Breath-hold perfusion and permeability mapping of hepatic malignancies using magnetic resonance imaging and a first-pass leakage profile model

��

�� !� �� "� #� #�� $� �� %�� &� ��'

�� !"#" �� $�� "�� %�� &% ��'�� (��) #�� (�� !* +�,� �

Received 3 April 2001; Revised 10 July 2001; Accepted 24 July 2001

ABSTRACT: We have applied a novel pharmacokinetic model of the distribution of contrast media to dynamiccontrast-enhanced MRI data from patients with hepatic neoplasms. The model uses data collected during the passageof a bolus of contrast medium and allows breath-hold image acquisition. The aims of the study were to investigate thefeasibility of permeability mapping using the first pass technique and breath-hold acquisitions, and to examine thereproducibility of the technique and the effect of the liver’s dual vascular supply on the assumptions of the model.Imaging was performed in 14 patients with hepatic neoplasms. Dynamic data clearly demonstrated differences in thetiming and shape of the contrast medium concentration–time course curve in the systemic arterial and portal venoussystems. Mapping of the arrival time (T0) of contrast medium allowed identification of tissue supplied by the hepaticarteries and portal vein. Hepatic tumours all showed typical hepatic arterial enhancement. Repeated measurements ofendothelial permeability surface area product (kfp) and relative blood volume (rBV), performed in five patients,showed excellent reproducibility with variance ratios (Vr) of 0.134 and 0.113, respectively. Measurement ofenhancing tumour volume was also highly reproducible (Vr = 0.096) and this was further improved by the use of T0

maps to identify pixels supplied by the hepatic artery (Vr = 0.026). Estimates of kfp and rBV in normal hepatic tissuesupplied by the portal vein were highly inaccurate and these pixels were identified by use of the T0 parameter andexcluded from the analysis. In conclusion, dynamic MRI contrast enhancement combined with a pharmacokineticmodel of the distribution of contrast media in the first pass allows us to produce highly reproducible parametric mapsof kfp and rBV from hepatic tumours that are supplied by the hepatic arterial system using breath-hold acquisitions.Copyright 2002 John Wiley & Sons, Ltd.

KEYWORDS: liver neoplasm; dynamic MRI; pharmacokinetic model; perfusion; endothelial permeability

($%)*+,-%(*$

The quantification of endothelial permeability in tumourshas attracted increasing attention in recent years with therecognition that the growth of tumours is dependant onthe development of new blood vessels via a processknown as angiogenesis.1 This recognition has led to therealization that the characteristics of the tumoral micro-vasculature such as vascular density and endothelialpermeability can provide important information aboutprognosis and tumour behaviour.2–5 In particular there

has been considerable research activity in developingnovel therapeutic agents that utilize the characteristics ofthe angiogenic process to specifically target the devel-oping microvasculature.6,7

We have recently described a new technique forsimultaneous calculation of blood volume and endothe-lium permeability images from large three-dimensionaldata sets in patients with brain tumours.8,9 The novelfeature of this model is that it uses only data collectedduring the first passage of the bolus of contrast mediathrough the target tissue so that data acquisition isextremely fast compared with the conventional methodfor measuring permeability.10–12 Since data acquisitioncan be performed in a single breath-hold, it has thepotential to be used in abdominal studies whererespiratory motion complicates the collection of dynamicdata for the measurement of contrast enhancement.

Unlike normal brain the microcirculation in normalliver is highly permeable to transitional metal ions andcompounds of gadolinium or manganese, such as

NMR IN BIOMEDICINENMR Biomed. 2002;15:164–173Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nbm.729

*Correspondence to: A. Jackson, Division of Imaging Science andBiomedical Engineering, University of Manchester, Manchester M139PT, UK.Email: [email protected]

Abbreviations used: Gd-DTPA, gadinium diethylene triamine penta-acetic acid; kfp, permeability surface area product; LP, leakage profile;rBV, relative blood volume; rf, radiofrequency; RPR, ratio of the peakconcentration of contrast agent to the concentration in recirculationphases; To, arrival time; VIF, vascular input function; Vr, varianceratios.

Copyright 2002 John Wiley & Sons, Ltd. NMR Biomed. 2002;15:164–173

feromoxides13,14 and gadolinium diethylene triaminepenta-acetic acid (Gd-DTPA).15 In addition the bloodsupply to the liver is derived jointly from the hepaticarteries and the portal venous system so that systemicallyadministered contrast agents arrive at the liver in twodistinct phases.16 In this study, we have applied the first-pass analysis technique in a patient group with hepaticneoplasm to investigate the feasibility of permeabilitymapping using the first pass technique and breath-holdacquisitions, to examine the reproducibility of thetechnique and the effect of the liver’s dual vascularsupply on the assumptions of the model.

.�%/)(�#0 �$+ ./%�*+0

��

Fourteen patients with known liver tumours wereincluded in the study. All patients gave informed consentand the Central Manchester Healthcare Trust MedicalEthics Committee and the South Manchester Ethicalcommittee approved the study. Table 1 shows the demo-graphic data and histological diagnosis for each patient.

-��

MR images were obtained on a 1.5 T Philips ACS NT-PT6000 MR system with maximum gradient strength of23 mT/m and maximum slew rate of 105 mT/m/ms(Philips Medical Systems, Best). Prior to the MRexamination a 16G catheter was inserted into an ante-cubital vein using local anaesthetic (lignocaine, s.c.). Theimaging protocol for dynamic contrast enhanced studiesconsisted of three consecutive three-dimensional radio-frequency (rf) spoiled (T1-weighted) field echo acquisi-tions with an array of flip angles (� = 2, 10 and 35°) toallow calculation of T1 maps. The third sequence wasrepeated (n = 6, duration T = 24.6 s) for baseline acquisi-tion. The same sequence was then repeated to produce aT1-weighted dynamic data set (T1dy) with a time resolu-tion of �t = 4.1 s and a duration of 41 s. All images wereacquired in the axial plane and MR parameters are shownin Table 2. Contrast agent (0.1 mmol/kg of gadodiamide,Gd-DTPA-BMA; Omniscan, Nycomed, Oslo) was givenas a manual intravenous bolus injection over a period of 4s and was immediately followed by a bolus of 35 ml ofnormal saline injected at the same rate. The injection wasstarted as the dynamic imaging sequence was initiated.The images were all collected during breath holding. Thefirst three images providing the variable flip angle datawere collected in one breath-hold (19 s) and the baselineimages were collected in a second breath-hold (18 s). Thedynamic enhancement data was collected using a 41 sdata acquisition and the patient was asked to hold theirbreath for as long as was comfortable during thisacquisition. Data for the analysis is adequate if the datacollection covers the initial precontrast baseline to thefirst two collection points after the recirculation bolus.This can be achieved in as little as 25 s if the timing of thebolus arrival and the image collection coincide. Addi-tional data collected after the patient begins to breath waseasily identified by the presence of movement artefactand was discarded. All patients were rehearsed in the MRprotocol immediately prior to the examination. Patientswere not imaged at any specific time of day and noattempt was made to control their dietary intake prior toscanning.

Five patients were reimaged again 48–56 h after theinitial studies in order to establish the reproducibility ofthe technique (Table 1).

%�� +�� 1��

Patientnumber

Age/sex Diagnosis

Tumorlocation

1 59/M Cavernous hemangioma Caudate lobe2 46/F Cavernous hemangioma Left Lobe3 46/M Colorectal adenocarcinoma Multi-focal4* 55/M Colorectal adenocarcinoma Multi-focal5* 48/F Colorectal adenocarcinoma Right Lobe6* 38/M Colorectal adenocarcinoma Right lobe7* 64/M Colorectal adenocarcinoma Multifocal8* 57/F Colorectal adenocarcinoma Right Lobe9 44/F Colorectal adenocarcinoma Right Lobe

10 69/M Colorectal adenocarcinoma Multifocal11 43/F Ovarian serous carcinoma Right Lobe12 51/F Ovarian serous carcinoma Right Lobe13 68/F Hepatocellular carcinoma

(HCC) in non-cirrhotic liverRight lobe

14 51/M HCC in cirrhotic liver Right lobe

Asterisks indicate patients in whom imaging was performed on twooccasions.

%�� '� (�� 2 �� 3� ��4�� 5�6� �� 5%6

Scan Sequences TR/TE/� Matrix �t (s) T (s)

FE(2°) 3D T1W-FE 4.1/1.2/2° 128 � 128 � 25 — 4.2FE(10°) 3D T1W-FE 4.1/1.2/10° 128 � 128 � 25 — 4.2T1dy(35°) 3D T1W-FE 4.1/1.2/35° 128 � 128 � 25 5.1 50.0T2*dy 2D T2*W-FEEPI 221/30/35° 128 � 128 � 9 1.86 50.0

3D, three dimensional; T1W, T1-weighted.


BREATH-HOLD PERMEABILITY IMAGING OF HEPATIC MALIGNANCY 165

+��

All images were transferred to an independent work-station for analysis (Sun Microsystems, Palo Alto, CA).

.�3�� The spatial positions of the liverin the three separate breath-hold acquisitions werecompared by performing spatial coregistration of thehepatic component of each data set to the baseline scan.The coregistration algorithm chosen uses a calculation ofthe dot product of gradient edge strengths and is thereforesuitable for comparison of images with divergent contrastcharacteristics. The transformation matrices from thecoregistration procedure were used to identify patientswith significant misregistration (greater than one voxel inany direction) between the breath-hold sequences.

Misregistration of this magnitude was seen in twocases and misregistered data was realigned using a scaledsinc kernel.17

-�� Maps ofproton density (M0) and intrinsic longitudinal relaxationrate (R10 � 1/T10) maps were calculated by fitting thesteady-state T1-FE signals S(�) with the Ernst formula(assuming TE � T2*):

S�� M0 � sin� � �1 � E10��1 � cos� � E10� �1�where � has three discrete values (� = 2, 10 and 35°) andE10 = exp(�TR�R10).

Four-dimensional maps of longitudinal relaxation rate[R1(t)] maps were calculated using signal intensity datafrom pre- and post-contrast T1-FE images [S(t) � S(0)]:

R1�t� � ��1�TR� � ln1 � �A � B��1 � cos� � �A� B��

�2�where � = 35°, TR = 4.3–7.0 ms, A = [S(t) � S(0)]/(M0�sin�), B = (1 � E10)/(1 � cos��E10).

Four-dimensional Gd-DTPA-BMA concentrationmaps were then calculated for each dynamic phase:

C�t� � R1�t� � R10��1 �3�where 1 is the relaxivity of Gd-DTPA-BMA determinedexperimentally, 1 = 4.39 s�1 mM�1 (at 37°C).

-�� 7

%�� In recent work we have introduced a newparameter, which we call the leakage profile of contrastagent.8 The leakage profile is based on a well-establishedpharmacokinetic model as shown below:18

dC1�dt � ktransCp�t� � C1�t��V1� �4�where V1 is the extra-vascular contrast distribution

volume, C1 is the concentration of contrast in the extra-vascular space, ktrans is the endothelial permeabilitysurface area product and Cp is the concentration ofcontrast agent in plasma.

If we assume that back-flux can be ignored during thefirst vascular pass of contrast bolus, eqn (1) becomes:

V1 � dC1�dt � ktrans � Cp �5�Thus, the extravascular component of the total contrastconcentration in a voxel is equal to the transfer constant,now called k first pass (kfp), times the integral of theplasma concentration of contrast agent.

C�t� � kfp

� t

Cp�t��dt� �6�

We call this function the leakage profile (LP):

LP�t� �� t

Cp�t��dt� �7�

+�� 3��

Calculation of kfp using this approach requires accurateidentification of the temporal variation of contrast agentconcentration in the blood vessels supplying the tissue.This function is commonly referred to as the arterial, ormore correctly the vascular input function (VIF). In thecase of the liver the organ is supplied by two distinctvascular systems. Approximately 10–15% of the bloodsupply to normal hepatic parenchyma is derived from theaorta via the hepatic arteries. The majority (75–85%)comes from the portal vein via the splenic and intestinalcirculation. This creates a complex situation making theidentification of the correct VIF for any individual voxeldifficult unless one or other of the vascular supplies isheavily dominant. In contrast to hepatic parenchyma thearterial supply to hepatic tumours is usually derivedwholly or largely from the hepatic artery. A VIF wastherefore determined from the aorta or hepatic artery inall cases. It must be noted, however, that this VIF isentirely inappropriate for use in the hepatic parenchymasupplied predominantly via the portal venous system(vide infra). In order to provide an indicator to the sourceof vascular input in each voxel, a parametric map of thetime of contrast arrival (T0) was generated and was usedto identify voxels where the vascular supply arisespredominantly from the portal system.

-�� 7

Contrast concentration changes in tissue over time on T1-


166 A. JACKSON ET AL.

weighted images represent a cumulative effect of intra-and extravascular contrast agent. Failure to compensatefor this effect will result in the overestimation of ktrans invoxels with significant intravascular contrast. This will inturn lead to a bias in the estimation of ktrans, which willappear artificially high in tumours with a larger perfusedblood volume. This effect has been called ‘pseudoperme-ability’ and can be addressed in a number of ways. Theseinclude the use of simple cut-off threshold values forktrans and attempts to correct data using rBV mapscalculated from T2*-weighted data. T2*-weighted datacan be collected in a separate acquisition or simulta-neously using dual echo techniques to derive both T1 andT2 weighted measurements.

We have used a high temporal resolution T1-weighteddata collection and employed post-processing techniquesto decompose the data for each voxel into intra andextravascular components. The post processing strategyhas the following steps:

1. A vascular input function [Cp(t)] is obtained foreach patients from voxels in the hepatic artery orabdominal aorta.

2. The ratio of peak Cp(t) and steady-state Cp(t),marked by the beginning of recirculation of contrastbolus, is measured from the VIF curve for use laterin the analysis. This value represents the ratio of thepeak concentration of contrast agent to the con-centration in recirculation phases and is abbreviatedto RPR.

3. The measured Cp(t) curves are integrated tocalculate the tumour LP.

4. For each voxel the leakage profile is used to definethe shape of the extravascular contrast component,with kfp providing the scaling factor [eqn (6)]. Theresidues from each voxel between the extravascularcomponent defined by the LP and the actual datadefine the intravascular component. In order toperform this step the timing of the leakage profile ismodified so that the contrast arrival time for the LPcoincides with the contrast arrival time observed inthe voxel.

5. An iterative process is then used to optimize theseparation of these intra- and extravascular compo-nents. The iterative process determines the accuracyof the iterative decomposition by assessing thegoodness of fit of the decomposed intravascularcontrast concentration data to the RPR measured instep 2. The measured RPR is then used to modify theestimate of the intravascular contrast concentrationin the recirculation phase to conform to the ratiomeasured in the input function. The process pro-ceeds through this iterative loop until the differencebetween two successive estimations of kfp is smallerthan � (= 10�4 min�1 in the current study).

The decomposed concentration–time course data werethen used to derive estimates of kfp using eqn (6) and rBV

expressed as the area under the curve of the intravascularcontrast concentration. Parametric images of kfp and rBVwere thresholded using T0 measurements in order toremove normal liver tissue supplied by the portal system.Volumes of interest within tumours were defined frommaps of contrast enhancement generated by subtractionof pre-contrast from post contrast images. Non-enhan-cing tumour tissue was not included in the volume ofinterest.

.��

In five cases maps of kfp and rBV were available from thesame patient on separate examinations. This data wasused to assess the reliability of repeated measurements ofkfp and rBV. The reproducibility of semi-automated andmanual measurement of tumour volume was assessed.Manual tumour volume estimates were derived bymanual placement of a volume of interest on maps ofenhancement generated by subtraction of precontrastfrom enhanced images. The semi-automated method useda thresholding approach based on the parametric maps ofT0. Histogram analysis of these maps demonstrates aclear bimodal distribution corresponding to systemicarterial and portal venous supply. A threshold midwaybetween the mean values of these identifies voxelssupplied primarily by the arterial tree; these thresholdedimages contain multiple small areas of low T0 due to theinclusion of parenchymal branches of the hepatic arteries.These were removed using a morphological closingoperator followed by a conditional dilation.

A single-tailed Wald–Wolfowitz runs (W-W) tests19

was used to test for differences in the distributions(histograms) of kfp and rBV values measured on days 0and 2. Median and 97.5% centiles of kfp and rBV for thewhole tumor volume were calculated for each patient andfor each observation and were used as an indicator of thespread of the data.20,21

Reproducibility of volume measurements and of themedian and 97.5% centile values of kfp and rBV wasassessed by measurement of the variance calculated as:

V ��x1 � x2�

n � �2

where x1 and x2 are the values measured on the first andsecond visits and n is the number of subjects. This figurerepresents a direct estimate of error on repeated measure-ments. The ratio of the variance (Vr) was also calculatedas Vr = V�1/x where x is the mean value Since this studyused a relatively small number of cases the potential erroron the estimate of V was assessed using standard Fstatistics.



)/0,#%0

Examination of the data from individual patients showedremarkably reliable spatial co-registration between datacollected on separate breath holds. In 12 cases misregis-tration was less than one voxel in any direction. In thetwo remaining cases data collected during one breath-hold was slightly misaligned by one to three voxels. Sincreslicing of the data produced adequate realignment withno evidence of distortion due to non-rigid tissuedeformation.

Examination of the vascular supply to the liverdemonstrated little difference in the timing or shape ofthe leakage profile derived from the aorta and the hepaticartery (Fig. 1). Portal vein flow was typically delayed by10–17 s after the arterial supply and demonstrated noevidence of any first-pass peak (Fig. 1). In all cases theshape of the portal vein curve was a smooth increase incontrast concentration. All enhancing tumour tissuedemonstrated rapid increase in contrast concentrationthat peaked whilst the portal vein concentrations werestill rising (Fig. 2). Normal liver tissue showed a slowdelayed rise in concentration mirroring the timing of theportal vein contribution.

Examination of T0 maps clearly demonstrates earlyarterial passage of contrast, an intermediate phase asblood passes into the capillary beds of arterially suppliedtissues such as the spleen and hepatic neoplasms and alate portal phase due to blood from the hepatic portalsystem supplying the normal hepatic parenchyma(Plate 1).

Maps of T0, kfp and rBV were subjectively of highquality with homogeneous distributions of parametricvalues in normal tissues and clear delineation of smallstructures such as the branches of the portal vein andhepatic arteries (Plate 2). Examination of the maps fromnormal hepatic tissue and from portal system tributaries

shows extremely high values of kfp and very low valuesof rBV in areas of long T0. These values are incorrect andreflect failure of the post-processing algorithm whichassumes a VIF with a distinct first pass effect such as isseen in the arterial tree (see discussion). In order to avoiderrors in the measurement of kfp and rBV, theseparametric images were masked using a threshold valueof T0 (10 s). Tissues supplied by the hepatic portal systemare represented by null values (not a number) in the finalparametric maps (Plate 3). Masked images demonstratehepatic pixels only where branches of the hepatic arteryresult in short T0 or where pathological tumour tissue issupplied by the hepatic arterial system.

In two patients with cavernous haemangiomas therewas very little perfusion during the first pass and valuesof rBV could not be measured. Values of kfp weremeasurable but were very small. The distribution of

8�� -./ � -�/ �'�� 0�� 0��'��1�� 2 �� ! � �� 0��)�� 2 -�/&��3�� '� �'�� -��4�� 5��/ ��'�� 0� ��

8�� '� "��6�� '�� '�� )� �� -�� +/



��

�� ! �"� � �� #�� #�� ! �"� �� ! � �� ! ��

Copyright 2002 John Wiley & Sons, Ltd. NMR Biomed. 2002;15

BREATH-HOLD PERMEABILITY IMAGING OF HEPATIC MALIGNANCY

�� ! � �� "� � �� $�� % �� %� � �� &�� !

Copyright 2002 John Wiley & Sons, Ltd. NMR Biomed. 2002;15

A. JACKSON ET AL.

median values of rBV and kfp for all cases is shown inFig. 3.

)��

Comparison of the pixel value distributions of kfp andrBV from the two investigations in individual patientsdemonstrated no significant differences in any case.Figure 4(A) and (B) illustrates the relationship betweenmedian values of kfp and rBV measured for each of thepaired examinations. Figure 4(C) and (D) also illustratesthe relationship between the upper 5% of pixel values forrBV and kfp on the two scans. Figure 5(A) and (B)illustrates the relationship between tumour volumesmeasured on the two scans using the manual and semi-automated techniques.

The measures of variance and ratio of the variance foreach variable are shown in Table 3. F statistics show that

the potential error on the estimates of variance isexpected to be below 85%.

+(0-,00(*$

Quantification of the rate of contrast agent leakageprovides a potential mechanism for the assessment ofcapillary endothelial permeability and allows the gen-eration of parametric images that can provide informationabout the heterogeneity of microvascular function withinabnormal tissues.10,22,23 There are many potentialmethodological approaches but it is important to realizethat none is entirely satisfactory.22 Methods that expressthe characteristics of contrast leakage rate purely fromobserved changes in signal intensity are prone tovariations in scanning protocol, amplifier gain, injectiontechnique, cardiovascular variation between individualsand other factors.10,22,24,25 This has led to the develop-ment of techniques that use imaging data to derivephysiological indicators such as endothelial permeabilitysurface area product (ktrans) and relative blood volume(rBV), which should in theory be free from theseeffects.22 In principle the calculation of ktrans requiresaccurate measurements of variations in contrast agentconcentration with time from the extracellular spacewithin the tissue and from the intravascular space in theblood vessels which supply it. With this data ktrans can becalculated simply using eqn (4).

Applying these principles to MRI data is complicatedby the restrictions imposed by the imaging method.Firstly, the derivation of contrast agent concentrationfrom signal intensity data requires accurate measurementof the T1 value of the tissue prior to contrast adminis-tration which can be technically complex and time-consuming.26–28 Secondly, the measurement of time-varying responses in contrast concentration assumesaccurate spatial registration of images throughout thedata collection period. Thirdly, the measurement of thetemporal variations in the intravascular contrast concen-tration is complicated by the mechanism of contrastdistribution following venous injection.25,29–31 Lastly,the measurement of the temporal variations in contrastconcentration in the extracellular, extravascular space iscomplicated by the need to acquire data from voxelswhich commonly contain not only permeable capillarybeds but also an unknown proportion of medium or largeblood vessels.8 In the liver these problems are furthercompounded by the presence of a dual vascular supply.16

The technique we describe uses a series of three T1-weighted images acquired with varying flip angles toallow calculation of R10. This technique has beendescribed and validated elsewhere.32 In the liver itproved possible to acquire this set of images in a singlebreath-hold and the resulting data showed no evidence ofmisregistration. However, it is not possible to performthis acquisition in the same breath-hold as the dynamic

8�� 9� �� '�� 7 �� 2�� '0�� 0�� 2 �� '� �� -�� /� �� -�� 8'��/� �� '� �� -��/� %"" �� -�� /� ��%"" �� -��/2 .� 9 ��0�� '��

%�� 9� .�� 3�� 576 �� 3�� 57�6 �� 5� : ;6� -� �� '��4� �� 3� � ��

Median V Vr

kfp 0.39 0.047 0.134rBV 5.80 0.730 0.113kfp 95% 0.41 0.076 0.170RBV 95% 8.42 1.542 0.161Volume 27.65 4.540 0.096Automated volume 34.28 1.412 0.026



contrast data, so that coregistration of the data sets relieson the patient taking comparable breaths for each scan. Itwas surprising that we found misregistration of more thanone pixel between these scans in only two of 14 studies.In the other studies automated coregistration of the datasets was straightforward and there was no evidence oftissue deformation that would invalidate the use ofautomated routine rigid body coregistration approaches.

The measurement of changes in the intravascularcontrast concentration over time is central to thesetechniques and requires careful consideration. If agradual infusion of contrast is used then a relativelylow temporal resolution may be adequate. However theuse of contrast infusions produces difficulties in attempt-ing to differentiate the effects of intravascular and

extravascular contrast agent on signal changes fromindividual voxels (vide infra).31 If a bolus injection isused then the contrast concentration changes in theintravascular compartment are complex since the bolusremains compact as it passes around the body andsignificant peaks of contrast concentration are seenduring the first two or even three circulations of thebolus.29,31 Once the contrast has undergone two or threecirculations (30–40 s) it is uniformly mixed throughoutthe circulating blood volume. The tissue uptake and renalexcretion of contrast during the passage of the bolus willdepend on the peak intravascular contrast concentrationobtained and the duration of the bolus. Since this peakconcentration is transitory this approach demands a veryhigh temporal resolution. We can decrease temporal

8�� 1� �� )�� '��0�� -��/ �� 7 -0��/ ��'�� '�� -��/ �� '�� :�� '�� -��/



resolution a little since we know that the profile of thecontrast concentration–time course curve will conform toa gamma variate function in normal vasculature. Thisallows us to use a curve fitting routine to provide datainterpolation in the temporal domain and to employ alower actual temporal resolution than would otherwise berequired. In this study we have used a time resolution ofapproximately 4 s, which is at the upper limit of theacceptable range. The requirement for such hightemporal resolution places significant restrictions on theimage acquisition protocol and, in particular, restricts thevolume and spatial resolution of the acquisition. Despitethis, the ability of modern MRI scanners to produceimages using very short TR and TE, together with thebeneficial effects of volume acquisitions on signal-to-noise ratio has allowed us to image a relatively largevolume of tissue at an acceptably high spatial resolution.

The use of high temporal resolution data, combinedwith a rapid bolus injection allows us to use the shape ofthe residue function in each voxel to drive a decomposi-tion process which will estimate the contribution ofintravascular and extravascular contrast to the observedsignal changes.8 The separation of these effects isimportant if we desire true measurements of ktrans whichare not artificially elevated by the presence of intra-vascular contrast, often referred to as ‘pseudoperme-ability’. The optimization scheme we have used todecompose the contributions of intra- and extravascularcontrast relies on an assumption that the concentrationsof contrast due to leakage into the intercellular spaceduring the first passage of the contrast bolus arenegligible. This will not be true in tissues with very highextraction fractions and the size of the error due to thisassumption is unknown. Very high extraction fractionscould lead to the underestimation of kfp and correspond-

ing underestimation of rBV, particularly if the rate ofextraction is sufficiently high to generate a peak incontrast concentration in the extravascular space duringthe passage of the bolus. One way to characterize themagnitude of these potential errors is to compare theresults of this technique with estimates of extractionfraction in muscle where the capillary permeability hasbeen extensively examined in both human and animalmodels and we have found that the first pass techniqueproduces values which are in close agreement withprevious studies (unpublished results).

The first pass technique identifies the contributions ofintra- and extravascular contrast to signal change by apost-processing algorithm based on the assumption thatintravascular contrast concentration changes will show apattern of temporal variation identical to that seen in theVIF. The algorithm uses features of the VIF (LP andRPR), which are assumed to be constant for any vessel, toestimate the relative contributions of intra- and extra-vascular contrast. In the liver this assumption is untruesince normal hepatic parenchyma is supplied both by thehepatic arteries and the portal venous system. Thecharacteristics of the contrast changes in these twovascular supplies are very different. The VIF extractedfrom the hepatic arteries is similar to that seen in theaorta, although slightly delayed and of lower magnitude.The VIF observed in the portal system is significantlydelayed and shows a slow progressive rise to a plateauphase with no residual evidence of any first pass bolus.By choosing to use a breath-hold technique we havelimited the length of the dynamic acquisition so that it isnot possible to measure both the arterial and portalvenous VIFs. Furthermore, since the portal venous VIFhas lost all traces of the first-pass effect the optimizationmethod we have used would not be able to differentiate

8�� ;� �� )�� '��0�� '�'� ��'� ��'�� '�� '�� -��/ �� ;�'�� -��/ ��8'��2 <��



between intravascular and extravascular contrast con-centration changes. The results of this are that measuredvalues of kfp and rBV in hepatic tissue supplied by theportal venous system will be unreliable. Furthermore, thealgorithm will interpret contrast changes in branches ofthe portal venous system as representing contrast leakagegiving rise to falsly elevated values of kfp and artificiallylow values of rBV (Plates, 2 and 3). This problem is notinsoluble since the method we have used is a special caseof the expectation-maximization algorithm that avoidsthe complexities and inherent instabilities of nonlinearoptimization approaches.33 The use of an alternateoptimization routine should allow correct measurementof hepatic parenchymal permeability assuming thehepatic portal vein input function can be measuredwithin the data collection period. Despite these problemsthe differentiation between tissues supplied predomi-nantly by the systemic and arterial circulation and thosesupplied by the portal venous circulation is made simpleby the large time delay between the arrival of arterial andportal venous contrast. The use of the T0 parameterallows a simple binary classification of voxels and theaccuracy with which the contrast changes conforms to theLP provides adequate evidence that the arterial supply isdominant. This has enabled us to produce parametricmaps of kfp and rBV from normal abdominal tissue andfrom hepatic neoplasms.

In five cases we have been able to repeat themeasurements on two occasions, several days apart,without any therapeutic intervention. This has allowed usto calculate the reliability of the technique, which showsthat we can expect to confidently detect changes in meanvalues of kfp and rBV on the order of 15%, althoughchanges in the upper 5% of the data would need to be onthe order of 20%. Although the measurement of tumourvolume in the liver is difficult due to poor margination ofthe tumours on conventional images, the results wepresent suggest that a change of 10% can be detected withconfidence. This can be improved by the use of a semi-automated technique which utilizes the marked differ-ences in T0 between tumour and normal tissue so thatchanges of less than 5% can be confidently identified.

The results of this study demonstrate that we canproduce high-quality reproducible measurements oftumour volume, endothelial permeability and relativeblood volume from hepatic neoplasms that are suppliedby the systemic arterial system. This approach offers thebenefits of rapid image acquisition allowing breath-holdimaging and provides reproducible independent esti-mates of kfp and rBV for each voxel. However it is alsoclear that the assumptions of the analysis fail in normalliver parenchyma. Furthermore the alteration in theconfiguration of the contrast concentration profile afterpassage through the intestinal circulation suggests thatthe decompositional approach to analysis would not beapplicable in normal liver even if the portal VIF could bemeasured. The development of alternative techniques

that address these problems will be challenging. Toprovide reliable measurements of ktrans in both normalliver and hepatic tumours will require independentestimation of rBV in each voxel to provide regionalcorrection for measurements of ktrans. In addition thesource of the vascular supply for each voxel must bedetermined and the data then analysed with the appro-priate VIF. This will require separate estimation of thearterial and portal venous VIF functions, which, due tothe relative timing of contrast arrival in these twosystems, is impossible using breath-hold acquisitions.These restrictions indicate that we will need hightemporal resolution data collection that extends fromthe beginning of the arterial phase until the portal venouscontrast concentrations have reached a plateau. This willimpose major restrictions on the design of the imageacquisition sequence. Possible strategies include the useof navigator echoes to correct for the movements ofnormal respiration, segmented k-space collection to allowreconstruction of data with varying temporal resolutionand to support sampling strategies to reduce movementinduced blur, and the use of dual echo acquisitions toallow simultaneous collection of data for the estimationof rBV.

In conclusion we have demonstrated that the use ofdynamic MRI contrast enhancement combined with apharmacokinetic model of contrast distribution in the firstpass allows us to produce highly reproducible parametricmaps of kfp and rBV from hepatic tumours that aresupplied by the hepatic arterial system. The use of T0

maps allows clear identification of the source of bloodsupply to each pixel and comparison with the LPindicates whether a mixed blood supply is present. Theuse of T0 mapping also improves the reproducibility oftumour volume estimates.

)/8/)/$-/0

1. Bicknell R, Lewis C, Ferrara N. Introduction, in TumourAngiogenesis. Oxford University Press: Oxford, 1997; 15.

2. Amoroso A, et al. Vascular endothelial growth factor: a keymediator of neoangiogenesis. A review. Eur. Rev. Med. Pharmac.Sci. 1997; 1(1–3): 17–25.

3. Brasch R, et al. Assessing tumor angiogenesis using macromol-ecular MR imaging contrast media. J. Magn. Reson. Imag. 1997;7(1): 68–74.

4. Delorme S, Knopp MV. Non-invasive vascular imaging: assessingtumour vascularity. Eur. Radiol. 1998; 8: 517–527.

5. Aronen HJ, et al. Echo-planar MR cerebral blood volume mappingof gliomas. Clinical utility. Acta Radiol. 1995; 36(5): 520–528.

6. Lund EL, et al. Tumor angiogenesis—a new therapeutic target ingliomas. Acta Neurol. Scand. 1998; 97(1): 52–62.

7. Winlaw DS. Angiogenesis in the pathobiology and treatment ofvascular and malignant diseases. Ann. Thorac. Surg. 1997; 64(4):p. 1204–1211.

8. Li K, et al. Improved 3D quantitative mapping of blood volumeand endothelial permeability in brain tumors. J. Magn. Reson.Imag. 2000; 12: 347–357.

9. Zhu XP, et al. Quantification of endothelial permeability, leakagespace and blood volume in brain tumors using combined T1 and



T2* contrast-enhanced dynamic MR. J. Magn. Reson. Imag. 2000;11: 575–585.

10. Tofts PS. Modeling tracer kinetics in dynamic Gd-DTPA MRimaging. J. Magn. Reson. Imag. 1997; 7: 91–101.

11. Brix G, et al. Pharmacokinetic parameters in CNS Gd-DTPAenhanced MR imaging. J. Comp. Assist. Tomog. 1991; 15: 621–628.

12. Larsson H, et al. Quantitation of blood-brain carrier defect bymagnetic resonance imaging and gadolinium-DTPA in patientswith multiple sclerosis and brain tumors. Magn. Reson. Med. 1990;16: 117–131.

13. Mergo P, et al. MRI in focal liver disease: a comparison of smalland ultra-small superparamagnetic iron oxide as hepatic contrastagents. J. Magn. Reson. Imag. 1998; 8: 1073–1078.

14. Seneterre E, et al. Detection of hepaticmetastasis: feromoxides-enhanced MR imaging versus unenhanced MRI imaging and CTduring arterial portography. Radiology 1996; 200: 785–792.

15. Scharf J, et al. Assessment of hepatic perfusion in pigs bypharmacokinetic analysis of dynamic MR images. J. Magn. Reson.Imag. 1999; 9: 568–572.

16. Foley WD, et al. Multiphase hepatic CT with a multirow detectorCT scanner. Am. J. Roentgenol. 2000; 175(3): 679–685.

17. Thacker NA, et al. Improved quality of re-sliced MR images usingre-normalized sinc interpolation. J. Magn. Reson. Imag. 1999;10(4): 582–588.

18. Tofts PS, Kermode AG. Measurement of the blood brain barrierpermeability and leakage space using dynamic MR imaging:fundamental concepts. Magn. Reson. Med. 1991; 17: 357–367.

19. Nie NH, Hull CH. SPSS update 7–9. McGraw-Hill: New York,1981; 234–235.

20. Wastell DG, Barker GR. Intraclass correlations: A two-facet casestudy and some comments on the concept of reliability. Bull.Psychonom. Soc. 1988; 26(6): 583–586.

21. Bland JM, Altman DG. Measurement error and correlationcoefficients. Br. Med. J. 1996; 313: 41–42.

22. Tofts PS, et al. Estimating kinetic parameters from dynamiccontrast-enhanced T(1)- weighted MRI of a diffusable tracer:standardized quantities and symbols. J. Magn. Reson. Imag. 1999;10(3): 223–232.

23. Brix G, et al. Regional blood flow, capillary permeability, and

compartmental volumes: Measurement with dynamic CT—initialexperience. Radiology 1999; 210: 269–276.

24. Hawighorst H, et al. Uterine cervical carcinoma: comparison ofstandard and pharmacokinetic analysis of time-intensity curve forassessment of tumor angiogenesis and patient survival. CancerRes. 1998; 58: 3598–3602.

25. Taylor JS, et al. MR imaging of tumor microcirculation: promisefor the new millennium. J. Magn. Reson. Imag. 1999; 10(6): 903–907.

26. Diesbourg LD, et al. Quantification of myocardial blood flow andextracellular volumes using a bolus injection of Gd-DTPA: kineticmodeling in canine ischemic disease. Magn. Reson. Med. 1992;23(2): 239–253.

27. Tong CY, et al. Measurement of the extraction efficiency anddistribution volume for Gd-DTPA in normal and diseased caninemyocardium. Magn. Reson. Med. 1993; 30(3): 337–346.

28. Donahue KM, et al. Studies of Gd-DTPA relaxivity and protonexchange rates in tissue. Magn. Reson. Med. 1994; 32(1): 66–76.

29. Parker GJM, Tanner S, Leach M. Pitfalls in the measurement oftissue permeability over short time-scales using a low temporalresolution blood input function. In Proceedings of 7th ISMRM,Sidney, 1998; 1207.

30. Ellinger R, et al. The impact of the arterial input function (AIF) onquantitative evaluation of dynamic susceptibility contrast en-hanced MR studies. In Proceedings of the 6th Scientific Meeting ofthe International Society of Magnetic Resonance in Medicine,Sidney, 1997; 1207.

31. Li KL, et al. Improving estimates of endothelial permeabilitysurface area product using constrained fitting parameters for theestimation of the plasma tracer concentration function (PTCF). InProceedings of the 8th Scientific Meeting of the InternationalSociety of Magnetic Resonance in Medicine, Denver, 2000.

32. Zhu X, et al. 3D T1 mapping by means of fast field echo technique.Proceedings of the 7th Scientific Meeting of the InternationalSociety of Magnetic Resonance in Medicine, Philadelphia, 1999;2143.

33. Dempster A, Laird NM, Rubin D. Maximum likelihood fromincomplete data via the EM algorithm. J. R. Stat. Soc. 1977;B39(1): 1–38.



Documents

Breath-hold perfusion and permeability mapping of hepatic malignancies using magnetic resonance imaging and a first-pass leakage profile model