web page under Services. Further information about thisOnce the online version of the published article for which permission is being requested is located, click

can be obtained via RightsLink, a service of the Copyright Clearance Center, not the Editorial Office.Strokein Requests for permissions to reproduce figures, tables, or portions of articles originally publishedPermissions:

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

Serial Study of Apparent Diffusion Coefficient andAnisotropy in Patients With Acute Stroke

Qing Yang, PhD; Brian M. Tress, MD, FRACR; P. Alan Barber, FRACP;Patricia M. Desmond, MSc, FRACR; David G. Darby, PhD, FRACP;

Richard P. Gerraty, MD, FRACP; Ting Li, PhD; Stephen M. Davis, MD, FRACP

Background and Purpose—We sought to characterize the evolution of apparent diffusion coefficient (ADC) and apparentdiffusion anisotropy (ADA) in acute stroke and to evaluate their roles in predicting stroke evolution and outcome.

Methods—We studied 26 stroke patients acutely (,24 hours), subacutely (3 to 5 days), and at outcome (3 months). Ratiosof the ADC and ADA within a region of infarction and the normal contralateral region were evaluated and comparedwith the Canadian Neurological Scale, Barthel Index, and Rankin Scale.

Results—Heterogeneity in ADC and ADA evolution was observed not only between patients but also within individuallesions. Three patterns of ADA evolution were observed: (1) elevated ADA acutely and subacutely; (2) elevated ADAacutely and reduced ADA subacutely; and (3) reduced ADA acutely and subacutely. At outcome, reduced ADA withelevated ADC was observed generally. We identified 3 phases of diffusion abnormalities: (1) reduced ADC and elevatedADA; (2) reduced ADC and reduced ADA; and (3) elevated ADC and reduced ADA. The ADA ratios within 12 hourscorrelated with the acute Canadian Neurological Scale (r50.46, P50.06), subacute Canadian Neurological Scale(r50.55, P50.02), outcome Barthel Index (r50.62, P50.01), and Rankin Scale (r520.77, P,0.0005) scores.

Conclusions—Combined ADC and ADA provide differential patterns of stroke evolution. Early ADA changes reflectcellular alterations in acute ischemia and may provide a potential marker to predict stroke outcome.(Stroke.1999;30:2382-2390.)

Key Words: cerebral edeman cerebral infarctionn magnetic resonance imaging, diffusion-weightedn stroke, ischemicn stroke outcome

Diffusion-weighted imaging (DWI) has become an im-portant MRI method for the early diagnosis and char-

acterization of ischemic stroke.1–10 Early reports1,2 showedthat acute infarcts appear hyperintense on DWI because of areduction in the apparent diffusion coefficient (ADC) ofwater within minutes after the onset of ischemia,3 reflectingearly disruption of energy metabolism.

However, the mechanism behind the ADC reduction inacute ischemia is not fully understood. In biological tissue,the ADC of water molecules is much lower than its free watervalue because of physical restrictions from membranes,fibers, and macromolecules such as proteins.11 The amount ofdissolved organic molecules may also alter the viscosity ofwater. In acute ischemia, disruption of energy metabolismwith failure of ion pumps causes cell swelling and accumu-lation of intracellular sodium and water (cytotoxic edema).7,8

The net migration of water from extracellular space (ECS,where ADC is presumed to be high) into intracellular space(ICS, where ADC is presumed to be low) has been considered

the dominant mechanism for the ADC reduction.1,6 Anotherpostulated mechanism is that the ADC reduction is a result ofdecreased membrane permeability caused by collapse oftransmembrane ion gradients.12 Furthermore, recent studieshave provided compelling evidence that the observed ADCreduction is dominated by the water ADC reduction in theICS because of a decrease in energy-dependent cytoplasmiccirculation or an increase in water viscosity.13–15On the otherhand, the time course correlation of the ADC reduction withthe decrease in ECS volume and the increase in ECStortuosity suggests that cell swelling may also cause ADCreduction of water or other molecules in the ECS.15,16Recentnumerical modeling has suggested both cellular swelling andmembrane permeability to be important factors influencingADC in spinal cord white matter.17

The temporal evolution of diffusion abnormalities is alsoimportant in tracking stroke progression. Studies using ani-mal models have revealed profound ADC changes at earlyacute stages.2–4 Moreover, there is an increase in the lesion

Received July 1, 1999; final revision received August 18, 1999; accepted August 18, 1999.From the Departments of Radiology (Q.Y., B.M.T., P.M.D, T.L.) and Neurology (P.A.B., D.G.D., R.P.G., S.M.D), Royal Melbourne Hospital and

University of Melbourne, Victoria, Australia.Presented in part at the Annual Meeting of the American Society of Neuroradiology, San Diego, Calif, May 23–28, 1999.Correspondence to Dr Qing Yang, Department of Radiology, University of Melbourne, Royal Melbourne Hospital, Parkville, Victoria 3050, Australia.

E-mail QY@radior.medrmh.unimelb.edu.au© 1999 American Heart Association, Inc.

Stroke is available at http://www.strokeaha.org

2382 by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

volume over time as measured by decreased ADC values.3 Inhuman stroke, while some studies6,7 report the persistence ofa reduced ADC for at least 4 days after stroke onset, othershave found heterogeneity of ADC within the infarct whilepart or all of the lesion displayed pseudonormal or high ADCvalues by 24 to 48 hours.5,18 This disparity may be related todifferent patient populations or stroke etiologies,5,18 differentdata acquisition and analysis techniques,19 and anisotropicdiffusion effects.20,21 The diffusion of water molecules inbiological tissue is anisotropic, particularly in tissue contain-ing nerve fibers and white matter tracts, where ADC is highin the direction parallel to fiber tracts and low in perpendic-ular directions.22,23 Anisotropic diffusion can cause conspic-uous hyperintensities on DWI that may be confused withacute ischemia9 and influence the accurate quantitation ofADC.20,21 Such effects can be minimized by calculating thetrace ADC and isotropic DWI images from DWI measure-ments in 3 orthogonal directions.20,24Recent studies25–27haveevaluated the time course of the trace ADC in human stroke,eliminating the influences of anisotropic diffusion.Computer-assisted segmentation analysis has further demon-strated ADC heterogeneity within the infarct and partial ADCelevation by 5 to 9 hours after ictus, although the averageADC in the lesion remained low.27

Measurements of apparent diffusion anisotropy (ADA)may provide additional information about cellular changes inischemic stroke.22,23,28–33To our knowledge, no systematicstudy of the evolution of combined ADC and ADA in acutehuman stroke has been reported. In this study we evaluatedthe serial changes of ADC and ADA abnormalities in patientswith acute ischemic stroke. We sought to improve ourunderstanding of the mechanisms underlying the ADC andADA changes in ischemia and to determine whether thesedata are useful in characterizing and predicting the evolutionof acute ischemic stroke.

Subjects and MethodsPatients with sudden onset of focal neurological deficit consistentwith hemispheric ischemic stroke were recruited from the StrokeUnit of the Royal Melbourne Hospital. Stroke onset was defined asthe last time the patient was known to be without neurologicaldeficit. Patients were excluded if they had cerebral hemorrhage,preexisting significant nonischemic neurological deficit (includingdementia or extrapyramidal disease), or a history of prior stroke thatwould hinder interpretation of clinical and radiological data. Patientstreated by putative neuroprotective or thrombolytic drugs wereexcluded. The study was performed with the approval of the ethicscommittee at our institution, and written informed consent wasobtained from the patient or next of kin.

Clinical AssessmentThe Canadian Neurological Scale (CNS), a validated neurologicalimpairment score,34 was measured just before the acute and subacuteMRI studies. Outcome clinical assessments were performed on thesame day as the outcome MRI study and consisted of a repeated CNSand scores derived from the Barthel Index (BI) and the Rankin Scale(RS).35 The BI is a validated functional disability score, and the RSis a validated handicap scale. These clinical scales were used becausethey measure different aspects of recovery after stroke. All clinicalassessments were performed by a neurologist or neurology residenttrained in their administration and were administered without knowl-edge of the MRI results.

Imaging ParametersAll patients were scanned on a 1.5-T clinical whole-body scanner(Signa Horizon SR120, GE Medical Systems) with the use of anoptimized protocol including a T1-weighted sagittal localizer, DWIsequence, contrast-enhanced perfusion imaging (CEPI) sequence,dual proton-density and T2-weighted fast-spin-echo sequence, spin-echo echo planar imaging (EPI) sequence, phase-contrast MRangiography (MRA), and finally a contrast-enhanced T1-weightedsequence. Similar slice locations were used to facilitate comparisons.The total “table time” for these sequences was approximately 20minutes. Only the DWI results are reported in this study.

DWI scans were performed with a single-shot, spin-echo EPIsequence with the Stejskal-Tanner diffusion-encoding method.36 TheDWI parameters were as follows: 40320-cm field of view, 2563128matrix size, 16 axial slices, 6-mm slice thickness, and 1-mm gapcovering the whole brain. The first 19 patients were studied with atrace DWI sequence with 5 diffusion b values (0 to 1000 s/mm2) ineach of 3 orthogonal directions24 and repetition time/echo time(TR/TE) of 6000/110 ms. The remaining 7 patients had diffusiontensor imaging (DTI)22,23with 3 b values (0 to 1000 s/mm2) in eachof 6 directions and TR/TE of 10 000/110 ms. Scanning time was 1minute 18 seconds for the trace DWI and 2 minutes 10 seconds forthe DTI.

Image ProcessingPostprocessing of images was performed on a UNIX workstationwith the use of customized software developed in IDL (InteractiveData Language, Research Systems Inc). Since the calculation ofdiffusion anisotropy is sensitive to image noise and artifacts, noisereduction and correction for image distortions were performed on allimages. Raw images were filtered with a 939 gaussian kernel(s50.5). The average background noise was subtracted to reduce thenonlinear influences on the DWI signal attenuation. The EPI se-quence has first-order eddy current compensation to minimize imagedistortions. However, higher-order eddy current distortions due tothe strong diffusion gradients may still cause artifacts, particularly onADA quantification. Therefore, corrections for such distortions wereapplied with a modified approach based on the translation-shear-scaling model.37 Although the single-shot EPI sequence eliminatedmotion artifacts from each “snapshot” image, possible head move-ment during the DWI scanning time (1 to 2 minutes) would becaptured, causing mismatch between images. Such mismatch mayvary between images and slices and cannot be easily corrected withthe use of standard rigid-body coregistration algorithms. Therefore,any data set with interimage mismatch was aligned by a dynamicvisual-manual adjustment based on both ratio and difference maps,followed by an automated fine-tuning approach. All image correc-tions employed a cubic convolution interpolation method, whichclosely approximates the theoretically optimum sinc interpolationfunction38 to preserve image resolution. Motion and distortionartifacts were double-checked in animation mode between images ofdifferent b values in each direction and between images of the sameb value at different directions. The DWI signal intensity attenuationcurve was also dynamically checked, particularly in the region ofinterest (ROI). Any individual image with noticeable artifacts wasexcluded from the fitting process for calculation of the ADC map ineach of the 3 (for trace DWI) or 6 directions (for DTI).

Both the trace DWI and the DTI allow the calculation of ADCs ineach of 3 orthogonal directions, which then provide the averageADC (ADCav, noted as ADC hereafter) and the orientation-dependentstandard deviation index of ADA (ADAsd).24 In addition, DTI allowsthe calculation of the full diffusion tensor22 and hence a moreaccurate orientation-independent anisotropy index, such as the frac-tional anisotropy (ADAfr).23 The trace DWI and DTI sequences weretested on a standard water phantom at room temperature. Thecalculated ADC value of free water was 2.260.131023 mm2/s, withan ADAsd value of 0.0460.02 and ADAfr value of 0.1660.05. Thesenon-zero ADA values are mainly due to signal-to-noise ratio.22 BothADA indices range from 0 to 1, representing completely isotropic toextreme anisotropic diffusion.

Yang et al Serial Study of Diffusion Abnormalities in Human Stroke 2383

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

Data AnalysisHeterogeneity of ADC within the lesion has been reported5 andfurther confirmed with computer-assisted segmentation analysis.27

Our experience confirmed this finding. The heterogeneous ADC andADA distribution within the lesion can also be visualized byadjusting the image contrast level and window, particularly withinlarge infarctions. Therefore, segmentation analysis of the lesion isnecessary to properly study the evolution of ADC and ADA indifferent tissue areas. Automated segmentation analysis requiresspecialized computer software,5,27 which is not widely available.Furthermore, it would be difficult to automatically follow the sametissue area over serial studies because of the heterogeneous progres-sion and edematous swelling of the lesion. Therefore, we used asegmentation approach based on anatomic location and tissue typesimilar to that used in animal model studies.2,3 Since different tissuestructures have different ADA values, proper evaluation of ADAchanges needs to be linked to specific tissue types. The ADA of thewhite matter is much greater than that of the gray matter, and theADA map is very useful in delineating white matter. When oneconsiders the image resolution and slice thickness of the DWIimages, partial volume effects (PVE) were inevitable, particularlybetween the gray matter and the white matter or the cerebrospinalfluid (CSF) in cerebral gyral areas. Therefore, lesions were seg-mented and grouped as white matter (WM), cortical or deep graymatter (GM), and mixed gyral gray/white matter (GWM) regions.The isotropic DWI (with b51000), T2-weighted image (which is DWIwith b50), and ADC and ADA maps of serial studies were visualizedsimultaneously. A ROI was initially selected by free-hand tracing on theacute ADC map to outline the infarct with reduced ADC. The ROI wasthen projected on other images and further edited for reliable segmen-tation, guided by anatomic knowledge and histogram analysis. No serialmeasurement was performed for later expanded areas of infarction.Efforts were made to track each segmented tissue area with all 4 kindsof images, guided by anatomic knowledge, while edematous swellingwas also considered. In a few cases in which asymmetric headangulation caused difficulties for comparison with the contralateral sideon the same slice, multiple slices were analyzed. Additionally, it was notpractical to reproduce the exact same slice locations over serial MRIstudies for all patients, especially when head movements occurredduring the scan, although a standard landmark was always used. Thus,some raw DWI data may be subject to PVE within half the slicethickness. However, efforts were made to minimize additional PVEduring the data analysis. No segmentation was performed for smallinfarctions, which were evaluated separately from large lesions. Theaverage ADC and ADA values within each segmented ROI wereobtained with histogram analysis to eliminate contamination fromindividual noise. Care was also taken to avoid contamination fromsulcal CSF. To attain a reliable measure of the evolution of the ADCand ADA changes between patients and over serial studies, ADC andADA ratios were calculated by dividing the mean values in the lesionROI by those in the corresponding contralateral ROI.20

Linear regression analyses were performed between the ADC andADA ratios and the clinical scores (CNS, BI, RS) with Pearson’scorrelation coefficient (r) and significance level (P) of the F test.Results were considered statistically significant at levels ofP,0.05.Since the evolutions of ADC and ADA are heterogeneous andcomplicated, no statistical analysis was performed to test for ratiodifference, but all data points are presented.

ResultsTwenty-six patients (16 men, 10 women; mean age, 69612years; range, 42 to 92 years) were studied, with 9 smalllesions (lacunar, striatocapsular, and small cortical infarc-tions) and 17 large lesions (major arterial territory infarc-tions). All 26 patients had acute studies, and 25 had subacutestudies. One patient missed the subacute study but completedthe outcome study. Twenty patients (including all 9 patientswith small lesions) had outcome studies. Three patientsrefused, and 1 was unable to tolerate the outcome MRI scan,

but all of them had the outcome clinical assessments. Onepatient died as a result of complications from stroke (assignedoutcome BI55 and RS50), and 1 died of an unrelated cardiacevent after the subacute study (outcome BI and RS notavailable). Time from stroke onset to the acute study was11.567.2 hours (range, 2.5 to 23.5 hours), with 10 studieswithin 6 hours and 17 studies within 12 hours. Time to thesubacute study was 3.761.2 days (range, 1.9 to 6.8 days).Time to the outcome study was 90626 days (range, 35 to 154days).

Trace DWI and DTIA total of 71 MRI studies were performed with 51 trace DWIand 20 DTI scans. The mean ADC value in all normal ROIswas 0.8560.14 (31023 mm2/s), which agrees well with otherreports.20,25–27

All 71 DWI scans provided ADAsd maps, while 20 DTIscans provided additional ADAfr maps. Since ADAsd isorientation dependent, it underestimates the diffusion anisot-ropy of water in tissues depending on tissue types and fiberorientations relative to the 3 orthogonal diffusion-encodingdirections.22 However, our experience indicated similarADA sd and ADAfr changes in terms of elevation or reductionby comparing the ischemic lesion with the correspondingcontralateral region. To further clarify this phenomenon, themean values of both indices in the segmented infarcts andnormal ROIs (including CSF) from the 20 DTI scans wereplotted, as shown in Figure 1a. This illustrates the range ofquantitative ADA values of different tissue types measuredby both indices, in good agreement with other reports.22,30,33

The non-zero ADA values in CSF agree well with those

Figure 1. Relationship between ADAfr and ADAsd values (a) andbetween ADAfr and ADAsd* ratios (b), as described in the text,for WM (rectangle), GM (circle), GWM (diamond), and CSF (star).Solid and open symbols represent normal and infarcted tissues,respectively. The solid curve in panel a represents the empiricalrelationship ADAfr512exp(24.2 ADAsd). The solid line in panel bshows a 1:1 relationship supported by linear regression analysis.

2384 Stroke November 1999

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

measured in a water phantom. Figure 1a demonstrates theempirical relationship between ADAsd and ADAfr, as de-scribed by a fitted curve: ADAfr512exp(24.2 ADAsd). A2-dimensional histogram analysis of the associated ADAsd

and ADAfr maps further supported such an approximaterelationship. The ADAsd values were converted into ADAsd*values, which are approximately equivalent to ADAfr accord-ing to the above relationship. The ratios of ADAsd* in thelesion to that in the contralateral region were calculated andcompared with the corresponding ADAfr ratios, as shown inFigure 1b. A 1:1 relationship between the ADAfr and thecorrected ADAsd* ratios was found by linear regressionanalysis (slope51.0060.02,r50.94,P,10212). This allowedus to extend the above correction to all the ADAsd valuesmeasured by the trace DWI. The corrected ADAsd* ratioswere included together with the ADAfr ratios measured by theDTI to increase statistical power in subsequent analysis.

ADC and ADA Evolution in Small InfarctionsThe time courses of the ADC and ADA ratios in 9 patientswith small lesions are shown in Figure 2. All 9 acute lesionsdisplayed reduced ADC; 8 had elevated ADA, while 1 WMlesion at 23.5 hours had low ADA ratio. Subacutely, the ADCratios tended toward normal in 2 small cortical GM lesionsand remained low in the others. The initially elevated ADAratios persisted in 2 WM lesions, became reduced in 4 WMlesions, and approached normal in the 3 small cortical GMlesions. At outcome, the ADC ratio became elevated in 8patients, while 1 had persistent ADC reduction at 85 days.The ADA ratio became further reduced in all lesions. Overall,the ADC and ADC changes during infarct evolution weregreater in WM than in GM lesions.

ADC and ADA Evolution in Large InfarctionsA total of 69 segmented lesion ROIs were sampled in serialstudies of 17 patients. The time courses of the ADC and ADAratios in all the segmented ROIs are shown in Figure 3.General patterns of ADC and ADA evolution are similar to

those in the small lesions of the same tissue types, except thatno acutely elevated ADA was observed in the deep GM. Theevolution of ADC and ADA ratios in the GWM tissues is alsosimilar to that of small WM lesions. However, earlier ADCpseudonormalization or elevation by 24 to 48 hours can beseen in both WM and GM lesions. Figure 4 demonstrates theserial ADC and ADA maps of a patient with initially reducedADC and elevated ADA in the WM-dominated lesion. Theseevolved to elevated ADC and reduced ADA by 42 hours afterstroke onset. Heterogeneous ADC distribution within thelesion can be seen. In addition, the initial lesion expanded by42 hours, with a peripheral rim of reduced ADC and elevatedADA resembling the initial diffusion abnormalities of theinfarct core. At 93 days, both the infarct core and the rimdisplayed different levels of ADC elevation and ADA reduc-tion (Figure 4). Such expanded infarct areas were not in-cluded in the serial analysis. Figure 5 demonstrates differ-ences in both ADC and ADA evolutions within the samelesion, in which the initial infarct had relatively homogeneousADC reduction and ADA elevation but displayed significantheterogeneities between the WM and peripheral GWM areasat 5 days. Although there are some differences between theADA fr and ADAsd maps, similar features of ADA changes inthe lesion can be seen compared with the contralateral side.This patient’s head was tilted to the left at the subacute stage,reflecting the actual clinical situation. However, no severehead movement occurred during the DWI sequence, and anyinterimage mismatch was eliminated.

For most of the small and segmented large lesions, 3 patternsof ADA evolution were observed generally: (1) elevated ADAacutely and subacutely; (2) elevated ADA acutely and reducedADA subacutely; and (3) reduced ADA acutely and subacutely.At outcome, elevated ADC with reduced ADA was observedgenerally. The exceptional cases, including 1 WM lesion (Figure3, solid squares in top row) and 3 small cortical GM lesions(Figure 2, bottom row), had both ADC and ADA ratios slightlydifferent from normal by outcome studies. Figure 6 demon-strates the serial ADC and ADA maps, isotropic DWI, and

Figure 2. Time course of ADC and ADAratios in small lesions after stroke onset.Solid lines join the serial data points foreach patient. Time is on logarithmic scale.

Yang et al Serial Study of Diffusion Abnormalities in Human Stroke 2385

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

T2-weighted image in the WM lesion case. Despite the usualprogression of the peripheral GWM lesion area (white arrow-head in Figure 6, solid diamonds in middle row of Figure 3), theacute WM lesion (white arrow) progressed gradually, approach-ing normal ADC and ADA and slight T2-weighted image

hyperintensity (black arrow) by outcome. This is the only case inthis study in which the control ADC and ADA values weresampled from surrounding nonlesional tissues to avoid an oldlesion on the contralateral side (double black arrows), whichmight have led to less reliable ratios. The absolute ADC valueswere 0.56, 0.73, and 0.87 (31023 mm2/s) in the WM lesion and0.71, 0.78, and 1.21 (31023 mm2/s) in the peripheral GWMlesion area at 4.5 hours, 51 hours, and 85 days, respectively. Theabsolute ADC value was 0.8660.05 (31023 mm2/s) in thenormal tissues of this patient, which confirmed the slow ADCrecovery in the WM lesion area. Examination of the MRA andCEPI results in this patient indicated early reperfusion. However,the slight T2 hyperintensity in this WM area at outcome suggestssome permanent pathological changes.

Relationship Between ADC and ADAThe combined ADC and ADA ratios in all lesions over serialstudies are shown in Figure 7. Three phases of diffusionabnormalities are distinguishable: (1) elevated ADA andreduced ADC; (2) reduced ADA and reduced ADC; and(3) reduced ADA and elevated ADC. Most infarctions dis-played a transition from phases 1 to phase 2 between the acuteand subacute stages, then to phase 3 by outcome. However,phase 3 had occurred by the subacute stage in some WM andGM regions of large lesions, while phase 2 persisted from thesubacute to outcome stages in 1 small WM lesion. It is notedthat the 3 small cortical GM lesions (solid circle in Figure 7)

Figure 3. Time course of ADC and ADAratios in segmented large lesions afterstroke onset. Solid lines join the serial datapoints for each segmented ROI. The solidrectangle and diamond correspond to thesegmented WM and GWM lesions in thepatient who had slow ADC and ADA recov-ery, as shown in Figure 6.

Figure 4. Serial ADC (top row) and ADAsd (bottom row) maps ofa patient who displayed an early ADC elevation of the infarctcore with later development of ischemia in the peripheral rim by42 hours. Note heterogeneity of ADC within the lesion and dif-ferent ADC levels between the core and periphery at outcome.

2386 Stroke November 1999

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

and the slowly evolving WM lesion (Figure 6, solid square inFigure 7) tended to transform from phase 1 via the normalADC and ADA cross point into phase 3. For all small andlarge lesions, there is a correlation between ADC and ADAratios for transition from phase 1 to phase 2 (r520.41,P50.0005) and from phase 2 to phase 3 (r520.52,P,0.00005). However, the overall transition between all 3phases is nonlinear, presumably reflecting different patholog-ical processes during infarct evolution.

Correlation With Clinical ScoreFor large infarctions with multiple segmented ROIs, theaverage ADC and ADA ratios weighted by the segmentedtissue volume were calculated to represent an average diffu-sion abnormality of the entire lesion slice. Correlation anal-yses were performed between the patient’s clinical scores(CNS, BI, RS) and the average ADC and ADA ratios for thetime intervals of,12 hours, 12 to 24 hours, 2 to 10 days(subacute stage), and.35 days (outcome). Correlations werefound between ADA ratios within 12 hours of stroke onsetand the acute CNS (r50.46, P50.06), subacute CNS(r50.55, P50.02), outcome BI (r50.62, P50.01), and RS(r520.77,P,0.0005) scores. Additional analysis confirmedthere was no significant correlation between the ADA ratiosand the entire acute DWI lesion volume or outcome T2 lesion

volume. This suggests that the correlations between the ADAratio and clinical scores are not biased by the lesion size.Furthermore, there was no significant correlation betweenclinical scores and ADA ratios at later stages or ADC ratiosat any stage. This suggests that early ADA changes mayprovide an important marker in predicting stroke outcome.

DiscussionIn this study of patients with acute ischemic stroke, we havefound different patterns of ADC and ADA evolution overtime and different phases of combined ADC and ADAabnormalities. In addition, we have observed differences inboth ADC and ADA evolution not only between patients butalso within individual lesions. Furthermore, we have foundthat ADA ratios within 12 hours of stroke onset correlatesignificantly with acute, subacute, and outcome clinicalscores.

Figure 5. Serial ADC (top row), ADAsd (middle row), and ADAfr

(bottom row) maps of a patient demonstrating heterogeneity inADC and ADA evolution of the ischemic lesion. The initial infarcthad relatively homogeneous ADC reduction and ADA elevationbut displayed significant heterogeneities between the WM andthe peripheral GWM areas at 5 days. The ADAsd and ADAfr

maps display similar features of ADA changes. Note that thetilted head position reflects actual clinical situation; however,motion artifacts have been eliminated (see imaging processingsection).

Figure 6. Serial ADC (top row) and ADAfr (second row) maps,isotropic DWI (third row), and T2-weighted image (bottom row)images of a patient demonstrating slow ADC recovery of theacute WM lesion (arrow on acute DWI), while the peripheralGWM lesion (arrowhead) progressed to elevated ADC by out-come. Note slight hyperintensity on the outcome T2-weightedimage (black arrow) and an old lesion on the contralateral side(double black arrows).

Yang et al Serial Study of Diffusion Abnormalities in Human Stroke 2387

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

The combined ADC and ADA information may provideinsight into cellular changes during ischemia evolution. Inacute ischemia, early disruption of energy metabolism leadsto failure of transmembrane ion pumps and cell swelling(cytotoxic edema).39 As infarction evolves, vasogenic edemamay develop as a result of the blood-brain barrier breakdown,causing excessive water accumulation and tissue swelling.40

However, both cytotoxic and vasogenic edema may presentsimultaneously in ischemic lesions, and early vasogenicedema may not always cause ECS swelling.4 Thus, it ishelpful to regard the progression of an ischemic lesion withspecific compartment-related swelling.

Different ADC models could be applied in attempt toexplain the combined ADC and ADA changes. In the ECSmodel, the ADC of water is presumed to be high in the ECSand low in the ICS, with the ADC reduction in acute ischemiaexplained by the net shift of water from the ECS into the ICS,as a result of cell swelling in cytotoxic edema.1,6,24 Inaddition, the shrinkage of the ECS with increased ECStortuosity28 may cause increased restriction of extracellularwater movement, leading to elevated ADA.31 While this mayqualitatively explain the reduced ADC with elevated ADA inphase 1, excessive water accumulation and ECS swelling invasogenic edema would lead to reduced ADA accompaniedby elevated ADC (phase 3). However, reduced ADC andreduced ADA (phase 2) were observed in many cases withT2-weighted image hyperintensities and extensive tissuedistortions, suggesting excessive water accumulation andECS swelling due to vasogenic edema. Thus, phase 2 couldnot be easily explained by this model.

In other models, the ADC reduction in acute ischemia maybe dominated by decreased ADC in the ICS due to decreasedmembrane permeability12 or cytoplasmic circulation or in-creased viscosity of water.13–15In these ICS models, the ADCof water in the ICS and ECS may be similar.15 To explain ourresults of combined ADC and ADA changes, we postulatethat elevated ADA may reflect enhanced restriction of intra-cellular water movement, due to decreased membrane perme-ability, or enhanced ICS weighting, due to water shift fromthe ECS into ICS caused by cell swelling. However, cellswelling may cause reduced restriction of ICS water move-ment and hence a reduction of the elevated ADA. The level ofADA elevation may reflect the degree of cellular swelling

and membrane degradation and hence the severity of theischemic injury. Additional ECS swelling and possible mem-brane fragmentation may lead to further ADA reductionbelow normal. Therefore, a decline in ADA most probablyreflects the process of cell swelling, extracellular edema, andcell lysis. The reduced ADA in phase 2 may be characterizedby the development of ECS swelling and membrane degra-dation, while the reduced ADC can be explained by the ICSmodels. In contrast, persistently elevated ADA (phase 1) maysuggest lack of ECS swelling and preserved membraneintegrity, although vasogenic edema may be present at theearly ischemic stage.4 The elevated ADC and reduced ADA(phase 3) most probably reflects cell lysis toward necrosiswith the destruction of membrane integrity.4,5 Recently,reduced ADA and elevated ADC have been found in chronicwhite matter lesions of ischemic leukoaraiosis, which isconsistent with axonal loss and gliosis.41

In patients with acute stroke observed in this study, theduration of the initial ADC reduction ranged from,48 hoursto 85 days. The observed differences in ADC and ADAevolution support that ADC heterogeneity within the lesionreflects different temporal rates of stroke progression.5,27 Thefeature of a lesion with elevated ADC in the infarct core anda rim of reduced ADC at a later stage (Figure 4) has beenreported in animal model studies3,4 with histopathologicalcorrelates of later development of infarction in the rim.4 Inthis study, the elevated ADA in the rim supports laterdevelopment of ischemic infarction rather than edema. Atoutcome, the ADC in the rim is only slightly elevated incomparison with the infarct core, which may suggest differentpathological processes. Furthermore, the outcome ADC ele-vation of the rim is similar to that of the small cortical GMlesions (Figure 2, bottom row). Although the measurement ofADC in small lesions or the rim area of larger lesions may beinfluenced by PVE, the slow recovery of the ADC and ADAin the WM lesion case (Figure 6) is less likely due to PVE.This may be a human case of reversible focal ischemicinjury42 or ischemic-induced spreading depression found inanimal models.43 On the other hand, the slow evolution ofADC and ADA in these cases (in situations of better collateralflow or early reperfusion) may relate to other processes suchas apoptosis, in which selective delayed cell death is likely tooccur in areas of milder ischemic injury for days after theinitial insult.44

The correlation of ADA changes within 12 hours with theacute, subacute, and outcome clinical scores supports thatearly ADA changes reflect severity of the ischemic injury andpredict stroke outcome. In contrast, the lack of correlationbetween clinical scores and ADA changes at later stagessuggests that other processes, such as ECS swelling invasogenic edema and cell lysis with membrane fragmenta-tion, may have developed and contributed to a heterogeneousevolution. Furthermore, no significant correlation betweenADC and clinical scores was found, suggesting that ADCalone does not have sufficient predictive power in terms ofhistopathological and clinical outcome in patients with acuteischemic stroke.5,19This contrasts with a study in which ADCmeasured within 60 hours of stroke onset correlated withstroke outcome at 4 months.45 Such disparity may relate to

Figure 7. Plot of ADA vs ADC ratios for all WM (square), GM(circle), and GWM (diamond) lesions over serial measurements.Solid symbols relate to special cases discussed in the text.Solid lines are linear regression results for transitions fromphase 1 to 2 (r520.41, P50.0005) and from phase 2 to 3(r520.52, P,0.00005), respectively.

2388 Stroke November 1999

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

different sampling methods. The other study reported system-atically higher ADC values by measuring the entire lesionslice, including the peripheral area.45 This may becomeparticularly important for small lesions, in which a measuredhigher ADC value may be coupled with more PVE and hencebetter outcome biased by the small lesion size. Second, itignored the heterogeneity of ADC within the lesion, whichmay relate to different pathophysiological properties, partic-ularly between the infarct core and periphery.3–5,27 Differentprocesses within the lesion may already have developedwithin 60 hours of stroke onset. Nevertheless, whether thereis an ADC threshold in predicting tissue viability and strokeoutcome requires further understanding of the mechanismsunderlying the ADC changes in acute ischemia. However,ADA may be the better diffusion property to track strokeprogression.

Recently, elevated ADA in the acute “ischemic penumbra”delineated by DWI and CEPI has been observed and relatedto tissue salvage.46 This may reflect possible membranechanges related to autoregulation of local blood flow in acuteischemia. It is possible that the ADA changes in ischemiamay be influenced by early reperfusion. This is beyond thescope of the present study. Further studies to correlate theADA changes with perfusion parameters and MRA resultsmay help to elucidate the underlying pathophysiology inacute ischemia.

It should be noted that the empirical relationship betweenADA fr and ADAsd was not expected theoretically. For anindividual cell or fiber structure, the ADAsd value is variable,depending on specific orientation of the cell relative to thediffusion-encoding directions. For a macroscopic ROI con-taining a large amount of differently orientated cells, it ispossible that the average effect may reduce the orientation-dependent influence. In addition, the use of relative ratioscomparing the same tissue types in the same subject mayfurther minimize this influence and contribute to the observedcorrelation between ADAfr and ADAsd* ratios. However, theempirical relationship observed in this study may not alwaysbe valid and should not be generalized. This may be subjectto further evaluation with more available DTI data.

ADA, as another diffusion property of water moleculesprovided by DWI, is readily available to provide additionalinsight into the cellular changes in acute ischemia. CombinedADC and ADA data are valuable in characterizing strokeevolution and predicting clinical outcome. These data mayalso help to monitor cellular changes of acute ischemia inresponse to putative drug treatments.

AcknowledgmentsThis work was supported in part by the National Health and MedicalResearch Council, National Stroke Foundation, Lord Mayors Fund(Melbourne), Schering P/L, Neurological Foundation of New Zea-land, and VJ Chapman Research Fellowship (Dr Barber). Previousfunding support and continuous proprietary research support fromGE Medical System are also acknowledged (Dr Yang).

References1. Moseley ME, Kucharczyk J, Mintorovitch J, Cohen Y, Kurhanewicz J,

Derugin N, Asgari H, Norman D. Diffusion-weighted MR imaging ofacute stroke: correlation with T2-weighted and magnetic susceptibility-

enhanced MR imaging in cats.AJNR Am J Neuroradiol. 1990;11:423–429.

2. Knight RA, Ordidge RJ, Helpern JA, Chopp M, Rodolosi LC, Peck D.Temporal evolution of ischemic damage in rat brain measured by protonmagnetic resonance imaging.Stroke. 1991;22:802–808.

3. Reith W, Hasegawa Y, Latour LL, Dardzinski BJ, Sotak CH, Fisher M.Multislice diffusion mapping for 3-D evolution of cerebral ischemia in arat stroke model.Neurology. 1995;45:172–177.

4. Pierpoli C, Righini A, Linfante I, Tao-Cheng JH, Alger JR, Di Chiro G.Histopathologic correlates of abnormal water diffusion in cerebral ische-mia: diffusion-weighted MR imaging and light and electron microscopicstudy.Radiology. 1993;189:439–448.

5. Welch KMA, Windham J, Knight RA, Nagesh V, Hugg JW, Jacobs M,Peck D, Booker P, Dereski MO, Levine SR. A model to predict thehistopathology of human stroke using diffusion and T2-weightedmagnetic resonance imaging.Stroke. 1995;26:1983–1989.

6. Warach S, Chien D, Li W, Ronthal M, Edelman RR. Fast magneticresonance diffusion-weighted imaging of acute human stroke.Neurology.1992;42:1717–1723.

7. Warach S, Gaa J, Siewert B, Wielopolski P, Edelman RR. Acute humanstroke studied by whole brain echo planar diffusion weighted MRI.AnnNeurol. 1995;37:231–241.

8. Hossmann KA, Fisher M, Bockhorst K, Hoehn-Berlage M. NMR imagingof the apparent diffusion coefficient (ADC) for the evaluation of meta-bolic suppression and recovery after prolonged cerebral ischemia.J CerebBlood Flow Metab. 1994;14:723–731.

9. Sorenson AG, Buonanno FS, Gonzalez RG, Schwamm LH, Lev MH,Huang-Hellinger FR, Reese TG, Weisskoff RM, Davis TL, SuwanwelaN, Can U, Moreira JA, Copen WA, Look RB, Finklestein SP, Rosen BR,Koroshetz WJ. Hyperacute stroke: evaluation with combined multisectiondiffusion-weighted and hemodynamically weighted echo-planar MRimaging.Radiology. 1996;199:391–401.

10. Barber PA, Darby DG, Desmond PM, Yang Q, Gerraty RP, Jolley D,Donnan GA, Tress BM, Davis SM. Prediction of stroke outcome withechoplanar perfusion- and diffusion-weighted MRI.Neurology. 1998;51:418–426.

11. LeBihan D, Breton E, Lallemand D, Grenier P, Cabanis E, Laval-JeantetM. MR imaging of intravoxel incoherent motions: application to diffusionand perfusion in neurologic disorders.Radiology. 1986;161:401–407.

12. Helpern JA, Ordidge RJ, Knight RA. The effect of cell membrane waterpermeability on the diffusion coefficient of water. In: Proceedings of XIAnnual Meeting of the Society of Magnetic Resonance in Medicine;August 8–14, 1992; Berlin, Germany. 1992;1:1201.

13. Wick M, Nagatomo Y, Prielmeier F, Frahm J. Alteration of intracellularmetabolite diffusion in rat brain in vivo during ischemia and reperfusion.Stroke. 1995;26:1930–1934.

14. van der Toorn A, Dijkhuizen RM, Tulleken CAF, Nicolay K. Diffusion ofmetabolites in normal and ischemic rat brain measured by localized1HMRS. Magn Reson Med. 1996;36:914–922.

15. Duong TQ, Ackerman JJH, Ying HS, Neil JJ. Evaluation of extra- andintracellular apparent diffusion in normal and globally ischemic rat brainvia 19F NMR. Magn Reson Med. 1998;40:1–13.

16. van der Toorn A, Sykova E, Dijkhuizen RM, Vorisek I, Vargova L,Skobisova E, Campagne ML, Reese T, Nicolay K. Dynamic changes inwater ADC, energy metabolism, extracellular space volume, and tor-tuosity in neonatal rat brain during global ischemia.Magn Reson Med.1996;36:52–60.

17. Ford JC, Hackney DB, Lavi E, Phillips M, Patel U. Dependence ofapparent diffusion coefficients on axonal spacing, membrane perme-ability, and diffusion time in spinal cord white matter.J Magn ResonImaging. 1998;8:775–782.

18. Chien D, Kwong KK, Gress DR, Buonanno FS, Buxton RB, Rosen BR.MR diffusion imaging of cerebral infarction in humans.AJNR Am JNeuroradiol. 1992;13:1097–1102.

19. Warach S, Moseley M, Sorensen AG, Koroshetz W. Time course ofdiffusion imaging abnormalities in human stroke [letter]; Welch KMA,Levine SR, Chopp M, Knight RA, D’Olhaberriague L, Boska MD,Nagesh V, Windham JP, Peck D. [response].Stroke. 1996;27:1254–1256.

20. Ulug AM, Beauchamp N Jr, Bryan RN, van Zijl PC. Absolute quanti-tation of diffusion constants in human stroke.Stroke. 1997;28:483–490.

21. Warach S, Boska M, Welch KMA. Pitfalls and potential of clinicaldiffusion-weighted MR imaging in acute stroke.Stroke. 1997;28:481–482.

Yang et al Serial Study of Diffusion Abnormalities in Human Stroke 2389

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from

22. Pierpaoli CJ, Basser PJ. Toward a quantitative assessment of diffusionanisotropy [published correction appears inMagn Reson Med. 1997;37:972].Magn Reson Med. 1996;36:893–906.

23. Basser PJ, Pierpaoli CJ. Microstructural and physiological features oftissues elucidated by quantitative-diffusion tensor MRI.J Magn Reson.1996;B111:209–219.

24. van Gelderen P, de Vleeschouwer MHM, DesPres D, Pekar J, van ZijlPCM, Moonen CTW. Water diffusion and acute stroke.Magn ResonMed. 1994;31:154–163.

25. Schlaug G, Siewert B, Benfield A, Edelman RR, Warach S. Time courseof the apparent diffusion coefficient (ADC) abnormality in human stroke.Neurology. 1997;49:113–119.

26. Schwamm LH, Koroshetz WJ, Sorensen AG, Wang B, Copen WA,Budzik R, Rordorf G, Buonanno FS, Schaefer PW, Gonzalez RG. Timecourse of lesion development in patients with acute stroke: serialdiffusion- and hemodynamic-weighted magnetic resonance imaging.Stroke. 1998;29:2268–2276.

27. Nagesh V, Welch KMA, Windham JP, Patel S, Levine SR, Hearshen D,Peck D, Robbins K, D’Olhaberriague L, Soltanian-Zadeh H, Boska MD.Time course of ADCw changes in ischemic stroke: beyond the human eye!Stroke. 1998;29:1778–1782.

28. Sykova E, Svoboda J, Polak J, Chvatal A. Extracellular volume fractionand diffusion characteristics during progressive ischemia and terminalanoxia in the spinal cord of the rat.J Cereb Blood Flow Metab. 1994;14:301–311.

29. Thornton JS, Ordidge RJ, Penrice J, Cady EB, Amess PN, Punwani S,Clemence M, Wyatt JS. Anisotropic water diffusion in white and graymatter of the neonatal piglet brain before and after transient hypoxia-is-chemia.Magn Reson Imaging. 1997;15:433–440.

30. Carano RAD, Li F, Irie K, Fisher M, Sotak CH. The temporal evolutionof diffusion anisotropy in the ischemic rat brain. In: Proceedings of theSixth Scientific Meeting of the International Society of Magnetic Res-onance in Medicine; April 18–24, 1998; Sydney, Australia; p 530.

31. Armitage PA, Bastin ME, Marshall I, Wardlaw JM, Cannon J. Diffusionanisotropy measurements in ischaemic stroke of the human brain.MAGMA. 1998;6:28–36.

32. Yang Q, Desmond PM, Tress BM, Barber PA, Darby DG, Gerraty RP,Davis SM. Apparent diffusion anisotropy in normal and ischemic humanbrain. In: Proceedings of the Sixth Scientific Meeting of the InternationalSociety of Magnetic Resonance in Medicine; April 18–24, 1998; Sydney,Australia; p 1250.

33. Zelaya F, Flood N, Chalk JB, Wang D, Strugnell W, Doddrell DM,Strugnell W, Benson M, Ostergaard L, Semple J, Eagle S. An evaluation

of the time dependence of the anisotropy of the water diffusion tensor inacute human ischemia.Magn Reson Imaging. 1999;17:331–348.

34. Cote R, Battista RN, Wolfson C, Boucher J, Adam J, Hachinski V. TheCanadian Neurological Scale: validation and reliability assessment.Neu-rology. 1989;39:638–643.

35. Wade DT. Measurement in Neurological Rehabilitation. Oxford,England: Oxford Medical Publications; 1994.

36. Stejskal EO, Tanner JE. Spin diffusion measurements: spin-echoes in thepresence of a time-dependent field gradient.J Chem Physiol. 1965;42:288–292.

37. Haselgrove JC, Moore JR. Correction for distortion of echo-planarimages used to calculate the apparent diffusion coefficient.Magn ResonMed. 1996;36:960–964.

38. Park S, Schowengerdt R. Image reconstruction by parametric cubic con-volution. Comput Vis Graph Image Proc. 1983;23:258–272.

39. Siesjo BK, Bengtsson F. Calcium fluxes, calcium antagonists, andcalcium-related pathology in brain ischemia, hypoglycemia, andspreading depression: a unifying hypothesis.J Cereb Blood Flow Metab.1989;9:127–140.

40. Loubinoux I, Volk A, Borredon J, Guirimand S, Tiffon B, Seylaz J.Spreading of vasogenic edema and cytotoxic edema assessed by quanti-tative diffusion and T2 magnetic resonance imaging.Stroke. 1997;28:419–427.

41. Jones DK, Lythgoe D, Mark A, Horsfield MA, Simons A, Williams SCR,Markus HS. Characterization of white matter damage in ischemic leuko-araiosis with diffusion tensor MRI.Stroke. 1999;30:393–397.

42. Minematsu K, Li L, Sotak CH, Davis MA, Fisher M. Reversible focalischemic injury demonstrated by diffusion-weighted magnetic resonanceimaging in rats.Stroke. 1992;23:1304–1311.

43. Rother J, de Crespigny AJ, D’Arceuil H, Iwai K, Moseley ME. Recoveryof apparent diffusion coefficient after ischemia-induced spreadingdepression relates to cerebral perfusion gradient.Stroke. 1996;27:980–987.

44. Pulera MR, Adams LM, Liu H, Santos DG, Nishimura RN, Yang F, ColeGM, Wasterlain CG. Apoptosis in a neonatal rat model of cerebralhypoxia-ischemia.Stroke. 1998;29:2622–2630.

45. van Everdingen KJ, van der Grond J, Kappelle LJ, Ramos LMP, MaliWPTM. Diffusion-weighted magnetic resonance imaging in acute stroke.Stroke. 1998;29:1783–1790.

46. Yang Q, Tress BM, Barber PA, Li T, Darby DG, Gerraty RP, DesmondPM, Davis SM. Prognosis of tissue viability in human stroke usingdiffusion and perfusion imaging. In: Proceedings of the Annual Meetingof the American Society of Neuroradiology; May 23–28, 1999; SanDiego, Calif; p 119.

2390 Stroke November 1999

by guest on October 20, 2014http://stroke.ahajournals.org/Downloaded from