| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Experimental Studies |
1 From the Laboratory for Minimally Invasive Tumor Therapy (A.H., H.P., A.U.H., S.L., S.N.G.), Department of Radiology (A.U.H., C.J.W., J.K., V.D.R., S.N.G.), Beth Israel Deaconess Medical Center, 1 Deaconess Rd, WCC 308B, Boston, MA 02215; and the Department of Interventional Radiology, Gustave Roussy Cancer Institute, Villejuif, France (T.d.B.). Received December 15, 2005; revision requested February 9, 2006; revision received June 1; accepted June 21; final version accepted September 22. Supported by National Cancer Institute Dana Farber/Harvard Cancer Center Renal Cancer SPORE grant 1 P50 CA10194-01. Address correspondence to S.N.G. (e-mail: sgoldber{at}caregroup.harvard.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Materials and Methods: All animal protocols and experiments were approved by the institutional animal care and use committee before the study was initiated. R3230 mammary adenocarcinoma was implanted in 11 rats. Tumors (18–20 mm) were scanned with dynamic 16-section CT at baseline and after administration of arsenic trioxide, which is known to cause acute reduction in blood flow. The concentration of arsenic was titrated (0–6 mg of arsenic per kilogram of body weight) to achieve a defined blood flow reduction (0%–75%) from baseline levels at 60 minutes, as determined with correlative laser Doppler flowmetry. The mean blood flow was calculated for each of four 5-mm sections that covered the entire tumor, as well as for the entire tumor after multiple sections were processed. Measurements obtained with both methods were correlated with laser Doppler flowmetry measurements. Interobserver agreement was determined for two blinded radiologists, who calculated the percentage of blood flow reduction for the "most representative" single sections at baseline and after arsenic administration. These results were compared with the interobserver variability of the same radiologists obtained by summing blood flow changes for the entire tumor volume.
Results: Overall correlations for acute blood flow reduction were demonstrated between laser Doppler flowmetry and the two CT perfusion approaches (single-section CT, r = 0.85 and r2 = 0.73; multisection CT, r = 0.93 and r2 = 0.87; pooled data, P = .01). CT perfusion disclosed marked heterogeneity of blood flow, with variations of 36% ± 13 between adjacent 5-mm sections. Given these marked differences, interobserver agreement was much lower for single-section CT (standard deviation, 0.22) than for multisection CT (standard deviation, 0.10; P = .01).
Conclusion: Multisection CT perfusion techniques may provide an accurate and more reproducible method of tumor perfusion surveillance than comparison of single representative tumor sections.
© RSNA, 2007
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Commercial CT software is used to obtain quantitative perfusion measurements, including tissue blood flow, blood volume, mean transit time, and capillary permeability (6). CT perfusion measurements have been shown to change as a result of antiangiogenesis therapy (6,7). Measurements obtained with such software have been demonstrated to be reliable (8–12) and reproducible (13), given the assumption that measurement of a representative section is sufficient to evaluate an entire tumor.
Nevertheless, tumors are heterogeneous. Accordingly, section thickness and selection, variables that have not been well studied in the past, might influence the overall results, especially for comparisons of sections from two examinations performed with a substantial time interval. By the same token, CT perfusion of the entire tumor volume—for example, a multisection approach used to cover the entire z-axis of the tumor—could potentially reduce this error. Thus, the purpose of our study was to compare prospectively single-section and multisection CT perfusion for tumor blood flow determination in an animal model.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
A parent tumor (approximately 1 cm in diameter) was initially harvested from a live carrier. Within 30 minutes of its dissection and removal, the tumor was homogenized with a tissue grinder (model 23; Kontes Glass, Vineland, NJ), and the tumor cells were suspended in 7 mL of a culture medium (RPMI 1640; INC Biomedicals, Aurora, Ill). An aseptic technique was used. In prior control experiments performed in our laboratory, this process resulted in a concentration of 1 x 108 cells per milliliter, with more than 95% cellular viability. During direct imaging, 0.2–0.3 mL of the tumor cell suspension was injected slowly with an 18-gauge needle into the mammary fat pad.
Animals were observed every 3–4 days to measure tumor growth. Tumors were allowed to grow for 14–24 days, until the desired treatment size (1.8–2.0 cm) was achieved. Tumor size was chosen to match size optimally with maximum z-axis coverage (four-section CT acquisition at 5-mm thickness = 20 mm). Two authors (A.H., H.P.) performed tumor induction, monitoring, and randomization. Tumors were measured with calipers daily until the desired size was reached. Solid tumor architecture was confirmed with ultrasonography before CT. Tumor size and composition were also confirmed with CT during experimentation. Before CT and arsenic administration, the tail vein was catheterized with a 24-gauge cannula. On completion of the study, animals were euthanized with an overdose of barbiturate (Somlethal; JA Webster, Sterling, Mass).
Laser Doppler Flowmetry
Laser Doppler flowmetry was considered the reference standard for measurement of tumor blood flow. This technique has been validated as accurately reflecting tumor microcirculatory blood flow (17,18). Tumor microcirculatory blood flow was measured with 24-gauge needle fiberoptic probes (Perfiflux; Perimed, Bury St Edmonds, England). With these probes placed within a tumor, backscattered light from the laser beam is used to measure tissue blood flow continuously to a depth of 1.5 mm.
A small incision in the skin overlying the tumor was made to facilitate the placement of a laser Doppler fiber. The fiber was inserted approximately 2 mm in the tumor until the bare fiber was completely inside the tumor. The final placement of the laser fiber was confirmed with CT. Two authors (A.H., H.P.) performed all laser Doppler recordings.
CT Protocol
Rats were placed in the prone position on the patient couch of the CT scanner and secured with a restraining strap around the abdomen to limit movement. Scanning was performed with a 16-section CT scanner (Light Speed Plus; GE Healthcare, Milwaukee, Wis) that had 1.25-mm detectors, which enabled section thickness of up to 20 mm. The CT imaging protocol involved three steps: a scout scan, a study of the abdomen and pelvis without contrast material enhancement, and a dynamic study with contrast material enhancement used for CT perfusion.
The nonenhanced study was performed initially to identify the location of the tumor for planning purposes. The following parameters were used: helical acquisition, 2.5 x 2.5 mm; section thickness, 5 mm; speed, 27 mm/sec; pitch, 1.3; 120 kV; 240 mA; scan field of view, 24 cm; and matrix, 512 x 512 mm. The images were then inspected by one author (A.H.) on the CT console, and the target sections were selected by two authors (A.H., C.J.W.) to plan the subsequent dynamic studies (ie, at baseline and after blood flow modulation) and ensure that the entire tumor was covered by the 20-mm scan volume. For the dynamic study, CT scanning was initiated 2 seconds before manual administration of a 2-mL/kg bolus of contrast material (Optiray 320; Mallinckrodt Pharmaceuticals, St Louis, Mo [300 mg of iodine per kilogram]) administered at a rate of 0.05 mL/sec for approximately 5 seconds via the 24-gauge tail vein catheter. Transverse images were obtained at 1-second intervals, covering the entire tumor (120 kV; 240 mA; scan field of view, 50 cm; matrix, 512 x 512 mm). Dynamic scanning was maintained during the bolus injection of contrast material and continued for 65 seconds.
Pharmacologic Modulation of Blood Flow
Arsenic trioxide (Sigma, St Louis, Mo) was constituted in phosphate-buffered saline solution to a concentration of 4 mg/mL. The doses of arsenic, an agent known to reduce blood flow acutely in this animal model (19), were varied to achieve a range of reduced tumor blood flow, given the previously demonstrated relationship between arsenic trioxide concentration and blood flow reduction at 60 minutes in this tumor model (19). Two animals each received 2.5, 4.0, 5.0, and 6.0 mg/kg of arsenic trioxide to create a range of decreased perfusion in blood flow. One animal received no arsenic trioxide to serve as a control. Arsenic was administered in blinded fashion by two of the authors (A.H., H.P.).
Image Processing for CT Perfusion
Because the manufacturer recommends use of sections no more than 5 mm thick for CT perfusion analysis, the 20-mm scan volume for each study was reconstructed into four separate contiguous sections collimated to 5 mm each. The 88 images from the 22 dynamic studies (one for each observer [A.H., H.P.] from each of the 11 animals) were processed separately by means of CT perfusion software (Perfusion 2.0; GE Healthcare) and a body tumor perfusion algorithm (20). A processing threshold of 0–120 HU was selected to permit appropriate subsequent analysis of both unenhanced and contrast material–enhanced soft tissue.
To determine arterial input, each of the two observers (A.H., H.P.) placed a region of interest over the best-visualized artery in the section plane (aorta, iliac, or superficial femoral artery). Time-attenuation curves were automatically generated for the arterial input along with perfusion maps for all the tissues within the scanning plane over the 65-second perfusion acquisition. To determine tumor perfusion, regions of interest were drawn freehand around the peripheral margin of the entire tumor by both observers with an electronic cursor. The observers took care to exclude peritumoral skin and fat and intraluminal gas by viewing a cine loop to gauge the extent of movement during acquisition. A global time-attenuation curve for the selected tumor tissue and the mean blood flow for the tumor tissue within the region of interest were derived.
For the multisection method, the blood flow maps of all four single sections of a given dynamic CT study were saved in a high-spatial-resolution gray-scale format at a factor of 0.25 and transferred onto a desk computer. A factor of 0.25 was used to avoid loss of information, as instructed in the software manual, because the pixel values in the saved images may have been different from the original values in the functional map.
Every image was then loaded by an author (A.H.) into a public domain image processing program (ImageJ; National Institutes of Health, available at http://rsb.info.nih.gov/ij/download.html) to calculate the histogram of the perfusion values within the region of interest (number of pixels for each value). The pixel data from the four tumor sections were then combined in Excel to calculate the mean CT perfusion for the entire tumor volume blood flow. Calculations of blood flow were performed in duplicate and blinded fashion by two observers (A.H., H.P.).
Data and Statistical Analysis
The data were analyzed in three ways to ensure proper validation of results. First, the two reviewers' findings were compared to determine the degree of concordance for measuring changes in blood flow by generating CT perfusion images from specified single sections. This was achieved by randomly and consensually selecting one of the four single 5-mm-thick sections for each animal, both at baseline and after arsenic administration. The percentage change in blood flow between the two CT sections was then calculated independently by two radiologists (A.H., H.P.) who had 2 and 4 years of experience, respectively, in the field of tumor imaging. They were blinded to the results of laser Doppler flowmetry and were unaware of each other's findings.
Next, results were compared with those of the reference standard (laser Doppler flowmetry). The percentage of blood flow reduction after arsenic administration was calculated for each animal by the two observers, who used both single-section and multisection techniques. To minimize reader bias, these two methods were performed on separate days and in different randomized orders for the two readers. The percentage change in blood flow was then calculated for each animal and reader and was compared with the laser Doppler flowmetry readings.
Once validation was achieved, interobserver error was measured for both single-section and multisection CT techniques. For the single-section technique, the two observers independently selected the 5-mm sections that they determined best depicted the overall tumor blood flow at baseline and after arsenic administration and performed independent analyses of the change in blood flow. For the multisection technique, the mean entire tumor drop in blood flow was calculated independently by the two observers. Finally, the results for interobserver agreement for the single- and multisection techniques were then correlated and compared with statistical software (SAS version 9.1; SAS Institute, Cary, NC).
Linear regression analysis was performed for all comparisons, both separately for each observer and with pooled data. Differences between the observers in these regression lines were tested by using an interaction of intercept and slope with the observer term. Additionally, multiple regressions were performed to determine the independent contribution of section and volume, with laser Doppler flowmetry measurements serving as the reference standard. Comparisons of single-section and multisection CT with the reference standard were also evaluated with the Bland-Altman method to determine the limits of agreement. We used the Fisher test for equality of variances, with the mean square error (square of the estimated error or residual SD) serving as the estimate of the variances. The P value cutoff for statistical significance was set at .05 (two-tailed level of significance).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
|
|
Comparison of Single- and Multisection CT Perfusion with Laser Doppler Blood Flow Measurement
Acute blood flow changes calculated for both single-section and multisection CT perfusion correlated with laser Doppler flowmetry findings (Fig 3). There were no significant differences in multisection or single-section CT for either slope or intercept between the observers. Additionally, use of Bland-Altman method showed that there was no skewing, deviation, or significant results in testing to determine whether the slope was anything other than 1 or whether the intercept was anything other than 0 for either single-section or multisection techniques.
|
|
Interobserver Agreement
Both reviewers selected the same two sections as representative of the pre- and postarsenic studies in only four of 11 animals (36%). Accordingly, substantial differences in interreader agreement were found between single- and multisection CT perfusion techniques. While the mean values did not differ between readers for either technique (multisection CT, P = .32; single-section CT, P = .22; paired t test) or between technique for either reader (reader 1, P = .25; reader 2, P = .78; paired t test), the Bland-Altman method demonstrated much lower interobserver agreement (SD, 0.22) for the single-section technique than for the multisection technique (SD, 0.10; P = .01) (Fig 4). This corresponded to a decrease in linear correlation between the observers (multisection CT, r = 0.96 and r2 = 0.92; single-section CT, r = 0.79 and r2 = 0.63).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
Various authors have investigated the reproducibility of single-section CT perfusion findings (13,24–26) and have found a high degree of correlation between observers for defined CT perfusion single sections; these authors included Blomey et al (25) (r2 = 0.83), Miles and Griffiths (26) (r2 = 0.94), and Goh et al (13) (intraclass correlation coefficient = 0.89). However, none of them have investigated the degree of error, particularly in interreader agreement, when images must be compared to calculate modification of blood flow in two consecutive CT perfusion scans, a common clinical end point. This, in turn, can lead to underestimating a potential cause of tremendous variability—the error attributable to section selection. Indeed, in our study when the selection of the single "most representative" section was left to the preference of the two blinded observers, the r2 value dropped from 0.92 to 0.63, and a significantly greater SD was identified for single-section (0.22) than for multisection CT (0.09).
These findings strongly suggest that the selection of the appropriate section can be a major source of variability. Indeed, in day-to-day clinical practice this problem may be greater than we observed because, unlike in our study, sequential assessment of response is frequently made at prolonged follow-up intervals by different observers on tumors that may have changed in shape, size, or orientation. Furthermore, given that no two observers are likely to select the identical two dynamic sections every time (this occurred in our controlled study only 36% of the time) and given that tumor blood flow is heterogeneous (overall in our study, mean blood flow varied by approximately 50% among the four sections), the use of single "nonrepresentative" sections potentially limits the application for assessment of therapeutic response. Even when the images are matched on the follow-up examination, the selected dynamic section will not be exactly the same because (a) patients are not oriented in exactly the same position during initial and follow-up studies, (b) the images do not correspond precisely, and (c) the size and shape of the tumor may have changed.
To overcome this potential major flaw, we used a volumetric method to estimate perfusion, with use of a multisection approach to cover the entire tumor. This method offers intrinsic advantages over the conventional single-section method. Since the whole tumor was imaged, we took advantage of all available data and not just a data subset. Additionally, coverage of the entire volume is less dependent on patient positioning and section selection. There is no need to select matching images on the follow-up study (registration) since total volume is measured, making the volumetric perfusion measurements independent of registration.
We assessed the accuracy of this multisection CT perfusion technique for demonstrating acute changes in blood flow by validating it against laser Doppler flowmetry. Indeed, a higher correlation was found between multisection CT perfusion and laser Doppler flowmetry (r2 = 0.87) than between dynamic single-section CT and laser Doppler flowmetry (r2 = 0.76). Moreover, the SD from our regression line was only 9% for the multisection technique, which was less than half the 21% seen for the single-section technique. Thus, predictability was improved by adding multisection processing, and the multisection technique was therefore closer to the reference standard. Nevertheless, we acknowledge that further calibration of CT perfusion may be necessary to match the laser Doppler values precisely.
Our study had some limitations. One potential point of concern is the fact that although laser Doppler flowmetry is used to measure blood flow at a maximum depth of 1.5 mm, we observed a more favorable correlation for multisection CT than for single-section CT. It could be argued that theoretically in a heterogeneous tumor a single appropriately registered section should be similar to the very small zone of laser Doppler blood flowmetry measurement. Yet in practice, likely given errors in registration, the averaging effect of the multisection technique resulted in better correlation with laser Doppler flowmetry measurements. Hence, it is possible that even better correlations could be achieved with better matching of regional blood flow or with use of other reference standard procedures, such as microspheres, that permit more global blood flow measurements (27). However, we purposely chose laser Doppler flowmetry because it enabled continuous blood flow monitoring at the exact time of both CT perfusion scans. Regardless, these issues highlight an important potential benefit of CT perfusion over other methods. Once it is appropriately validated, CT perfusion could enable repeatable noninvasive monitoring of blood flow changes in different areas of a tumor—something that is not possible with microspheres or laser Doppler flowmetry.
Our study was performed in only one animal model with well-defined tumor margins and tumor size adequately covered by the four single sections. Hence, it will be necessary to confirm the results in other models before application in human subjects. Moreover, at the time of this writing, the expected changes in perfusion parameters contingent on antiangiogenesis therapy were uncertain (28). Whether the limits of agreement we found are acceptable for clinical practice should become clearer as data from human subjects emerge.
Another potential clinical limitation is that it may not always be feasible to cover the entire tumor, particularly when the lesion is larger than the z-axis coverage. While this problem is continuously decreasing as newer generations of CT scanners (especially flat-plate CT scanners) are covering greater volumes of tissue in reduced time periods, further clinical validation will be necessary to determine exactly how much volumetric coverage will be required for different tumors in clinical practice. Indeed, it can be argued that fewer sections or a 20-mm-thick section could reduce the error to clinically acceptable levels. We were unable to assess this possibility directly because the software we used recommends a scan thickness of no more than 5 mm. Nevertheless, this possibility underscores a main point of our investigation—namely, that greater attention to section thickness and volumetric coverage will be needed for future studies of CT perfusion.
Given our CT perfusion results, we postulate that the multisection technique provides better results than the use of thinner single sections for the three perfusion parameters quantifiable with CT (blood volume, mean transit time, and tumor permeability). Further validation is needed, and it must be acknowledged that the postprocessing of our data was somewhat tedious. More computer automation, particularly for fusion of the multisection data, will be required to make this technique more practical and straightforward for clinical use.
As to the practical implications, section thickness is likely to be an important variable for sequential CT perfusion scanning of tumors. Our data suggest that single-section CT perfusion may not be as robust for comparing temporal changes in tumor blood flow as a multisection technique. Although further refinement and validation is needed, multisection CT perfusion will likely be useful for obtaining more global and accurate information about tumor blood flow and changes after therapy.
|
ADVANCES IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: SD = standard deviation
2 Current address: Department of Radiology, Beaujon Hospital, Paris, France. 
Authors stated no financial relationship to disclose.
Author contributions: Guarantor of integrity of entire study, S.N.G.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, A.H., A.U.H., T.d.B., S.N.G.; experimental studies, A.H., H.P., C.J.W., J.K., S.L., S.N.G.; statistical analysis, A.H., A.U.H., S.L., V.D.R., S.N.G.; and manuscript editing, A.H., H.P., A.U.H., J.K., T.d.B., V.D.R., S.N.G.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. Sabir, R. Schor-Bardach, C. J. Wilcox, S. Rahmanuddin, M. B. Atkins, J. B. Kruskal, S. Signoretti, V. D. Raptopoulos, and S. N. Goldberg Perfusion MDCT Enables Early Detection of Therapeutic Response to Antiangiogenic Therapy Am. J. Roentgenol., July 1, 2008; 191(1): 133 - 139. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. B. Kruskal Can Contrast-enhanced US with Targeted Microbubbles Monitor the Response to Antiangiogenic Therapies? Radiology, February 1, 2008; 246(2): 339 - 340. [Full Text] [PDF]
|
|
|
|
|
|
A. Hakime, A. Hines-Peralta, H. Peddi, M. B. Atkins, V. P. Sukhatme, S. Signoretti, M. Regan, and S. N. Goldberg Combination of Radiofrequency Ablation with Antiangiogenic Therapy for Tumor Ablation Efficacy: Study in Mice Radiology, August 1, 2007; 244(2): 464 - 470. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
