Medicine, Science & Technology

Absolute Quantification of Cerebral Blood Flow With Magnetic Resonance, Reproducibility of the Method, and Comparison With H215O Positron Emission Tomography

Description
Absolute Quantification of Cerebral Blood Flow With Magnetic Resonance, Reproducibility of the Method, and Comparison With H215O Positron Emission Tomography
Published
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Original Research  Absolute Quantification of Cerebral Blood Flow inNormal Volunteers: Correlation between Xe-133SPECT and Dynamic Susceptibility Contrast MRI Linda Knutsson, PhD, 1 *  Siv Bo¨rjesson, Eng, 2 Elna-Marie Larsson, MD, PhD, 3 Jarl Risberg, PhD, 2 Lars Gustafson, MD, PhD, 4 Ulla Passant, MD, PhD, 4 Freddy Ståhlberg, PhD, 5,6 and Ronnie Wirestam, PhD 6 Purpose:  To compare absolute cerebral blood flow (CBF)estimates obtained by dynamic susceptibility contrast MRI(DSC-MRI) and Xe-133 SPECT. Materials and Methods:  CBF was measured in 20 healthy  volunteers using DSC-MRI at 3T and Xe-133 SPECT. DSC-MRI was accomplished by gradient-echo EPI and CBF wascalculated using a time-shift-insensitive deconvolution algo-rithmandregionalarterialinputfunctions(AIFs).Toimprovethe reproducibility of AIF registration the time integral wasrescaled by use of a venous output function. In the Xe-133SPECT experiment, Xe-133 gas was inhaled over 8 minutesand CBF was calculated using a biexponential analysis. Results:  The average whole-brain CBF estimates obtained by DSC-MRI and Xe-133 SPECT were 85    23 mL/(min100 g) and 40    8 mL/(min 100 g), respectively (mean   SD,  n     20). The linear CBF relationship between the twomodalities showed a correlation coefficient of   r   0.76 and was described by the equation CBF(MRI)    2.4   CBF(Xe)  7.9 (CBF in units of mL/(min 100 g)). Conclusion:  AreasonablepositivelinearcorrelationbetweenMRI-based and SPECT-based CBF estimates was observedafter AIF time-integral correction. The use of DSC-MRI typi-cally results in overestimated absolute perfusion estimatesand the present study indicates that this trend is further enhanced by the use of high magnetic field strength (3T). Key Words:  cerebral blood flow; dynamic susceptibility contrast; magnetic resonance imaging; perfusion; Xe-133SPECT   J. Magn. Reson. Imaging 2007;26:913–920.© 2007 Wiley-Liss, Inc. REPRODUCIBLE ABSOLUTE QUANTIFICATION of ce-rebral blood flow (CBF) is desirable, for example, insequential tumor therapy monitoring, for determiningtissue at risk in acute ischemic stroke, and when a global reduction of CBF can be expected as, for exam-ple, in patients with dementia. Dynamic susceptibility contrast MRI (DSC-MRI) is a promising technique for assessment of relative CBF (1), although quantificationof CBF in absolute terms remains challenging. For ex-ample, the use of a delay-sensitive singular value de-composition (SVD) algorithm for deconvolution (2–4)mightleadtounderestimationofCBFinthepresenceof arterial delay, and a time-shift-insensitive deconvolu-tion technique is therefore recommended (5,6). CBF underestimationduetotheeffectsofarterialdispersioncan be reduced by the use of local or regional arterialinput functions (AIFs) (7,8). For absolute quantificationof CBF and cerebral blood volume (CBV), the arterialconcentration time integral is of importance, and itsaccuracy is influenced by a number of factors, such aspartial volume effects (PVEs) (9), arterial signal satura-tion at peak concentration (10), and local geometricdistortion (leading to arterial signal relocation) duringthe bolus passage (11). Finally, simulations have indi-cated that the response to a given amount of contrast agent might differ between artery and capillary (12,13).In order to assess the accuracy of DSC-MRI for abso-lute CBF quantification, the method has previously  been compared with other CBF modalities. For exam-ple, Hagen et al (14) compared DSC-MRI-based CBF (obtainedwithouttheuseofdeconvolution)withXe-CT,and Østergaard et al (15) compared DSC-MRI withpositron emission tomography (PET) assuming that thearea under the AIF was proportional to the injectedcontrast agent dose. Lin et al (16) compared DSC-MRI with PET in healthy volunteers and patients using a global scaling factor based on the ratio between the 1 Center for Medical Imaging and Physiology, MR division, Lund Univer-sity Hospital, Lund, Sweden. 2 Department of Psychiatry, Lund University Hospital, Lund, Sweden. 3 Department of Radiology, Aalborg Hospital, Aalborg, Denmark. 4 DepartmentofPsychogeriatrics,LundUniversityHospital,Lund,Swe-den. 5 Department of Radiology, Lund University Hospital, Lund, Sweden. 6 Department of Medical Radiation Physics, Lund University Hospital,Lund, Sweden.Contract grant sponsor: Swedish Research Council; Contract grant numbers:13514and621-2004-3831;Contractgrantsponsor:SwedishCancer Society; Contract grant sponsor: Crafoord Foundation; Con-tract grant sponsor: T & E Segerfalk Foundation; Contract grant spon-sor: Swedish Society of Medicine.*Address reprint requests to: L.K., Center for Medical Imaging andPhysiology, MR division, Lund University Hospital, SE-221 85 Lund,Sweden. E-mail: Linda.Knutsson@med.lu.seReceived December 17, 2006; Accepted July 3, 2007.DOI 10.1002/jmri.21093Published online in Wiley InterScience (www.interscience.wiley.com).  JOURNAL OF MAGNETIC RESONANCE IMAGING 26:913–920 (2007)© 2007 Wiley-Liss, Inc.  913  area of the venous output function (VOF) from eachpatient and the mean area of the VOFs obtained fromthe healthy volunteers. Chen et al (17) made a cross-calibration between CBF maps obtained from DSC-MRIand PET when comparing two deconvolution tech-niques.Grandinetal(18)comparedCBFaswellasCBV measured by DSC-MRI and PET, at rest and after acet-azolamide injection. Furthermore, Wirestam et al (19)compared absolute CBF from DSC-MRI and Xe-133SPECT in 10 volunteers by using a simultaneous dualfast low-angle shot (SD-FLASH) pulse sequence provid-ing a single brain-tissue slice (20).ItisreasonabletoconcludethatthepotentialofDSC-MRI for absolute CBF quantification is not yet fully established, and additional information is indeed war-ranted. For example, the majority of the previous eval-uations were carried out at 1.5T and the use of higher field strength has become increasingly common duringrecentyears.Furthermore,althoughPETisregardedasthe gold standard, comparisons with other referencemethods may serve as potentially useful supplements.Inthepresentstudy,CBFwasmeasuredinmultislicemode using Xe-133 SPECT and DSC-MRI at 3T in a group of 20 healthy volunteers and whole-brain as wellas regional CBF results from the two methods werecompared.IntheDSC-MRIanalysis,deconvolutionwas based on a time-shift-insensitive SVD algorithm withregional AIFs as input. Xe-133 SPECT allows for sepa-rate analysis of fast and slow flow components, and a pure white matter (WM) perfusion index, not influenced by gray matter (GM) flow contamination, was regardedas a potentially interesting aspect in the comparison with DSC-MRI. MATERIALS AND METHODS Subjects  CBF was measured using Xe-133 SPECT and DSC-MRIin 20 healthy volunteers. The subjects were 7 men and13 women, 43–81 years old (average age 66 years) at the time of the SPECT examination. The time interval between the Xe-133 SPECT experiment and the DSC-MRI measurement was on average 19 months (range,9–25 months), and in all cases the SPECT measure-ment was performed before the DSC-MRI experiment.Prior to the CBF measurements, each individual wasexamined by specialists in psychiatry and dementia. The medical history of each subject was reviewed usinga semistructured interview. Each subject underwent a general physical and neurological examination to ex-clude psychiatric illness, abuse of alcohol and drugs,hypertension, coronary and peripheral vascular dis-ease,systemicmetabolicalterations,braintrauma,andneurological disorders. None of the subjects used any medication that could have influenced the CBF andmetabolism. The volunteers were instructed to main-tain their usual caffeine intake on the day of the exam-ination. The project was approved by the local ethicscommittee and informed consent was obtained from all volunteers.  Xe-133 SPECT   A three-head scintillation camera (Picker Prism3000XP) was used for the Xe-133 SPECT acquisition.During the examination, Xe-133 gas (500 MBq/L in air) was inhaled over 8 minutes followed by 22 minutes of  breathing ordinary air. Ten slices were reconstructed with a nominal slice thickness of 7.1 mm and the imagematrix was 64  64. CBF was calculated using a biex-ponential analysis developed by Obrist et al (21) andmodified by Risberg et al (22). The calculation of CBF  was based on a slope index (SI) when evaluating whole- brain CBF, and this index was calculated in a similar  way as the initial slope index of the 2D Xe-133 inhala-tion method (22). The low-energy gamma rays registered in Xe-133SPECTresultsinscatteredradiationandlimitedspatialresolution (  10 mm). Hence, an accurate regional as-sessment of CBF is somewhat difficult, and it should bepointed out that the primary aim of the present study  was to establish the average absolute CBF level. As a complement, however, GM perfusion estimates wereobtainedusingtheSIvalues.WMregionsareevenmorecomplex to investigate since slow as well as fast flow components can be expected. An Xe-133 SPECT pa-rameter reflecting only the slow flow component of WM,referred to as F   WM , was calculated from the mono-expo-nentialslopeofthelast14minutesofthecurvefittedtodata observed during 30 minutes (23). To account for fast as well as slow flow components, the parameter  w     SI  (1   w)    F   WM  was also calculated for WM, where w  is the relative weight factor of the fast component (rep-resented by SI) in each pixel. DSC-MRI  Pulse Sequence and Imaging Procedure   A 3T head scanner (Magnetom Allegra, Siemens AG,Erlangen, Germany) was used for the DSC-MRI ex-amination. All volunteers received a contrast-agent dose of 0.1 mmol per kg bodyweight (Magnevist,Schering, Berlin, Germany) injected automatically at a rate of 5 mL/s in an arm vein. The first passage of the contrast agent bolus was monitored in 23 slices by use of a gradient-echo echo-planar imaging (GRE-EPI) pulse sequence with flip angle 90°, echo time 21ms, slice thickness 5 mm, and image matrix 128   128. The temporal resolution was 1.5 seconds and 60images per slice were recorded in the time series.GRE-EPI was selected in favor of spin-echo (SE) EPIin this study, since recent findings by Kjølby et al (13)indicate that SE sequences exhibit a nonlinear andtissue-dependent    R2-versus-concentration rela-tionship in tissue. Theory of CBF Calculation   When combining the central volume theorem and Zier-ler’s area-to-height relationship (24–26), CBF can beobtained using the following equation: 914 Knutsson et al.  CBF    1  H  large  R  max   0  C   t   dt   1  H  small   0  R   t   dt   0  C  art   t   dt   Theconcentrationintissue(C)andinartery(C art  )can be obtained from the relationship C(t)    k     (1/TE)   ln[S(t)/S 0 ], where S(t) is the signal at time t, S 0  is the baseline signal, and k is a constant. R(t) is the tissueresidue function, ie, the fraction of the injected tracer still present in the vasculature at time t after an as-sumed arterial input of infinitely short duration. R(t)can be obtained by deconvoluting the measured tissueconcentration time curve with an arterial input func-tion, and R  max   is the peak value of this function. H large and H small  are the hematocrit values in large and small vessels, respectively, and    is the brain density. In thepresent study, the value (1  H large )/[   (1  H small )]   0.705 cm 3 /g was employed (26).Factor analysis of dynamic studies (FADS) was em-ployed for automatic identification of a number of local AIFswithintheimagedbrainvolume(8,27).Concentra-tion–time curves that showed distorted shape at peak concentration (due to signal saturation or arterial sig-nal displacement during the bolus passage) were ex-cluded from the automatic FADS-based search for local AIFs. The FADS algorithm was additionally used toidentifyarteriesandveinsforsubsequenteliminationof large-vessel hyperintensity typically seen in GRE-EPI- based perfusion maps. Thereafter, CBF maps were cal-culated by deconvolution using the local AIF locatedgeometrically closest to a certain voxel. The deconvolu-tion was performed using a block-circulant singular  value decomposition method (5). To partially account for position-dependent PVEs, all accepted local arterialinput functions were rescaled (with retained shape) toshow the same area under the curve (28). Special care was taken to obtain a reproducible and accurate arte-rial concentration time integral, and two alternativearterial concentration time integral estimates wereevaluated as described below. The first choice was to use the largest arterial con-centrationtimeintegralofallthelocalAIFsidentifiedby the FADS algorithm in each volunteer. Since the arte-rial time integral equals the venous time integral, a second alternative was to select the superior sagittalsinus for quantification of the concentration time inte-gral. The superior sagittal sinus has a large diameter and it runs almost perpendicular to several of the im-aged slices, ie, PVEs can be minimized, and, further-more, the vessel is easily localized. First, a signal curvein the superior sagittal sinus was selected manually. Typically, such a signal curve is distorted due to localsignal relocation and/or signal saturation at peak con-centration. By selecting a small vein at another loca-tion, an approximation to the correct shape of the su-perior sagittal sinus concentration–time curve can beacquired. This curve with correct shape was then time-shifted(ifnecessary)andmultipliedbyanamplificationfactor in order to match the base and flanks of thesuperior sagittal sinus concentration–time curve withthe corresponding parts of the curve from the smaller  vein(seeFig.1).Thismatchingwasdoneautomatically:First, the two curves were interpolated to a digital tem-poral resolution of 0.3 seconds by fitting a quadraticfunction (y     a     bx     cx  2 ) to the three-point neighbor-hood (x[ i   1], x[ i  ], x[ i   1]) surrounding the interval x[ i  ]  u    x[ i   1]. Thereafter, the points that were to be in-cluded in the matching were selected, ie, the pointscorresponding to the nondistorted parts of the superior sagittal–sinus curve. The interpolated concentrationcurvefromthesmallerveinwasthenshiftedintimeandmultiplied by an amplification factor. A range of timeshifts and amplification factors were investigated, al-lowing the sum of the squares of the differences be-tween the included points to be minimized. Evaluation   The 23 CBF slices from the MRI data were recon-structed into 10 slices to match the number of CBF slices from the Xe-133 SPECT measurement. In thepresentstudy,thewhole-brainaverageCBF,calculatedfrom a comprehensive selection of regions of interest (ROIs), obtained by the DSC-MRI technique and the Xe-133 SPECT method were compared. In addition tothe global CBF comparison, a partial regional compar-ison was also carried out, ie, bilateral ROIs in thalamusGM, cortical GM, and frontal WM were evaluated sepa-rately with respect to CBF. The positioning of standardROIs in the CBF maps was accomplished by use of anin-house-developed computer program. RESULTS Whole-Brain CBF Estimates  Figure 2 shows the relationship between the average whole-brain CBF estimates obtained by DSC-MRI and Xe-133 SPECT. The DSC-MRI results from the two dif- Figure 1.  The srcinal concentration–time curves observed inthe superior sagittal sinus (bold solid line) and in a smaller  vein (solid line). The curve from the smaller vein was carefully amplified and time-shifted to match the base and flanks of thesrcinal superior sagittal sinus concentration curve (as de-scribed in the text), resulting in the corrected concentration– time curve for the superior sagittal sinus (dashed line). CBF Correlation between Xe-133 SPECT and DSC-MRI   915  ferent arterial concentration time integral estimates aredisplayed separately in Fig. 2a and Fig. 2b: Calculationof DSC-MRI-based CBF by use of the largest arterialconcentration time integral of all the local AIFs identi-fiedbytheFADSalgorithmdidnotresultinanypositivecorrelation with Xe-133 SPECT (Fig. 2a). The corre-sponding CBF results obtained by use of the superior sagittal–sinus time integral correction approach aregiven in Fig. 2b. The results of the linear regressionanalyses of the whole-brain CBF estimates (DSC-MRI vs. Xe-133 SPECT) are displayed in Table 1. All DSC-MRI-based CBF values presented below are based on the superior sagittal–sinus time integral cor-rection method. A Bland–Altman plot of the average whole-brain CBF estimates obtained by DSC-MRI (withsuperior sagittal-sinus time integral correction), and Xe-133SPECTisdisplayedinFig.3.TheXe-133SPECT measurements resulted in an average whole-brain CBF of 40    8 mL/(min 100 g) (mean    SD,  n     20), whilethe DSC-MRI measurements gave a corresponding av-erage whole-brain CBF of 85    23 mL/(min 100 g). Average whole-brain CBF as a function of the age of thesubject is displayed in Fig. 4. The observed trendlinescorresponded to a whole-brain average CBF reduc-tion of 2.6% per decade based on the SPECT data (Fig.4a) and 2.8% per decade based on the DSC-MRI data (Fig. 4b). CBF Estimates in Selected WM and GM Regions  In the selected GM regions, the average Xe-133 SPECT- based CBF was 44    10 mL/(min 100 g) while thecorresponding DSC-MRI-based CBF estimate was137  31 mL/(min 100 g). The MRI-versus-SPECT CBF relationship in GM is displayed in Fig. 5 and the resultsof the corresponding linear regression analysis aregiven in Table 2. In frontal WM the DSC-MRI experi-ment resulted in a CBF estimate of 49    11 mL/(min100 g), while Xe-133 SPECT showed 12  1.5 mL/(min100 g) using F   WM  and 22    5 mL/(min 100 g) using w     SI  (1-w)    F   WM . The corresponding relationships in WM are displayed in Fig. 6a,b and the results of the Figure 2.  Whole-brain cerebral blood flow estimates in mL/(min 100 g) obtained using Xe-133 SPECT and DSC-MRI in 20healthy volunteers.  a:  Arterial concentration time integraltaken to be the largest arterial concentration time integral of all the local AIFs identified by the FADS algorithm.  b:  Arterialconcentration time integral obtained from the corrected supe-rior sagittal sinus concentration time integral (the correctionscheme is described in the text and outlined in Fig. 1). Table 1Average Whole-Brain CBF Estimates: Results from Linear Regression Analysis of DSC-MRI versus Xe-133 SPECT DataWhole-Brain CBFLinear Equation [CBF in units of mL/(min 100g)] Linear Equation Assuming ProportionalityMaximum local AIF time integral CBF(MRI)  0.13  CBF(Xe)  73 ( r    0.069) CBF(MRI)  1.7  CBF(Xe) ( r    0.07)Corrected superior sagittal sinus timeintegral CBF(MRI)  2.4  CBF(Xe)  7.9 ( r    0.76) CBF(MRI)  2.2  CBF(Xe) ( r    0.76) Figure 3.  Bland–Altman plot showing the difference betweenthe Xe-based and MRI-based CBF estimates plotted against the average CBF from the two modalities. 916 Knutsson et al.  corresponding linear regression analyses are displayedin Table 2. When data from all the separately investigated re-gions (thalamus GM, cortical GM and WM) in all volun-teers were pooled together, the analysis showed thefollowing: The use of F   WM  for WM resulted inCBF(MRI)  2.7    CBF(Xe)  18 ( r   0.94) and, under theassumption of proportionality, CBF(MRI)    3.08   CBF(Xe) ( r     0.93). Correspondingly, the use of  w     SI  (1-w)  F   WM  for WM resulted in CBF(MRI)  3.3   CBF(Xe)-15 ( r     0.92) and, under the assumption of proportionality, CBF(MRI)    2.95    CBF(Xe) ( r     0.91).Note that the assumption of proportionality only led tosmall deteriorations of the degree of correlation.Finally, examples of corresponding color-coded CBF maps from the two modalities are shown in Fig. 7. DISCUSSION  The primary aim of this study was to compare the whole-brain CBF estimates in a population of volun-teers measured by DSC-MRI and Xe-133 SPECT. Theuse of the largest time integral among the local AIFsidentified by FADS did not result in adequate correla-tion, most likely due to substantial intersubject varia-tions in PVEs influencing the arterial curve area. Thelinear correlation between DSC-MRI and Xe-133SPECT was, however, quite satisfactory when the pro-posed concentration time integral correction was per-formed. Furthermore, the degree of correlation was ingood agreement with a previous study in which DSC-MRI using an older sequence type was compared with a previous-generation Xe-133 SPECT system (19). It isalso encouraging to note that, in the present study, thedegree of correlation was only slightly reduced whenproportionality between the two modalities was as-sumed. TheDSC-MRI-basedabsoluteCBFvaluesobtainedinthis study were indeed higher than the corresponding Xe-133 SPECT estimates and higher than previously reported values from PET studies. Considering that ef-forts were made to correct the arterial time integral(area under the curve), the CBF overestimation seen with DSC-MRI is most likely related to the fundamentalproblem with DSC-MRI in that the response to a givencontrast agent concentration differs between large ves-sels and capillaries (12,13). However, several additionalfactors may have contributed to the incorrect time in-tegral assessment for large vessels: PVEs, signal satu-ration, and signal relocation are important practicalreasons for the distortion of arterial curves at peak concentration and also for local AIFs with reasonableshape to show incorrect estimations of the concentra-tion time integral. Hence, as implied by Fig. 2a, a reli-abletimeintegralcorrectionmethodiscrucialforquan-tification or semiquantification of CBF using DSC-MRI.In this study we corrected the large-vessel area usinginformation from a smaller vein and from the superior sagittal sinus. Although the PVEs in the superior sag-ittal sinus can be assumed to have been small or mod-erate, it cannot be ruled out that the flanks of theuncorrected curve were influenced by PVEs (9), leadingto a remaining time integral error after correction. The observed overestimation of CBF measured by DSC-MRIwasnotunexpected.PreviousattemptstouseDSC-MRI for absolute quantification of CBV and CBF have normally resulted in overestimated values com-paredwithothertechniques.Thefollowingstudieswereall carried out at 1.5T: Rempp et al (26) used an SD-FLASH pulse sequence and obtained a CBF of 70 mL/(min 100 g) in GM and 34 mL/(min 100 g) in WM. Vonken et al (29) employed a segmented EPI pulse se- Figure5.  Cerebralbloodflow(inmL/(min100g))inthalamusgray matter (diamonds) and cortical gray matter (squares),obtained using Xe-133 SPECT and DSC-MRI in 20 healthy  volunteers. Figure 4.  Average whole-brain CBF versus the age of the sub- jectatthetimeoftheexaminationfor( a  )Xe-133SPECTand(  b )DSC-MRI. CBF Correlation between Xe-133 SPECT and DSC-MRI   917
Search
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks