| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Developments |
1 From the Departments of Radiology (S.M.K., J.D.E.) and Community and Family Medicine (D.M.D.), Duke University Medical Center, Box 3808, Durham, NC 27710; and School of Medicine, University of Pennsylvania, Philadelphia (V.A.L.). Received March 27, 2003; revision requested June 18; final revision received September 16; accepted October 8. Address correspondence to S.M.K. (e-mail: susankealey@hotmail.com).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
© RSNA, 2004
Index terms: Brain, perfusion • Computed tomography (CT), perfusion study, 14.12112
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
While the feasibility and promise of dynamic CT perfusion imaging are evident for the assessment of patients with stroke, there are technical and clinical issues that remain to be answered with regard to the use of CT perfusion imaging in assessment of patients with stroke. One technical issue relates to the variability in maps that can be seen as a result of placement of user-defined arterial and venous input function regions of interest (ROIs). Deconvolution analysis is an important and commonly used method for studying CT perfusion data. At present, however, such analysis requires the user to visually select and manually place small ROIs onto a reference image to provide the algorithm with information needed to compute tissue perfusion parameter values (5). Dependence on user-defined input function ROIs for accuracy seems a likely source for variability during analysis. In our experience, small differences in the ROI location within a given artery or vein can dramatically change the resultant time-attenuation curve that the algorithm uses to compute the perfusion maps. Additionally, we have observed that perfusion map quality frequently varies from patient to patient without readily apparent cause. We suspected that user-defined input function ROI properties could be a factor and hypothesized that volume averaging effects caused by small differences in ROI location have a significant influence on absolute perfusion values and map quality as measured by means of the signal-to-noise ratio (SNR). Thus, we undertook this study to determine the influence of arterial and venous input function properties on perfusion parameter values and tissue SNR.
|
Materials and Methods |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
CT Perfusion Scanning
First, a transverse imaging plane was selected at the level of the basal ganglia by one neuroradiologist (J.D.E.) (Fig 1). The level of the basal ganglia was chosen to allow maximum coverage of all major vascular distributions on one image. Continuous cine acquisition was performed for 45 seconds in a single location during intravenous infusion of nonionic iodinated contrast material (40 mL of contrast material at 4 mL/sec for 10 seconds). In each case, CT scanning began 5 seconds after the start of infusion. Gantry rotation speed was one rotation per second, and the cine data were reconstructed retrospectively at half-second intervals prior to analysis.
|
|
|
The superior sagittal sinus was chosen as the venous outflow function in 36 patients. In four cases, this vessel was not visible at the level scanned, so another large vein such as the transverse sinus or straight sinus was selected instead.
The characteristic features of time-attenuation curves are described in Figure 3. The time-attenuation curves corresponding to both the arterial input function and the venous outflow function were first inspected visually to measure a number of different properties. First, since cardiac function is thought to be one variable that could influence blood flow map quality, the duration of the baseline interval (in seconds) before arrival of the contrast bolus in the artery and vein was measured to provide an indirect index relating primarily to patient cardiac function (Fig 3a). Second, the height, or peak enhancement (in Hounsfield units), of the arterial and venous curves above their mean baseline attenuation values was measured by identifying the peak attenuation value of each curve and subtracting the mean attenuation value of the ROI during the baseline interval (Fig 3b). Finally, the width of each arterial and venous time-attenuation curve (in seconds) was measured at the attenuation value equal to half the maximum attenuation value of the curve (Fig 3c).
|
|
|
|
Analysis of Variability within Subjects
A subgroup of 10 subjects chosen randomly from the larger group was analyzed to evaluate the effect of varying the location of the venous input ROI. We chose 10 subjects to have a stable estimation of variance and to allow some averaging to justify the use of a t statistic. For each of these 10 patients, CT perfusion maps (CBV, CBF, and MTT) were created twice by one author (either J.D.E. or S.M.K.) operating independently: (a) once with the venous input function ROI located centrally within the superior sagittal sinus and (b) a second time with the venous input function ROI located nearby but more peripherally within the superior sagittal sinus to intentionally create a volume averaging effect. Arterial ROI placement was kept constant. Venous peak enhancement values and baseline levels were determined as described earlier and recorded. Tissue ROIs were placed over gray matter and white matter as discussed earlier, and the mean perfusion parameter values and SDs were measured and recorded.
Statistical Methods
For the correlations performed between subjects, Pearson and Kendall correlation coefficients were computed, and corresponding P values were calculated. Owing to the large numbers of analyses performed in this part of the study, statistical significance was declared only for P values equal to or less than .01. Correlation of subject age with mean perfusion parameter values was performed by using a Pearson correlation coefficient with regression analysis. A Student t test was used to determine any differences between male and female subjects. For intrasubject comparisons with different venous input function curves, the absolute values of CBV, CBF, MTT, and perfusion map SNR for each venous ROI location were compared by using a two-tailed paired Student t test. Statistical significance was declared at the .05 level. The statistical software used was SAS (SAS, Cary, NC).
|
Results |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
The mean MTT was 5.2 seconds ± 1.4 (range, 2.5–8.5 seconds) in frontal lobe white matter and 2.9 seconds ± 0.8 (range, 1.8–5.1 seconds) in the putamen. The mean CBV was 1.5 mL per 100 g ± 0.9 (range, 0.6–5.5 mL per 100 g) in frontal lobe white matter and 3.1 mL per 100 g ± 1.8 (range, 1.8–11.6 mL per 100 g) in the putamen. The mean CBF was 22.1 mL per 100 g per minute ± 10.6 (range, 7.4–56.5 mL per 100 g per minute) in the frontal lobe white matter and 65.1 mL per 100 g per minute ± 22.6 (range, 30.2–129.5 mL per 100 g per minute) in the putamen.
Mean SNRCBV was 0.07 ± 0.03 in white matter and 2.03 ± 0.74 in gray matter. Mean SNRCBF was 1.05 ± 0.38 in white matter and 2.69 ± 5.55 in gray matter. Mean SNRMTT was 1.18 ± 0.39 in white matter and 1.71 ± 0.92 in gray matter. The relationship of arterial and venous peak enhancement to perfusion map parameters is shown in Tables 1 and 2.
|
|
Intersubject Analysis
Correlation of venous curve parameters with perfusion map parameters.—The venous peak enhancement value correlated significantly with the mean CBF value in the white matter location (r = –0.51; P < .001) and for the mean CBV value in both the white matter location (r = –0.46; P < .003) and the gray matter location (r = –0.44; P < .005). Correlation with the venous peak enhancement value that approached significance was found for the mean CBF value in the gray matter location (r = –0.32; P = .047). No other correlations of venous curve parameters with perfusion parameter values were significant.
Significant correlation between venous peak enhancement value and SNRMTT was found in both white matter (r = 0.63, P < .001) and gray matter (r = 0.62, P < .001). Venous peak enhancement values also correlated significantly with SNRCBV in both white matter (r = 0.44, P = .004) and gray matter (r = 0.56, P < .001) and with SNRCBF in white matter (r = 0.65, P < .001) (Fig 5). Venous curve baseline interval and time-attenuation curve width did not correlate with perfusion parameter values or with SNR values.
|
Intrasubject Analysis
Central placement of venous ROI.—For central placement of the venous input ROI, mean CBF was 12.14 mL per 100 g per minute ± 4.49 in white matter and 42.83 mL per 100 g per minute ± 9.31 in gray matter. Mean CBV was 1.10 mL per 100 g ± 0.33 in white matter and 2.02 mL per 100 g ± 0.39 in gray matter. Mean MTT was 6.15 seconds ± 3.61 in white matter and 3.30 seconds ± 1.04 in gray matter.
Mean SNRCBV was 1.98 ± 1.38 in white matter and 2.57 ± 1.10 in gray matter. Mean SNRCBF was 1.31 ± 0.46 in white matter and 1.67 ± 0.43 in gray matter. Mean SNRMTT was 1.89 ± 1.23 in white matter and 2.02 ± 0.37 in gray matter.
Peripheral placement of venous ROI.—For more peripheral placement of the venous ROI (intentionally introducing volume averaging effect), the mean peak enhancement was 455.3 (12% decrease compared with central placement of the ROI; P < .001; 95% CI: 395.3, 515.3). The mean CBF was 14.45 mL per 100 g per minute ± 5.48 in white matter (19% increase; P = .02; 95% CI: 5.45, 23.45) and 51.07 mL per 100 g per minute ± 10.87 in gray matter (19% increase; P = .004; 95% CI: 22.07, 80.07). Mean CBV was 1.17 mL per 100 g ± 0.47 in white matter (16% increase; P = .11; 95% CI: 0.37, 1.97) and 2.41 mL per 100 g ± 0.57 in gray matter (19% increase; P = .004; 95% CI: 1.41, 3.41). Mean MTT was 5.78 seconds ± 3.29 in white matter (6% decrease; P = .08; 95% CI: 2.28, 9.28) and 3.27 seconds ± 1.3 in gray matter (1% decrease; P = .4; 95% CI: 1.27, 5.25).
Mean SNRCBV was 2.03 ± 1.40 in white matter (2.5% increase; P = .18; 95% CI: 1.73, 2.33) and 2.43 ± 0.99 in gray matter (5.4% decrease; P = .18; 95% CI: 1.43, 3.43). Mean SNRCBF was 1.21 ± 0.39 in white matter (8% decrease; P = .05; 95% CI: 0.81, 1.61) and 1.58 ± 0.26 in gray matter (5.4% decrease; P = .24; 95% CI: 0.78, 2.38). Mean SNRMTT was 1.53 ± 0.71 in white matter (19% decrease; P = .07; 95% CI: 0.53, 2.53) and 1.77 ± 0.46 in gray matter (12% decrease; P = .04; 95% CI: 1.07, 2.47).
|
Discussion |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
The commercial deconvolution-based software used in this project (CT Perfusion; GE Medical Systems), like all deconvolution software, requires operator placement of a small ROI over an appropriate cerebral artery to provide the algorithm with an arterial input function time-attenuation curve. Additionally, the algorithm also requires placement of an ROI thought to represent "pure blood"—that is, a reference ROI that corresponds to a volume without partial volume averaging effects with nonvascular tissue. Traditionally, this ROI has been placed over a large vein and is usually referred to as the venous time-attenuation curve. This "venous" ROI aids in computation of the CBV value for each tissue pixel and correction of the arterial time-attenuation curve for the effects of partial volume averaging so that the corrected arterial input function reflects concentration of iodinated contrast material. This automatic correction of the arterial time-attenuation curve to parallel the venous input function is the likely explanation for the lack of correlation between the arterial input factors and the CT perfusion parameter values in our study.
In both the first and second parts of the present study, our finding that the measured mean tissue CBF and CBV values are significantly related to the peak enhancement of the user-defined venous curve is notable. Both our correlation across subjects and our comparison within subjects agree in this observation. Our intrasubject comparison proved that slightly different ROI locations within the same venous structure resulted in significantly different tissue CBV and CBF values. This fact is important, since clinically useful definitions of ischemia based on absolute values of CBF and CBV depend on a high degree of measurement accuracy and a low degree of inter- and intraobserver variation. Although not a specific goal of this research study, it seems that variability related to manual ROI placement would likely be a major source of intra- and interobserver variability with respect to absolute values of CBV and CBF. It would be valuable to specifically assess the extent of observer variability with particular reference to ROI placement in a future study.
Our observation of increased tissue CBF and CBV values related to decreased peak enhancement and peripheral (compared with central) location of the venous ROI within the venous structure may be explained by recalling that, in deconvolution analysis, tissue CBV is computed as the integral of tissue time-attenuation curve divided by the integral of the venous time-attenuation curve. If there is partial volume averaging error in the venous time-attenuation curve, both peak enhancement and the result of integration are decreased proportionally by the same amount. This is true because of the linearity of attenuation with volume in CT. Thus, as the resulting integral of the venous time-attenuation curve is decreased proportionally with the degree of partial volume averaging, there is a corresponding increase in the ratio of tissue integral to venous integral (measured tissue CBV). Thus, the changes we observed in measured CBV in our study are consistent with the theory underlying the method. The increase in CBF that was seen in our study related to volume averaging effect in the venous ROI can be explained by recalling that, according to the central volume principle (6), CBF = CBV/MTT. That is, CBF changes in proportion to CBV when MTT is constant. The error leading to increase in CBV is primary; the increase in measured CBF follows the increase in measured CBV. We did not observe any statistically significant changes in MTT in our study.
Our finding of a small but significant correlation of peak curve enhancement on perfusion map SNR values was supported by our data from the intrasubject part of the study, which showed a trend toward decreased SNR that approached significance. Thus, both parts of the study agree and suggest that optimization of the venous ROI to minimize volume averaging effect should improve the resultant maps’ SNR values.
Our finding of no correlation between arterial and venous baseline intervals and mean perfusion parameter values was notable and suggests the possibility that the mean values measured with the deconvolution algorithm we used do not depend to a significant degree on patient cardiac output (the main factor that influences baseline interval). However, our finding of statistically significant negative correlation of arterial baseline interval with SNR on the CBV maps in the white matter location suggests that decreased cardiac output may be a factor that can negatively influence CBV map scan quality.
While our study provides new empirical data that show the influence of user-selected arterial and venous time-attenuation curves on measured mean perfusion parameter values and map SNR values, we recognize that our study is limited in certain important ways. First, while our study shows how small changes in ROI location can affect measured values of CBF and CBV, it does not preclude the possibility that increases in peak curve enhancement due to other factors (eg, an increase in contrast material infusion rate) could also influence perfusion parameter values and/or SNR. In fact, at least one prior study (12) has shown that increase of contrast material infusion rate can increase perfusion map SNR value. Comparison of both different total contrast material doses and different rates of infusion (eg, 4 mL/sec vs 8 mL/sec) is warranted to determine any effects of these variables on perfusion map quality. Determination of the optimal infusion rate for clinical CT perfusion studies should, in our view, involve consideration of SNR gains as well as patient safety and comfort.
Second, in our study, we have identified ROI placement as an important potential source of user-related variability, but our study design and scope do not provide enough information to empirically determine the best method for ROI placement. Future work will be needed to determine the optimal strategy for identification and placement of arterial and venous ROIs. Nevertheless, it is our opinion on the basis of this project that careful manual placement of a small ROI (eg, 4–6 pixels in size) within the central part of the artery or vein should result in minimal volume averaging effect. Additionally, by carefully examining several possible adjacent central locations and the resultant time-attenuation curves, one should be able to choose the ROI with the greatest peak enhancement value. This strategy should work to limit variability between and among readers. Future work should address inter- and intraobserver reliability by using such an optimized technique, and this technique should be compared with other methods of assessing cerebral perfusion, such as positron emission tomography. Ultimately, future development of a semiautomated system whereby the user locates the vascular structures of interest and the computer algorithm chooses the final optimal time-attenuation curve seems to us to be a worthwhile and achievable goal.
In conclusion, our study provides evidence that CBV, CBF, and SNR values are related to peak enhancement values of the user-selected time-attenuation curve functions required by deconvolution-based analysis software. Specifically, our data suggest that taking care to choose the venous ROI with the greatest peak enhancement value should help to limit variability in and increase quality of CT perfusion images.
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, J.D.E.; study concepts, J.D.E., S.M.K., V.A.L.; study design, all authors; literature research, J.D.E., S.M.K., V.A.L.; clinical studies, J.D.E., S.M.K., V.A.L.; data acquisition and analysis/interpretation, J.D.E., S.M.K., V.A.L.; statistical analysis, J.D.E., D.M.D.; manuscript preparation, J.D.E., S.M.K.; manuscript definition of intellectual content, all authors; manuscript editing, J.D.E., S.M.K.; manuscript revision/review and final version approval, all authors
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMaterials and MethodsResultsDiscussionREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A.S. Vagal, J.L. Leach, M. Fernandez-Ulloa, and M. Zuccarello The Acetazolamide Challenge: Techniques and Applications in the Evaluation of Chronic Cerebral Ischemia AJNR Am. J. Neuroradiol., May 1, 2009; 30(5): 876 - 884. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. Kudo, M. Sasaki, K. Ogasawara, S. Terae, S. Ehara, and H. Shirato Difference in Tracer Delay-induced Effect among Deconvolution Algorithms in CT Perfusion Analysis: Quantitative Evaluation with Digital Phantoms Radiology, April 1, 2009; 251(1): 241 - 249. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
V. Goh, S. Halligan, A. Gharpuray, D. Wellsted, J. Sundin, and C. I. Bartram Quantitative Assessment of Colorectal Cancer Tumor Vascular Parameters by Using Perfusion CT: Influence of Tumor Region of Interest Radiology, June 1, 2008; 247(3): 726 - 732. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H.M. Silvennoinen, L.M. Hamberg, L. Valanne, and G.J. Hunter Increasing Contrast Agent Concentration Improves Enhancement in First-Pass CT Perfusion AJNR Am. J. Neuroradiol., August 1, 2007; 28(7): 1299 - 1303. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Waaijer, I.C. van der Schaaf, B.K. Velthuis, M. Quist, M.J.P. van Osch, E.P.A. Vonken, M.S. van Leeuwen, and M. Prokop Reproducibility of Quantitative CT Brain Perfusion Measurements in Patients with Symptomatic Unilateral Carotid Artery Stenosis AJNR Am. J. Neuroradiol., May 1, 2007; 28(5): 927 - 932. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
B.D. Murphy, X. Chen, and T.-Y. Lee Serial Changes in CT Cerebral Blood Volume and Flow after 4 Hours of Middle Cerebral Occlusion in an Animal Model of Embolic Cerebral Ischemia AJNR Am. J. Neuroradiol., April 1, 2007; 28(4): 743 - 749. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. M. Provenzale, S. Mukundan, and D. P. Barboriak Diffusion-weighted and Perfusion MR Imaging for Brain Tumor Characterization and Assessment of Treatment Response. Radiology, June 1, 2006; 239(3): 632 - 649. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
I. van der Schaaf, E.-J. Vonken, A. Waaijer, B. Velthuis, M. Quist, and T. van Osch Influence of Partial Volume on Venous Output and Arterial Input Function AJNR Am. J. Neuroradiol., January 1, 2006; 27(1): 46 - 50. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
