| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Breast Imaging |
1 From the Medical Oncology Branch (A.T., S.B.W., S. M. Swain), Molecular Imaging Program (R.E., P.L.C.), and Biostatistics and Data Management Section (S. M. Steinberg, D.J.L.), Center for Cancer Research, National Cancer Institute; and Diagnostic Radiology Department, Warren G. Magnuson Clinical Center (D.M.T., C.K.C., B.J.W.), National Institutes of Health, 8901 Wisconsin Ave, Bldg 8, Room 5101, Bethesda, MD 20889-5015; and GE Healthcare Technologies, Hanover, Md (S.N.G.). Received June 2, 2006; revision requested August 1; revision received October 12; accepted November 3; final version accepted January 12, 2007. A.T. supported by the Clinical Research Training Program at National Institutes of Health (NIH), a public-private partnership supported jointly by the NIH, and a grant to the Foundation for the NIH from Pfizer Pharmaceuticals Group. Supported in part by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research. Address correspondence to S. M. Swain (e-mail: Sandra.M.Swain{at}Medstar.net).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
Materials and Methods: This study was conducted in accordance with the institutional review board of the National Cancer Institute and was compliant with the Privacy Act of 1974. Informed consent was obtained from all patients. Patients with inflammatory or locally advanced breast cancer were treated with one cycle of bevacizumab alone (cycle 1) followed by six cycles of combination bevacizumab and chemotherapy (cycles 2–7). Serial dynamic MR images were obtained, and the kinetic parameters measured by using three dynamic analytic MR methods (heuristic, Brix, and general kinetic models) and two region-of-interest strategies were compared by using two-sided statistical tests. A P value of .01 was required for significance.
Results: In 19 patients, with use of a whole-tumor region of interest, the authors observed a significant decrease in the median values of three parameters measured from baseline to cycle 1: forward transfer rate constant (Ktrans) (–34% relative change, P = .003), backflow compartmental rate constant extravascular and extracellular to plasma (Kep) (–15% relative change, P < .001), and integrated area under the gadolinium concentration curve (IAUGC) at 180 seconds (–23% relative change, P = .009). A trend toward differences in the heuristic slope of the washout curve between responders and nonresponders to therapy was observed after cycle 1 (bevacizumab alone, P = .02). The median relative change in slope of the wash-in curve from baseline to cycle 4 was significantly different between responders and nonresponders (P = .009).
Conclusion: The dynamic contrast-enhanced MR parameters Ktrans, Kep, and IAUGC at 180 seconds appear to have the strongest association with early physiologic response to bevacizumab.
Clinical trial registration no. NCT00016549 [ClinicalTrials.gov]
© RSNA, 2007
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
Conventional anatomic imaging depicts the physical size of tumors and is therefore considered a delayed indicator that might not enable reliable prediction of outcome. Use of dynamic MR imaging to characterize the tumor microvasculature is attractive because the technique is easy to implement, involves no radiation exposure (allowing repeated use), and generates semiquantitative information about vascular permeability and blood flow within tumors (6). Because dynamic MR imaging signal kinetics correspond to a tumor's vascular parameters, quantification of these signals may be used to evaluate response to angiogenic inhibition (1,3). The pathophysiologic basis for these contrast material kinetics has been attributed to the hyperpermeability of angiogenic vessels (7).
Compartment models such as the Brix model and the general kinetic model (GKM) (8,9), which are used to calculate the leakage of contrast material from the vascular space to the extravascular-extracellular space (leak rates) and the reflux of contrast material back to the vascular space (reflux rate), have been established. Previous dynamic MR imaging studies to examine the effects of anti–vascular endothelial growth factor treatment in athymic rats with human breast carcinoma xenografts have revealed decreases in fractional leak rates and reflux rates, compared with these rates in control animals, as early as 24 hours after treatment (10). Similar results were obtained with other preclinical models of ovarian cancer assessed with dynamic MR imaging (11).
In a previous clinical trial (12), we observed substantial decreases in pharmacokinetic parameters during treatment in patients with previously untreated inflammatory or locally advanced breast cancer who underwent one cycle of bevacizumab as a monotherapy (cycle 1) followed by six cycles of combination bevacizumab, doxorubicin, and docetaxel therapy (cycles 2–7). Thus, the purpose of our study was to retrospectively compare three dynamic MR imaging analytic methods to determine the parameter or combination of parameters most strongly associated with changes in tumor microvasculature during treatment with bevacizumab alone and bevacizumab plus chemotherapy in patients with inflammatory or locally advanced breast cancer.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
Measurable disease was quantified by using MR imaging according to the response criteria in solid tumors (RECIST) guidelines (14). The index lesion was the primary breast mass (if discrete), the enlarged axillary node, or both. One patient's axillary adenopathy was followed up with computed tomography because of the limited field of view at MR imaging. Disease response was assessed by the same reader (C.K.C.) by using the sequence that revealed the abnormality most clearly. For assessment of the lymph nodes, the non–fat-suppressed dynamic sequence (see Dynamic Contrast-enhanced MR Imaging Analysis) was used, whereas for assessment of the primary breast mass, measurements were obtained by using the contrast-enhanced fast spoiled gradient-echo sequence. According to RECIST criteria, the longest axial dimension was recorded regardless of the orientation. Residual disease was assessed with reference to the original prechemotherapy images. (Also see Region-of-interest [ROI] selection in Dynamic Contrast-enhanced MR Imaging Analysis section.) Patients were divided into responders, who had a partial response to therapy, and nonresponders, who had stable or progressive disease after therapy (12).
MR Imaging Techniques
Imaging was performed with patients in the prone position by using a 1.5-T MR system (GE Healthcare, Waukesha, Wis) with a dedicated receive-only four-channel dual breast coil. Baseline transverse gradient-echo images were obtained with a 25–35-cm field of view set to encompass both breasts and the axilla. First, diagnostic T2-weighted images were obtained by using 5225/100 (repetition time msec/echo time msec), a section thickness of 5 mm, and a matrix of 128 x 256 pixels. Next, dynamic contrast-enhanced MR images were obtained with a three-dimensional spoiled gradient-echo sequence by using 8/4.2, a 25° flip angle, 4–5-mm-thick sections through the entire breast, an acquisition time of 30 seconds per data set, and a matrix of 128 x 256 pixels. After three baseline nonenhanced image acquisitions, an automatic injector (Medrad Spectris, Indianola, Pa) was used to intravenously infuse gadopentetate dimeglumine (Magnevist; Berlex Laboratories, Wayne, NJ) at 0.3 mL/sec, for a total of 0.1 mmol per kilogram of body weight (typically 15–20 mL), followed by a 50-mL normal saline flush. The 0.3 mL/sec infusion rate was chosen to satisfy the Brix model. Continuous 30-second imaging data sets were obtained before, during, and after administration of the contrast medium for a total of 8 minutes to result in 20 repeated data sets.
Dynamic Contrast-enhanced MR Imaging Analysis
We compared three dynamic MR imaging analytic methods—the heuristic, Brix (8), and GKM (9) techniques—and two approaches to defining the ROI (10). The analysis and ROI selection were performed by three authors (A.T., R.E., and B.J.W.) with 1–6 years experience in breast MR imaging. These three methods were chosen because they represent three approaches to modeling: With the heuristic approach, one makes no assumption about tissue compartmentalization, and the two methods based on tissue compartmentalization differ in terms of the assumptions made about the influence of measured (GKM technique) versus modeled (Brix technique) arterial input function on the kinetic parameter values. The results of these three methods were obtained by transferring the acquired images to a personal computer and processing them with two analysis programs that were developed in house and based on the interactive display language (IDL, Boulder, Colo): the Dynamic program for the Brix and heuristics methods and the Cinetool program (GE Healthcare) with a KinMode analysis module for the GKM method. The GKM computer model was provided by one of the authors (S.N.G.), who programmed the module into the Cinetool research software.
The time–signal intensity data from each pixel on the image generated their own time–signal intensity curves, which were then evaluated according to the dynamic MR imaging method used with the heuristic model or one of the pharmacokinetic models (Brix or GKM) (15). The parameters used for the heuristic model were direct measurements of the slopes of wash-in and washout curves and the integrated area under the gadolinium concentration curve (IAUGC) for the first 90 or 180 seconds after contrast material injection. The parameters derived for the Brix model were the amplitude of enhancement and the reverse transfer constant (8). The parameters derived for the GKM method were the forward transfer rate constant (Ktrans), the backflow compartmental rate constant extravascular and extracellular to plasma (Kep), and the extravascular-extracellular volume fraction, with the assumption of a mean parenchyma T1 value of 850 msec. These rate transfer constants were originally described by Kety (9) and were modified to incorporate two-compartment models (9,16,17). Arterial input functions were obtained by drawing an ROI around the aorta from a single central section.
ROI selection.—Because there is no standardized method of plotting or summarizing the data within an ROI, we used two techniques to determine which strategy yielded the most clinically meaningful results. First, we chose an ROI from a single section of the most enhanced area of the tumor—that is, we performed "hot spot" measurements (18). These ROIs were hand drawn on color maps created with the Brix model and then exported to the two other models. Second, we used a whole-tumor pixel-by-pixel averaging technique: The software automatically drew the ROI by using a region-growing algorithm bounded by an assigned threshold value, encompassing approximately 90% of the tumor area. Whole-tumor regions included only numerical data from pixels with a goodness-of-fit index (R2 value) of at least 0.85, which effectively eliminated nonenhancing or poorly enhancing regions. We chose this technique on the basis of findings in a prior study that revealed no apparent advantage in subsampling tumor regions (19).
Statistical analyses.—In the current analysis, we used data from a single-arm, single-stage study designed for enrollment of 20 examinable patients to achieve 95% power for the detection of a change in any of four parameters measured from baseline to the end of cycle 1, equal to 1 standard deviation of the change. The two-tailed Wilcoxon signed rank test (or t test when appropriate) with an 
level of .05 was used to perform these analyses.
The primary outcomes of these analyses were changes in each dynamic MR imaging parameter. Initial explorations indicated that actual differences between the baseline measurement and the cycle 1, 4, or 7 measurement were more dependent on the baseline value (B) than were the corresponding relative differences—for example, the relative difference of (C1 – B)/B for each parameter, where C1 is the cycle 1 value. Thus, these relative differences, converted into percentages, were used as the primary data for the analyses since the evaluation of absolute differences might have revealed greater bias as a function of the magnitude of the baseline values themselves.
To test whether the percentage relative changes were associated with a statistically significant difference from zero, the Wilcoxon signed rank test was used. Comparisons between responders and nonresponders were performed purely as secondary exploratory evaluations with low power by using the exact Wilcoxon rank sum test.
Because of the large number of parameters evaluated, the various degrees of independence from one parameter to another, and the exploratory nature of the study, to interpret results in the context of the multiple comparisons performed, only P values of less than .01 were considered to indicate statistical significance and P values of .01–.05 were associated with trends. All reported P values were derived by using two-sided tests and are presented without adjustments for multiple comparisons. Analyses were performed by using 2001 SAS, version 8.2, software (SAS Institute, Cary, NC).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
Dynamic Contrast-enhanced MR Imaging Results
Serial GKM Ktrans parametric maps and gadolinium concentration–time curves (Fig 1) showed the distribution of high-vascular-permeability surface areas. A high degree of vascular permeability was observed on the initial maps (constructed at baseline and after cycle 1), but less vascular permeability was observed on the subsequent maps (constructed after cycles 4 and 7). The kinetics of gadolinium enhancement in the tumor, as shown in concentration-time curves, were presented for each map. The higher initial gadolinium concentration slope at baseline and after cycle 1 reflected increased compartmental wash-in of the contrast agent, and the time points after the wash-in peak reflected the compartmental outflow or washout.
|
The initial results obtained by using ROIs based on hand-drawn areas around the most enhanced area of the tumor (ie, hot spot) indicated trends toward differences in the heuristic parameters slope wash-in (P = .04), slope washout (P = .02), and IAUGC at 180 seconds (P = .049) between the responders and the nonresponders. Results also indicated trends toward differences in the GKM parameters Ktrans and Kep (P > .01 for both). In addition, when the relative differences in measurements obtained from baseline to cycle 1 were examined according to response, only the values for the heuristic IAUGC at 90 seconds (P = .05) and the slope washout (P = .05) differed. For measurements obtained from baseline to cycle 4, there was a trend toward differences in the heuristic IAUGC at 180 seconds only (P = .03). There were no significant differences.
At whole-tumor analysis, the results were considerably different (Table 1). Three parameters had a significant decrease as the relative change between baseline and cycle 1: GKM Ktrans (median relative change, –34%; P = .003), Kep (median relative change, –15%; P < .001), and IAUGC at 180 seconds (median relative change, –23%; P = .009). Between baseline and cycle 4 and between baseline and cycle 7, almost all parameters of the three models decreased (Fig 2, Table 1). For all parameters measured, the decrease in value was greater earlier in therapy (between cycles 1 and 4) than later in therapy (between cycles 4 and 7) (Table 2). We observed one trend toward a difference between the responders and the nonresponders when we compared the relative changes in all eight parameters from baseline to cycle 1 and from baseline to cycle 4: that for the median relative change in slope washout from baseline to cycle 1 (–104% for responders, 65% for nonresponders [P = .02]) (Table 3). Although the median relative change in slope wash-in from baseline to cycle 4 was significantly different between the responders (–60%) and the nonresponders (–31%) (P = .009), there was no significant difference in this parameter between the two groups from baseline to cycle 1 (Fig 3).
|
|
|
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
Although none of the assessed methods enabled successful prediction of the clinical response after cycle 1, we observed a significant difference in the heuristic slope wash-in between the responders and the nonresponders at cycle 4 (P = .009). This finding suggests that choosing the appropriate time to obtain dynamic MR imaging data may improve the results. However, we speculate that this parameter was the only one that enabled differentiation between the two groups because of the change in tumor volume throughout the course of treatment and because semiquantitative calculations are less sensitive to noise from heterogeneous tissues.
Our results indicate that methods in which arterial input functions are incorporated into the pharmacokinetic model or in which there is no attempt at physiologic correlation are the most reproducibly sensitive to angiogenic response to therapy. In that regard, our results from the heuristic model are consistent with previous results, which show that although the coefficient of variation for simple signal slopes (gradients) is higher than that for similar quantitative methods, IAUGC values may have a lower coefficient of variation than do other quantitative parameters (20). At comparisons of other heuristic parameters, such as maximal signal intensity change per time interval ratio and time to enhancement (time at which the signal intensity reaches 90% of its maximum), it has been shown that these parameters have a higher coefficient of variation (21).
In our study, the significance of the IAUGC likely resulted from the fact that simple signal integration is more accommodating of "noisy" data and less sensitive to data heterogeneity. Although the IAUGC is sensitive to change, it is limited because it cannot enable differentiation of the shape of various enhancement patterns. For example, a simple continuous increase in signal intensity can yield the same IAUGC value as a sharp increase and rapid decrease that is characteristic of angiogenesis enhancement patterns. Confirmation and validation of these findings in larger studies are needed.
Another important part of our analysis was the selection of the ROI on the MR image. We used the hot-spot and whole-tumor ROI selection methods. In a similar breast tumor study (18), a hot-spot method yielded results that were more useful than whole-tumor ROI method results for the differentiation of benign versus malignant disease (18). Although the results of another hot-spot technique indicated a larger difference between responders and nonresponders in a breast cancer neoadjuvant chemotherapy study (22), these results possibly were skewed because the whole-tumor data included the signal intensity of nonenhancing necrotic areas. In our evaluation, use of the hot-spot ROI method did not enable optimal detection of the small subtle changes that result from antiangiogenic therapy; however, when the whole tumor was evaluated, sensitivity improved. Moreover, determinations of the whole-tumor ROI were less subjective (23). Unlike the two previously described hot-spot methods, in which the analysis could have been biased owing to necrotic areas of the tumor, the ROI drawn by our automated ROI-drawing tool in the software supplied with the GKM algorithm included only numerical data from pixels with a goodness-of-fit index (R2) of at least 0.85, which effectively eliminated the nonenhancing or poorly enhancing regions.
Our study had several limitations: First, because of the rarity of inflammatory breast cancer, the study included data from a small number of patients. Second, because invasive tumor cells in inflammatory breast cancer are distributed more heterogeneously throughout the affected breast tissue, discrete tumor measurement is more challenging. Despite this problem, we were able to identify important changes in dynamic MR imaging parameters after bevacizumab treatment. Our third limitation was due to the study design: Because the patients received only one cycle of bevacizumab monotherapy before undergoing combination chemotherapy, the information regarding the effects of bevacizumab alone was limited. To be conservative, we required a stricter interpretation of the level of significance associated with a given P value because of the large number of tested parameters. Other methods in subsequent analyses may involve parameters that are predictive if they are examined as single parameters. Finally, we were restricted to using a low contrast material infusion rate (0.3 mL/sec) to accommodate the requirements of the Brix model. It is unknown whether faster bolus infusion would have produced different results.
In conclusion, our study results show that dynamic contrast-enhanced MR imaging can be used to reliably detect and characterize the effects of the angiogenesis inhibitor bevacizumab. We found that the MR-derived IAUGC and GKM parameters Ktrans and Kep were the most strongly associated with changes in inflammatory breast cancer response after treatment with bevacizumab alone and combination bevacizumab therapy and chemotherapy. This may be because the GKM parameters are based on independent measurements of arterial input function and T1 correction—rather than MR signal intensity alone—for determination of gadolinium concentrations. Thus, the physiologic characteristics of individual patients are taken into account in these values. With continued improvements in dynamic contrast-enhanced MR imaging, such as faster acquisitions, improved postprocessing algorithms, and improved ROI analysis, we expect this method to contribute valuable prognostic information for angiogenic therapy.
|
ADVANCES IN KNOWLEDGE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
|
IMPLICATIONS FOR PATIENT CARE |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Abbreviations: GKM = general kinetic model • IAUGC = integrated area under the gadolinium concentration curve • Kep = backflow compartmental rate constant extravascular and extracellular to plasma • Ktrans = forward transfer rate constant • RECIST = response criteria in solid tumors • ROI = region of interest
2 Current address: Washington Cancer Institute, Washington Hospital Center, Washington, D.C. 
Clinical trial registration no. NCT00016549 [ClinicalTrials.gov]
Authors stated no financial relationship to disclose.
Author contributions: Guarantors of integrity of entire study, A.T., S. M. Swain; 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.T., C.K.C., R.E., S.B.W., S.N.G., S. M. Swain; clinical studies, A.T., D.M.T., C.K.C., S.B.W., B.J.W., P.L.C., S. M. Swain; statistical analysis, A.T., S. M. Steinberg, D.J.L.; and manuscript editing, A.T., D.M.T., C.K.C., S.N.G., S. M. Steinberg, D.J.L., P.L.C., S. M. Swain.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONADVANCES IN KNOWLEDGEIMPLICATIONS FOR PATIENT CAREReferences |
|---|
This article has been cited by other articles:
|
|
 |
|
  S. Kim, L. Loevner, H. Quon, E. Sherman, G. Weinstein, A. Kilger, and H. Poptani Diffusion-Weighted Magnetic Resonance Imaging for Predicting and Detecting Early Response to Chemoradiation Therapy of Squamous Cell Carcinomas of the Head and Neck Clin. Cancer Res., February 1, 2009; 15(3): 986 - 994. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
= ROI data, C1 = cycle 1, C4 = cycle 4, LMB = left mouse button, RMB = right mouse button.
