Original Research

Feasibility Study of Exploring a T1-WeightedDynamic Contrast-Enhanced MR Approach forBrain Perfusion Imaging

Yudong Zhang, PhD,1 Jing Wang, PhD,2 Xiaoying Wang, MD,1,2* Jue Zhang, PhD,2,3*

Jing Fang, PhD,2,3 and Xuexiang Jiang, MD1

Purpose: To investigate the feasibility of T1-weighteddynamic contrast-enhanced (DCE) MRI for the measure-ment of brain perfusion.

Materials and Methods: Dynamic imaging was performedon a 3.0 Tesla (T) MR scanner by using a rapid spoiled-GRE protocol. T1 measurement with driven equilibriumsingle pulse observation of T1 (DESPOT1) was used toconvert the MR signal to tracer concentration. Cerebralperfusion maps were obtained by using an improvedgamma-variate model in 10 subjects and compared withthose with arterial spin label (ASL) approach.

Results: The cerebral blood volume (CBV) values werecalculated as 4.74 6 1.09 and 2.29 6 0.58 mL/100 g ingray matter (GM) and whiter matter (WM), respectively.Mean transit time (MTT) values were 6.15 6 0.59 s in GMand 6.96 6 0.79 s in WM. The DCE values for GM/WMcerebral blood flow (CBF) were measured as 53.41 6 9.23/ 25.78 6 8.91 mL/100 g/min, versus ASL values of49.05 6 10.81 / 23.00 6 5.89 mL/100 g/min for GW/WM. Bland-Altman plot revealed a small difference ofCBF between two approaches (mean bias ¼ 3.83 mL/100g/min, SD ¼ 11.29). There were 6 pairs of samples (5%,6/120) beyond the 95% limits of agreement. The correla-tion plots showed that the slop of Y (CBF_DCE) versus Xintercept (CBF_ASL) is 0.95 with the intercept of 4.53 mL/100 g/min (r ¼ 0.74; P < 0.05).

Conclusion: It is feasible to evaluate the cerebral perfu-sion by using T1-weighted DCE-MRI with the improved ki-netic model.

Key Words: brain perfusion; cerebral blood flow; cerebral

blood volume; dynamic contrast-enhanced MRI; meantransit timeJ. Magn. Reson. Imaging 2012;35:1322–1331.VC 2012 Wiley Periodicals, Inc.

CEREBRAL PERFUSION EVALUATION could provideimportant hemodynamic information in patients withischemia, hemadostenosis, or other intracranial dis-eases (1,2). Many efforts have been made to obtainabsolute brain perfusion (3–5). Positron emission to-mography (PET) and single photon emission com-puted tomography (SPECT) are well established tech-niques to obtain absolute cerebral blood flow (CBF),but both suffer from limitations such as potentialrisks associated with radiation exposure and prohibi-tive cost, which making longitudinal studies unlikely.In view of MR-based brain perfusion, T2*-weighteddynamic susceptibility contrast (DSC) MRI is a com-peting clinical approach for its pronounced signaldecrease produced by susceptibility effect during thefirst passage of an exogenous endovascular contrastagent (CA) through the capillary bed. Although theprominent signal change in T2*-weighted DSC-MRI isbeneficial to produce high contrast perfusion image,the low tissue signal-to-noise ratio (SNR) can make itdifficult to visually localize the feeding arteries. Eventhough many strategies have been proposed toimprove the quality of T2*-weighted perfusion imaging,the quantitation is still challenging (6–8).

T1-weighted, dynamic contrast-enhanced (DCE) MRIis not new for estimating leakage of tissue with blood–brain barrier deficiency (9). However, whether thismethod is capable of tracking CBF, in general, is notyet clearly clarified. T1-weighted MR imaging was ini-tially used to explore cerebral blood volume (CBV) byDean et al in 1992 (10) and by Hacklander et al in1996 (11), and quantitate the regional CBF by Moodyet al in 2000 (12) and by Larsson et al in 2008 (13).However, it is difficult to establish a prevalent stand-ard of T1-weighted cerebral perfusion, because themeasurements basically depend on the kinetic model,dosage of CA, field strength, scan protocol, and thecorresponding imaging parameters. Accordingly, the

1Department of Radiology, Peking University First Hospital, Beijing,China.2Academy for Advanced Interdisciplinary Studies, Peking University,Beijing, China.3College of Engineering, Peking University, Beijing, China.

Y.Z. and J.W. contributed equally to this study.

*Address reprint requests to: X.W.Department of Radiology, PekingUniversity First Hospital, No. 8, Xishiku Street, Xicheng District, Bei-jing, China, 100034. E-mail: cjr.wangxiaoying@vip.163.com or J.Z.,Academy for Advanced Interdisciplinary Studies & College of Engi-neering, Peking University, Yiheyuan Road No. 5, Beijing, China,100871 E-mail: zhangjue@pku.edu.cn

Received August 30, 2011; Accepted December 2, 2011.

DOI 10.1002/jmri.23570View this article online at wileyonlinelibrary.com.

JOURNAL OF MAGNETIC RESONANCE IMAGING 35:1322–1331 (2012)

CME

VC 2012 Wiley Periodicals, Inc. 1322

present work explored a parallel three-dimensional(3D) spoiled GRE sequence at 3.0 Tesla (T) to retrievethe temporal resolution limits, as well as combiningwith an improved gamma-variate model to determinethe characteristics of time-dependent curve (TDC)during the bolus passage of the brain capillaries. Theobtained values of cerebral perfusion were comparedwith the values from arterial spin label (ASL) MRI,which is recently reported as a noninvasive, conven-ient and reliable method referring to quantitativebrain perfusion (14,15). It hopes that this methodcould allow to obtain pixel-wise perfusion maps suchas T2*-weighted DSC-MRI, ASL, or PET imaging.

MATERIALS AND METHODS

Advances in Knowledge

The aim of this study was to overcome the temporalresolution limits of T1-weighted dynamic contrast-enhanced (DCE) by using parallel acquisition (SENSE)combined with an improved kinetic model for meas-uring regional cerebral perfusion. Also, pixel-wise cer-ebral blood flow (CBF) maps obtained by the proposedapproach demonstrate a good tissue contrast betweengray matter and white matter and no effects of theartifacts from susceptibility, which make it feasiblefor clinical application.

Implication for Patient Care

The proposed method provides a facultative way tononinvasively evaluate the cerebral perfusion forpatients with ischemia, hemadostenosis, or otherintracranial diseases.

Theory

The current kinetic model assumes that the injectedtracer in the brain tissue can only enter a single compart-ment, actually the plasma space. The first passage of anexogenous endovascular tracer through the capillary bedcan be described by a gamma-variate function that isbased on the dye-dilution theory. The function is wellknown for description of the first-pass effects of tracerkinetics and has useful analytical properties that facili-tate the measurement of characteristic curve parameters(16). We modified gamma-variate model to appreciate thefirst-pass effects of an exogenous endovascular CAthrough the capillary bed, the detail is as follows:

CðtÞ ¼ A1þe�kaðt�at�ta Þ t < tb

CðtÞ ¼ A1þe�kaðt�at�ta Þ e

�kbðt�tbÞ t � tb

(½1�

where C(t) is the measured tissue time-dependenttracer concentration, A stands for the maximumtracer concentration during an idealized instantane-ous arterial input of tracer to the tissue. ka is inflowfactor of bolus tracer, at is the arrival time of bolusinflux and ta represents the time of half tracers flow-ing into the capillary bed. kb stands for the outflowfactor of bolus tracer and tb is the time-course ofbolus passing through capillary bed.

The equation contains two fractions: (i) when thetime t is shorter than tb, the TDC is characteristicwith ‘‘S-shape’’ without venous outflow; (ii) when t isgreater than or equal to tb, the TDC can be describedas a gamma-variate function including the effect of ve-nous outflow, as showed in Figure 1. Accordingly, thetime delay between venous outflow (tb) and arterialinflux (at) is the mean transit time (MTT) of capillarybed. CBV (a unit of mL/100 g) can be estimated bythe area under curve (AUC) method showed in Eq. [2]:

CBV ¼ m

r�R10 CðtÞdtR10 CaðtÞdt

� 100 ½2�

where Ca(t) is the tracer concentration in the feedingartery, m equaling to (1-HctL)/(1-HctS) is the correctingfactor for hematocrit (Hct) in blood plasma. Here, weused a constant Hct of 0.42 for all subjects and asmall vessel Hct (HctS) to large vessel Hct (HctL) ratio

Figure 1. Schematic diagrams of the kinetic model. The leftcolumn represents a ‘‘piston model’’ for the first passage ofbolus tracers through the capillary bed. The right columnrepresents the corresponding time-dependent concentrationcurves. We hypothesized that the time difference between thebolus tracers first arriving at and first leaving the ‘‘piston’’should be the MTT. a: The time of bolus tracers first arrivingat the ‘‘piston’’ (t ¼ 0). b: The first-pass effects of tracerkinetics when t < MTT, represents a hemodynamic proce-dure of blood inflow without effect of venous outflow. Thetracer concentration in the capillary can be quantitativelymeasured using Eq. [1]. c: Diagram for the first-pass effectof tracer kinetics when t � MTT. The factor of venous outflowcan be described as an exponential function (Eq. 1). d: Thetime of bolus tracers leaving the ‘‘piston’’ (t ! 1).

Brain Perfusion Imaging 1323

of 0.85 (17) to convert the HctL to HctS. r is the den-sity of brain tissue (1.04 g/mL). Finally, CBF (a unitof mL/100 g/min) can be determined based on theknown CBV and MTT using the following Eq. [3]:

CBV ¼ CBV

MTT� 60 ½3�

Human Subjects and MR Imaging

This research was performed in accordance with theguidelines for medical research in the Helsinki Declara-tion and was approved by the Local Committee for Medi-cal Research Ethics. Between June 2010 and March2011, total 10 patients with contrast-enhancedMR exam-ination were included in this study (7 men and 3 women;mean age, 51 years; age range, 39–63 years). All of thepatients were revealed intact from the characteristics ofradiographic images together with physical findings.

The T1-weighted DCE-MRI examination was per-formed on 3.0T scanner (Signa ExciteTM; GE MedicalSystems, Milwaukee, WI) with an eight-channel phasedarray head coil. For dynamic MR imaging, a parallel 3Dspoiled GRE sequence was performed in the obliqueaxial plane by using the following parameters: 2.9/1.3ms (repetition time/echo time), a 15� excitation angle, areceiver bandwidth of 662.5 kHz, 16 slices, 7 mm slicethickness and 240 mm field of view (FOV). To improvetemporal resolution, a 128 � 128 matrix with scan per-centage of 90% and a SENSE factor of 2 in both phase-encoding (PE) directions were used. Images were inter-polated to a matrix size of 512 � 512 by zero fillingbefore reconstruction. A linear phase-encoded orderingscheme was performed. Then a bolus of paramagneticcontrast agent (Omniscan; GE Healthcare Ireland, IDABusiness Park, Carrigtohill, Co.Cork) was injected witha dose of 0.20 mL/kg (equivalent to 0.10 mM/kg) andwith the speed of 3 mL/s. A total of 60 phases wereobtained at 1.3-s intervals, with the injection occurringat the tenth phase, so that the bolus would typicallyarrive during the 5th to 20th images.

The baseline T1 images were recorded before DCE-MRI by driven equilibrium single pulse observation of T1

(DESPOT1) method (18) using following imaging param-eters: 3D-spoiled GRE sequence, echo time (TE) ¼ 1.3ms, repetition time (TR) ¼ 2.9 ms, flip angles (3� and12�), matrix size was 128 � 128 with a 7 mm slice thick-ness, 16 slices, 240mm FOV, and 62.5 kHz bandwidth.

ASL perfusion imaging was performed using a PICOREQUIPSS II pulsed ASL sequence with four-shot spiralreadout (19,20). Five axial 8 mm slices were acquired inthe caudal-to-cranial direction in 4 min. Other MR imag-ing parameters were as follows: TR 3000ms, TI1 800ms,TI2 1500 ms, TE 3.1, and 30 ms, flip angle 90�, matrix128 � 128, FOV 240 mm. Small bipolar gradients (b ¼ 2s/mm2) placed immediately before the first echo wereused to reduce signal from large vessels.

Estimate of CA Concentration

The MR signal is not linear with concentration for thecontrast-enhanced MRI. As the change of longitudinalrelaxation rate (delta R1) is proportional to the contrast

concentration both with regard to blood and tissue, onecommon method to determine Gd concentration is toconvert the MR signal intensity to R1. The time-depend-ent signal intensity of SPGR can be described as follows:

SðtÞS0

¼ 1� e�TR�R1ðtÞ

1� e�TR�R1ð0Þ �1� cosðaÞe�TR�R1ð0Þ

1� cosðaÞe�TR�R1ðtÞ ½4�

where S(t) is the time-dependent signal intensity aftera bolus injection of Gd-DTPA, S0 is the baseline signalintensity of pre injection, R1(t) is the correspondingtime-dependent R1, R1(0) is the baseline R1 and a isflip angle. With the known R1(0), R1(t) is calculated.Accordingly, the CA concentration can be obtainedfrom R1 change, as described in Eq. [5]:

CðtÞ ¼ DR1ðtÞr1

¼ R1ð1Þ � R1ð0Þr1

½5�

where the relaxivity r1 of Gd-DTPA at 3.0T wasendued with 4.0 s�1.mM�1, a value provided by themanufacturer.

Image Postprocessing

Pixel-to-pixel analysis of DCE data using a least-squaremethod was performed on Matlab (MathWorks, Natick,Mass). In term of input function, we manually placed aregion of interest (ROI) in a relatively large vein (e.g., thesuperior sagittal sinus) rather than in the feedingarteries. The selected vein was oriented perpendicularlyto the imaging plane, and the size of ROI was smallenough (proximate 10 mm2 area) to ensure that the vas-cular curves are not affected by partial volume effects(PVE). Then the TDC of target ROI was denoted as ve-nous-output-function (VOF). Pixels with non-physiolog-ically high perfusion of more than 120 mL/100 g/minwere excluded from the evaluation. In the case of tissueROIs, on one of the most cranial slices, two pairs ofROIs were placed in bilateral frontal gray matter (GM)/white matter (WM), two pairs in bilateral temporal GMand basal ganglia (BG), and two pairs in bilateral GM/WM, respectively. The size of the ROI for both GM andWM was 150–600 mm2 area. The ROIs were drawn onthe corresponding anatomical images (Fig. 2).

With regard to ASL images, CBF maps wereobtained by subtraction of the labeled from the con-trol images and quantified using the perfusion modeldescribed by Wong (20). The placement of tissue ROIswas similar to the DCE method mentioned above.

Statistical Analysis

Results were reported as mean 6 standard deviation.To verify the concordance of the proposed model withASL analysis, Bland-Altman plot was created usingthe average of CBF measured from both methodsas the x-axis, while the difference between them asy-axis. Mean difference and 95% limits of agreementare reported. The simple Pearson correlation coeffi-cient and the corresponding P value were calculatedto analyze the significance of DCE and ASL for theperfusion measurements in different patients.

1324 Zhang et al.

RESULTS

The characteristics of cerebral hemodynamics werequantitatively estimated in all patients. Figure 3shows the representative timeframes of an axial slicethrough the basal ganglia, from which we can serially

identify the preenhanced image in 8th frame, the earlyarterial phase in 16th frame, the peak arterial phasein 25th frame, the venous phase in 30th frame andsteady-state in 50th frame. The VOF is obtained inGalen vein (arrow). Figure 4 shows the TDC corre-sponding to an ROI placed in Galen vein as VOFduring the bolus passage, and the MR signals is con-verted to CA concentrations. Figure 5 shows a tissueTDC for an ROI localized in the left cerebral hemi-sphere. It illustrates that the model fit (red solidcurve) is well accordant with tissue data points (blackdotted line).

Table 1 summarizes the mean values of regionalCBF, CBV and MTT measured in 10 patients usingthe proposed methods, corresponding CBF measuredby ASL method is also reported. Correlation betweenthe new approach and ASL was significant (P < 0.05)in individual patients, with the Pearson correlationcoefficient of 0.74 in CBF. Bland-Altman plot (Fig. 6)shows that the mean bias of CBF measured with twomethods is 3.83 mL/100 g/min with a standard devi-ation of 11.29 mL/100 g/min. the 95% confidenceinterval (CI) of the bias ranges from �18.29 to 25.96mL/100 g/min. There are six pairs of samples (5%,6/120) beyond the 95% limits of agreement that illus-trates high concordance between CBF obtained byDCE and ASL. The correlation plots with Passing &Bablok regression analysis show that the slop of Y(CBF_DCE) versus X intercept (CBF_ASL) is 0.95 withthe intercept of 4.53 mL/100 g/min.

Typical kinetic parameter maps with color scale areshown in Figure 7 for one patient without positive in-tracranial findings. It illustrates that the modifiedgamma-variate model with a stepwise fitting strategy

Figure 2. The illustration of ROIs location in cerebral paren-chyma. R, right; L, left; FG, frontal gray matter; FW, frontalwhite matter; TG, temporal gray matter; TW, temporal whitematter; OG, occipital gray matter; OW, occipital white mat-ter; BG, basal ganglia.

Figure 3. Dynamical imagesfrom a slice through the basalganglia. a: Image obtainedbefore contrast arrival. b:Image obtained in the earlyarterial phase, where a brightsignal is seen in anterior andmiddle cerebral artery. c:Image obtained in the peak ar-terial phase, where the cere-bral capillaries are greatlyenhanced and the Galen vein(arrow) used as VOF can beidentified. d: Image obtainedin the venous phase. e: Thesituation of 60 s after the ar-rival of contrast in the largerveins (stead-state). f: Corre-sponding T1 map obtainedbefore the contrast injection.

Brain Perfusion Imaging 1325

allows a quantitative estimate of five kinetic parame-ters: parameter A, an idealized maximum concentra-tion of CA distributed in capillary; inflow factor ka andcorresponding time factor ta; outflow factor kb and timefactor tb. The time-course of bolus agents passingthrough capillary bed, namely MTT is also obtained.CBV map estimated by the ‘‘area under curve’’ methodreveals an excellent tissue contrast and spatial resolu-tion. The comparison of perfusion maps using modifiedgamma-variate model and ASL technique is showed inFigure 8, in which, CBF maps with DCE-MRI (first row)give a good tissue contrast between gray matter andwhite matter, and are free of the artifact from suscepti-bility effect, but reveal only minor artifacts from noisein the white matters. The ASL images (second row) alsoshow an excellent tissue contrast between gray matterand white matter, but display a weak spatial resolutionand signal loss in some regions of superior section(arrow), probably caused by the effect of transit delayor susceptibility artifacts.

DISCUSSION

The evaluation of cerebral perfusion is extremely use-ful in clinical patients with cerebrovascular disease.

However, a quantitative measurement of CBF stillremains challengeable (6–8,21). The documented datausing H15

2 O-PET showed that normal CBV is approxi-mately 2–5% and CBF is typically estimated in therange of 25–60 mL/100 g/min with a standard devia-tion of 10–30% (22,23). Furthermore, the value ofCBF may vary in different kinetic models, individuals,and imaging techniques, so the gold standard is diffi-cult to be established. Although it is not entirely newfor the purpose of T2-weighted DSC-MR perfusionmeasurement during a bolus of tracer first passingthrough the brain capillaries, the present studyrevealed a feasibility of T1-weighted DCE-MRI cerebralperfusion imaging in 10 clinical patients. Theobtained perfusion values were highly consistent withthose calculated by ASL. In addition, the proposedmethod was capable of creating pixel-wise perfusionwith prominent tissue contrast and spatial resolutionsuch as those seen in T2*-weighted DSC-MRI or ASLimaging.

Generally, the commonly used single-compartmentmodel with T1-weighted DCE-MRI for brain perfusionimaging may encounter some difficulties: (i) the con-version for MR signal to CA concentration; (ii) thepoor signal contrast in T1-weighted image, especiallyin low field strength, may prevent the production of

Figure 4. Result of time-dependent MR signals converted as concentration in one ROI placed in Galen vein as VOF. a:Enlarged image to facilitate the placement of ROI in the target vein (arrow). b: Time-dependent signal curve of the ROI withproximate 10 mm2 area. c: The converted time-dependent concentration curve.

Figure 5. a: The presented model was fitted (red solid curve) to the experimental tissue data (black dotted line). b: The fittingcurve without impact of venous outflow (t < tb). c: The situation of 7.89 s (the obtained MTT) after the arrival of contrast inthe capillary bed.

1326 Zhang et al.

high-contrast perfusion image; (iii) the limitation oftemporal resolution; (iv) effect of water exchangebetween intra- and extravascular space; (v) the opti-mization of kinetic model. In view of these challenges,we have some improvements in the present study:

Conversion for MR Signal to Concentration

T1-weighted MR signal is approximately linear withGd concentration over a clinically relevant range (24).However, this correlation may result in incorrect con-centration estimates, especially in regions with highcontrast concentration (e.g., the input arteries or kid-neys). In this study, we used a generally approvedapproach that uses known sequence parameters andan analytical calibration relationship between signaland T1 value. Although its validity is largely depend-ent on the properties of hardware, pulse sequence,effective flip angle, and in vivo environment, the pre-sented method relied on two assumptions: (i) thehardware during dynamic imaging is held at a steady-state meaning that the equilibrium longitudinal mag-netization (M0), effective flip angle, and gain factor arenot changed at each timeframe; (ii) the T2/T2* effects is

neglected. The first assumption is approximately truewhen an uninterrupted, dynamic imaging was per-formed. To our knowledge, T2/T2* effect is only domi-nant in the situation of extremely high concentration(>10 mM). Thus, we assumed that T2/T2* effect issmall enough to be neglected under the clinicaldosage.

MR Protocols

One limitation of T1-weighted brain perfusion imagingis its temporal resolution. The calculated perfusionvalues, including CBF, CBV, and MTT, largely dependon the employed temporal resolution (25). Althoughcentral k-space acquisition scheme is a pronouncedapproach to increase temporal resolution, it has somelimits: (i) greatly sacrifice information of spatial reso-lution relevant to the loss of phase-encoding informa-tion in the edge of k-space, (ii) unable to establish areliable mathematical relation between tracer concen-tration and signal intensity. In recent years, parallelacquisition techniques, which involve the use of thespatial information contained in the sensitivity pro-files of multiple elements of a receive coil (26), have

Table 1

Perfusion results measured with proposed method and ASL (Mean 6 SD)

ROI (N¼20�6) CBV (mL/100 g) MTT (s) CBF-New (mL/100 g/min) CBF-ASL (mL/100 g/min)

Frontal GM 4.51 6 0.79 5.94 6 0.62 54.51 6 5.79 51.14 6 12.66

Frontal WM 2.21 6 0.72 7.35 6 0.91 23.68 6 11.86 25.64 6 6.00

Temporal GM 5.45 6 1.37 6.24 6 0.59 61.63 6 10.24 49.59 6 11.93

Occipital GM 5.19 6 0.79 6.52 6 0.41 53.00 6 6.00 48.18 6 9.99

Occipital WM 2.37 6 0.39 6.57 6 0.39 27.88 6 3.63 20.36 6 4.53

Basal ganglia 3.81 6 0.32 5.92 6 0.54 44.50 6 4.89 47.29 6 8.63

Total 3.92 6 1.49 6.42 6 0.76 44.20 6 15.93 40.37 6 15.52

GM ¼ gray matter; WM ¼ white matter.

Figure 6. a: Bland-Altman analysis of the difference in 120 pairs of CBF values obtained by DCE and ASL. Horizontal andvertical axes, respectively, represent the average and difference of CBF obtained by DCE and ASL. Dotted lines represent themean difference and the 95% limits of agreement. Regional CBF in BG are marked as blank pane, FG as black circle, FW asdecussation, OG as blank circle, OW as black pane, and TG as small blank pane. Note that Bland-Altman analysis revealedexcellent accordance between the two methods, with only six-pair samples (5%, 6/120) beyond the 95% limits of agreement.b: Passing and Bablok regression shows a relatively good linear correlation between the regional CBF measured by DCE andASL (r ¼ 0.74; P < 0.05).

Brain Perfusion Imaging 1327

been widely used in dynamic contrast-enhanced liverMR imaging to benefit patients with limited breath-hold capability. The k-space is systematically under-sampled in parallel imaging while data are acquiredwith all coil elements in parallel. This enables areduction in the spatial-encoding steps and thus ashorter MR image acquisition time, while preservingthe spatial resolution of the MR image. In this study,

we explored a parallel 3D spoiled GRE sequence byusing a SENSE factor of 2 in both PE directions,which leaded to a fourfold decrease in image acquisi-tion time. Hence, a scan interval of 1.3 s based onparallel imaging is fast enough for quantitation of theAIF or VOF, which was used to normalize MR-basedcerebral perfusion. The increased time samples arehelpful to the gamma-variate model fitting and good

Figure 7. Pixel-wise kinetic parameters calculated by least-square method.

Figure 8. A series of CBF maps measured by modified gamma-variate model (the first row) plotted against ASL method (thesecond row). CBF maps with DCE-MRI demonstrate a good tissue contrast between gray matter and white matter, and arefree of artifacts from susceptibility effect, with only minor artifacts from noise in the white matters. Despite the low resolutionof the ASL images, CBF maps obtained by ASL also show an excellent tissue contrast between gray matter and white matter.However, the signal loss was found in some regions of superior section (arrow), which may be probably caused by the effectof transit delay or susceptibility artifacts.

1328 Zhang et al.

for the improvement of perfusion maps. Moreover, thebaseline T1 images were recorded using rapid 3Dspoiled GRE with two optimal flip angles, which is ofbenefit to shorten the acquisition time (approximately30 s), making the clinical application viable.

CBV and the Effect of Water Exchange

Currently there are two approved approaches to quan-titate CBV with MRI: T2*-weighted DSC by using thefirst passage of a bolus of contrast agent, and thesteady-state (SS) method. DSC approach producesrelative values of CBV rather than quantitative valuesdue to the PVE on AIF (27). Although the SS approachcan provide quantitative CBV values, it is sensitive tothe water change between the compartments (28). Interm of the presented CBV measurement, we, first,measured the VOF in relatively large vein to avoidPVE, and then used an AUC method with using thefirst passage of a contrast agent bolus instead of asingle static approach to obtain quantitative CBV val-ues. The obtained CBV values are 4.74 6 1.09 mL/100 g in gray matter and 2.29 6 0.58 mL/100 g inwhite matter, which are approximately accordant withthat reported by Kuppusamy et al with SS approach(29), by shin with a corrected SS approach (28), andby Greenberg et al with PET (30), but slightly lowerthan that reported by Rempp et al with DSC analysis(31). We supposed that the difference can be attrib-uted to the placement of ROI, age of subjects, MR pro-tocol and used methodology. With regard to the effectof water exchange, it was a large source of variabilityin the use of T1-weighted SS approach, which is heav-ily dependent on the approximation of exchange rate,tracer concentration and tissue type. In contrast tosingle measurement model, AUC method used in ourstudy might provide some advantages in reducingCBV bias by averaging the measured R1 error in eachtimeframe. This may be an interesting focus for futurestudy.

Kinetic Model

In traditional ‘‘maximum gradient’’ method to deter-mine blood perfusion, venous outflow is assumed tobe neglected. Thus, the Fick’ law is simplified andblood flow can be approximately calculated as the ra-tio of the maximum initial slope of the tissue time-concentration curve to the peak height of the AIF. Oneadvantage of this method is the conceptual simplicity.However, the assumption of no venous outflow at thetime of the maximum initial slope of the tissue time-concentration curve is not always true. When CAbegins to flow out of the tissue capillaries before themaximum initial slope of the tissue TDC is arrived,the underestimation of CBF will occur. Thus, the doseof CA has to be enough to overcome the effect of ve-nous outflow. Another approach for perfusion imagingis dependent on the concept of convolution. Becausewith a peripheral intravenous injection, AIF is not adelta function (contrast is not injected directly intothe arterial inlets of the brain). The ideal tissue time-concentration cannot be measured directly under this

situation. To get around this problem, the measuredtissue TDC has to be deconvolved by the AIF, which isthe basic idea of deconvolution analysis. Although thepronounced properties referring to accuracy, reprodu-cibility and vascular structure-independent, theestablished deconvolution methods, such as thestandard singular value decomposition (SVD) or fastFourier Transform (FFT), may be influenced by thesignal noise (32), tracer arrival delay (33) or PVE (27).

The gamma-variate model was derived from thedye-dilution theory, which initially introduced byThompson et al in 1964 (34). The modified model inthe subsequent study was commonly used for the ul-trasonic assessment of parenchymal cerebral echocontrast enhancement (16). Although it is of capabilityto appreciate the first-pass effects of tracer kineticsand has useful analytical properties that facilitate themeasurement of characteristic curve parameters, thegamma-variate model is a semi-quantitative method.It is known that, in the networking capillary tubes,the bolus tracer delivered from input artery to outputvein is a time-dependent process that may be variedby the factors of histological structure, blood vesseldiameter, vessel length or distal vascular resistance.This actual hemodynamic process is simplified in thepreviously established gamma-variate model, butmakes the quantification of TDC unviable. In thepresent work, this deficiency is well resolved by usinga segmented kinetic model. The proposed modelassumes that the first passage of an exogenous endo-vascular tracer through the capillary bed will experi-ence two segmented hemodynamic processes: thebolus influx without venous outflow and bolus influxwith venous outflow. In the early kinetic phase,because of the transit delay of bolus tracer betweeninput artery and output vein, we assumed that thereis no outflow of tracers which arrived at capillary bed,suggesting the TDC in this time course can beexpressed by a single gamma function, as shown inthe first part of Eq. [1]. Generally, the duration of thiskinetic process is a few seconds in brain blood capil-laries. In the following kinetic process, although thereis still a bolus influx in regional capillaries, a concom-itant outflow happens. This kinetic feature is deter-mined by using an extended gamma-variate function,as shown in the second part of Eq. [1]. Therefore,according to this segmented theory, the kinetic modelproposed here is able to appreciate the true hemody-namic features of the first passage of bolus tracerthrough the cerebral capillary bed, where, the param-eters ka and kb are explored to determine bolus influxand venous outflow, parameter at is used to deter-mine the bolus arrival at the capillary bed and tb isgiven to calculate the time of venous outflow. Actually,the idea of segmented kinetic model is not first pre-sented in this study, as early as 1998, a time-domain,closed-form solution of the tissue homogeneity modelwas proposed to measure the water exchange in aleaked kinetic model by St Lawrence and Lee (35),where a similar parameter TC was introduced todetermine the transit time through the capillary. Thesimilar method was used to determine impulse reten-tion function in a renal perfusion model by Lee et al

Brain Perfusion Imaging 1329

in 2008 (36). Unlike the traditionally deconvolutionmethods, the proposed model in our study needs notto deconvolve the pixel-wise tissue TDC with AIF,which makes it unnecessary to correct the time delaybetween the two curves. Moreover, we can place theROI in a relatively lager vessel, e.g., sinus venosus, tofacilitate the measurement of input function.

Although an improved gamma-variate curve fittingwas presented in this study, the results may beinevitably influenced by the shape of the first-passbolus peak especially in situations of differentamount of tracer injected and cardiac output (37).Also it is reported that the model-dependentapproaches may lead to more systematic errors withregard to different residue function (32). Thus, thereproducibility of the proposed model still needs fur-ther verification.

In conclusion, by using a rapid 3D spoiled GREsequence at 3.0T in combination with segmented ki-netic model, the hemodynamic features of bolus tracerthrough the brain capillaries were well determinedand prominent perfusion images were obtained. Thehigh consistence of CBF values between the newapproach and ASL method demonstrates the feasibil-ity of T1-weighted, DCE MRI based on the proposedmodel for brain perfusion estimation.

REFERENCES

1. Sasao A, Hirai T, Nishimura S, et al. Assessment of vascular sup-ply of hypervascular extra-axial brain tumors with 3T MR re-gional perfusion imaging. AJNR Am J Neuroradiol 2010;31:554–558.

2. Schaefer PW, Barak ER, Kamalian S, et al. Quantitative assess-ment of core/penumbra mismatch in acute stroke: CT and MRperfusion imaging are strongly correlated when sufficient brainvolume is imaged. Stroke 2008;39:2986–2992.

3. Iida H, Law I, Pakkenberg B, et al. Quantitation of regional cere-bral blood flow corrected for partial volume effect using O-15water and PET: I. Theory, error analysis, and stereologic compari-son. J Cereb Blood Flow Metab 2000;20:1237–1251.

4. Soustiel JF, Glenn TC, Vespa P, Rinsky B, Hanuscin C, MartinNA. Assessment of cerebral blood flow by means of blood-flow-volume measurement in the internal carotid artery - comparativestudy with a (133)xenon clearance technique. Stroke 2003;34:1876–1880.

5. Petrella JR, Provenzale JM. MR perfusion imaging of the brain:techniques and applications. AJR Am J Roentgenol 2000;175:207–219.

6. Salluzzi M, Frayne R, Smith MR. An alternative viewpoint of thesimilarities and differences of SVD and FT deconvolution algo-rithms used for quantitative MR perfusion studies. Magn ResonImaging 2005;23:481–492.

7. Zaharchuk G, Bammer R, Straka M, et al. Improving dynamicsusceptibility contrast MRI measurement of quantitative cerebralblood flow using corrections for partial volume and nonlinearcontrast relaxivity: a xenon computed tomographic comparativestudy. J Magn Reson Imaging 2009;30:743–752.

8. Carroll TJ, Rowley HA, Haughton VM. Automatic calculation ofthe arterial input function for cerebral perfusion imaging withMR imaging. Radiology 2003;227:593–600.

9. Tofts PS. Modeling tracer kinetics in dynamic Gd-DTPA MR imag-ing. J Magn Reson Imaging 1997;7:91–101.

10. Dean BL, Lee C, Kirsch JE, Runge VM, Dempsey RM, PettigrewLC. Cerebral hemodynamics and cerebral blood volume:MR assessment using gadolinium contrast agents and T1-weighted Turbo-FLASH imaging. AJNR Am J Neuroradiol 1992;13:39–48.

11. Hacklander T, Reichenbach JR, Hofer M, Modder U. Measure-ment of cerebral blood volume via the relaxing effect of low-dose

gadopentetate dimeglumine during bolus transit. AJNR Am JNeuroradiol 1996;17:821–830.

12. Moody AR, Martel A, Kenton A, et al. Contrast-reduced imagingof tissue concentration and arterial level (CRITICAL) for assess-ment of cerebral hemodynamics in acute stroke by magnetic res-onance. Invest Radiol 2000;35:401–411.

13. Larsson HB, Hansen AE, Berg HK, Rostrup E, Haraldseth O.Dynamic contrast-enhanced quantitative perfusion measurementof the brain using T-1-weighted MRI at 3T. J Magn Reson Imag-ing 2008;27:754–762.

14. Ye FQ, Berman KF, Ellmore T, et al. H(2)(15)O PET validation ofsteady-state arterial spin tagging cerebral blood flow measure-ments in humans. Magn Reson Med 2000;44:450–456.

15. Feng CM, Narayana S, Lancaster JL, et al. CBF changesduring brain activation: fMRI vs. PET. Neuroimage 2004;22:443–446.

16. Postert T, Hoppe P, Federlein J, et al. Contrast agent specificimaging modes for the ultrasonic assessment of parenchymal cer-ebral echo contrast enhancement. J Cereb Blood Flow Metab2000;20:1709–1716.

17. Eichling JO, Raichle ME, Grubb RL Jr, Larson KB, Ter-PogossianMM. In vivo determination of cerebral blood volume with radioac-tive oxygen-15 in the monkey. Circ Res 1975;37:707–714.

18. Deoni SC, Rutt BK, Peters TM. Rapid combined T-1 and T-2mapping using gradient recalled acquisition in the steady state.Magn Reson Med 2003;49:515–526.

19. Wong EC, Buxton RB, Frank LR. Implementation of quantitativeperfusion imaging techniques for functional brain mapping usingpulsed arterial spin labeling. NMR Biomed 1997;10:237–249.

20. Wong EC, Buxton RB, Frank LR. Quantitative imaging of perfu-sion using a single subtraction (QUIPSS and QUIPSS II). MagnReson Med 1998;39:702–708.

21. Kudo K, Sasaki M, Ogasawara K, Terae S, Ehara S, Shirato H.Difference in tracer delay-induced effect among deconvolutionalgorithms in CT perfusion analysis: quantitative evaluation withdigital phantoms. Radiology 2009;251:241–249.

22. Tropres I, Grimault S, Vaeth A, et al. Vessel size imaging. MagnReson Med 2001;45:397–408.

23. Rostrup E, Law I, Pott F, Ide K, Knudsen GM. Cerebral hemody-namics measured with simultaneous PET and near-infraredspectroscopy in humans. Brain Res 2002;954:183–193.

24. Taylor J, Summers PE, Keevil SF, et al. Magnetic resonancerenography: optimisation of pulse sequence parameters and Gd-DTPA dose, and comparison with radionuclide renography. MagnReson Imaging 1997;15:637–649.

25. Kamena A, Streitparth F, Grieser C, et al. Dynamic perfusion CToptimizing the temporal resolution for the calculation of perfusionCT parameters in stroke patients. Eur J Radiol 2007;64:111–118.

26. McKenzie CA, Lim D, Ransil BJ, et al. Shortening MR image ac-quisition time for volumetric interpolated breath-hold examina-tion with a recently developed parallel imaging reconstructiontechnique: clinical feasibility. Radiology 2004;230:589–594.

27. van Osch MJ, Vonken EJ, Bakker CJ, Viergever MA. Correctingpartial volume artifacts of the arterial input function in quantita-tive cerebral perfusion MRI. Magn Reson Med 2001;45:477–485.

28. Shin WY, Cashen TA, Horowitz SW, Sawlani R, Carroll TJ. Quan-titative CBV measurement from static T-1 changes in tissue andcorrection for intravascular water exchange. Magn Reson Med2006;56:138–145.

29. Kuppusamy K, Lin WL, Cizek GR, Haacke EM. In vivo regionalcerebral blood volume: quantitative assessment with 3D T1-weighted pre- and postcontrast MR imaging. Radiology 1996;201:106–112.

30. Greenberg JH, Alavi A, Reivich M, Kuhl D, Uzzell B. Local cere-bral blood-volume response to carbon-dioxide in man. Circ Res1978;43:324–331.

31. Rempp KA, Brix G, Wenz F, Becker CR, Guckel F, Lorenz WJ.Quantification of regional cerebral blood flow and volume withdynamic susceptibility contrast-enhanced MR imaging. Radiology1994;193:637–641.

32. Ostergaard L, Weisskoff RM, Chesler DA, Gyldensted C, RosenBR. High resolution measurement of cerebral blood flow usingintravascular tracer bolus passages: Part I. Mathematicalapproach and statistical analysis. Magn Reson Med 1996;36:715–725.

1330 Zhang et al.

33. Willats L, Connelly A, Calamante F. Improved deconvolution ofperfusion MRI data in the presence of bolus delay and dispersion.Magn Reson Med 2006;56:146–156.

34. Thompson HK, Whalen RE, Starmer CF, Mcintosh HD. Indicatortransit time considered as gamma variate. Circ Res 1964;14:502–515.

35. St Lawrence KS, Lee TY. An adiabatic approximation to the tissuehomogeneity model for water exchange in the brain: I. Theoreticalderivation. J Cereb Blood Flow Metab 1998;18:1365–1377.

36. Lee VS, Zhang JL, Rusinek H, et al. Functional assessment of thekidney from magnetic resonance and computed tomographyrenography: impulse retention approach to a multicompartmentmodel. Magn Reson Med 2008;59:278–288.

37. Weisskoff RM, Chesler D, Boxerman JL, Rosen BR. Pitfalls inMR measurement of tissue blood flow with intravascular trac-ers: which mean transit time? Magn Reson Med 1993;29:553–558.

Brain Perfusion Imaging 1331