| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, University of Michigan Health System, 1500 E Medical
Center Dr., Ann Arbor, MI 48109-5030.
2 Department of Neurology, University of Michigan Health Systems, Ann Arbor,
MI.
3 Department of Radiation Oncology, University of Michigan Health Systems, Ann
Arbor, MI.
Received March 3, 2008;
accepted after revision August 14, 2008.
Address correspondence to E. A. Smith
(ethans{at}med.umich.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Thirty-three patients who had undergone treatment for primary brain tumors in whom routine MRI showed new contrast-enhancing lesions were retrospectively studied. The final diagnosis was assigned using histopathology (n = 13) or imaging follow-up (n = 20; range, 2–27 months). Ratios of three metabolites (Cho, Cr, and NAA) were calculated, and the results were correlated with the final diagnosis using a Wilcoxon's rank sum analysis. A logistic regression model was then used to create a prediction model based on the most statistically significant ratio.
RESULTS. Elevations of the metabolic ratios Cho/Cr (p < 0.001) and Cho/NAA (p < 0.001) and a decrease in the ratio NAA/Cr (p = 0.018) were found in patients with recurrent tumor (n = 20) versus those with postradiation change (n = 13). A prediction model using the Cho/NAA ratio yielded a sensitivity of 85%, a specificity of 69.2%, and an area under the receiver operating characteristic curve of 0.92.
CONCLUSION. An elevated Cho/NAA ratio correlated with evidence of tumor recurrence and allowed creation of a prediction rule to aid in lesion classification. The results suggest that MR spectroscopy is a useful tool in assigning patients with nonspecific enhancing lesions to either invasive biopsy or conservative management.
Keywords: brain neoplasms • MR spectroscopy • postradiation changes • postradiation necrosis • recurrent glioma
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MR spectroscopy (MRS) findings have been shown to correlate well with pathologic specimens obtained at biopsy or resection [5–7]. The results of previous studies suggest that MRS can discriminate between post radiation change and recurrent neoplasm in patients who have been treated for a primary brain tumor and have nonspecific contrast-enhancing lesions on follow-up imaging [8–10].
Prior investigators evaluating the ability of multivoxel MRS—either 2D or 3D chemical shift imaging (CSI)—to differentiate between recurrent neoplasm and postradiation change have suggested specific ratios that could be used as cutoffs to define the different groups [9–12]. Although specific ratio cutoffs may be helpful in defining the ability of the test to detect statistically significant differences between the groups, a more clinically applicable measure of the clinical utility of a test is the assessment of the posttest probability of tumor recurrence in patients with previously treated primary brain neoplasms and found to have new contrast-enhancing lesions on follow-up anatomic MRI.
The purpose of this study was to expand the work performed by Weybright et al. [9] by developing a method using alterations in the ratios of standard brain metabolites—choline (Cho), creatine (Cr), and N-acetylaspartate (NAA)—to predict the probability of tumor recurrence in patients previously treated for brain tumors with new contrast-enhancing lesions. The goal of this method is to identify future patients in whom the need for invasive biopsy and its associated risks could be eliminated. Development of such a prediction rule will enhance the clinical utility of MRS in this population.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Lesion Classification
Clinical, neuroradiologic, and histopathologic follow-up after MRS was used
to classify the new contrast-enhancing lesions as either recurrent neoplasm or
postradiation change. Lesions were classified as tumor recurrence if they met
any of the following criteria: verification of active tumor at the site on
biopsy, surgical resection, or autopsy. Lesions were also classified as tumor
recurrence if they demonstrated size progression on subsequent MRI in a manner
consistent with tumor growth. At our institution, clinical and MRI
surveillance occurs every 3–6 months at the clinician's discretion.
Lesions classified as radiation injury met one of the following criteria:
evidence of radiation injury without tumor at histopathologic examination or
conventional MRI follow-up showing prolonged stability or spontaneous
regression of the contrast-enhancing lesion without evidence of progression on
follow-up imaging and clinical evaluation.
MRS Protocol
All clinical MRS examinations were performed on a 1.5-T scanner.
Conventional MRI was performed in conjunction with all MRS examinations and
included unenhanced and gadolinium-enhanced T1-weighted sequences in the axial
and sagittal planes, axial T2-weighted images with fat saturation, axial FLAIR
and diffusion-weighted sequences, and gadolinium-enhanced T1-weighted images
in the coronal projection. MRS was performed as the last sequence in the study
so that voxel placement could be performed over the region of
interest—in this study, the contrast-enhancing lesion. Because of this
imaging sequence, all MRS examinations were performed after the administration
of gadolinium (gadopentetate dimeglumine [Magnevist, Bayer Schering
Pharma]).
MRS was performed on a 1.5-T scanner using standard manufacturer-recommended pulse sequences and set tings. A birdcage-design transmit–receiver head coil was used. The 2D chemical shift image was acquired using a point-resolved spectroscopy sequence (PRESS) volume localization with a 16 x 16 phase-encoding matrix over a 16-cm field of view (FOV). Water suppression was achieved using chemical-selective suppression (CHESS) [13], and signals were digitized to 2,048 complex point pairs over a 2.5-kHz spectral bandwidth. Relative spectral features of the CHESS pulses are preset by the vendor (GE Healthcare) and are not readily adjustable by the user. The MRS acquisition also interleaved nonsuppressed water signals for automatic phase correction of water-suppressed 2D CSI spectra.
The parameters used for all 2D CSI examinations were as follows: PRESS; TR/TE, 1,500/144; FOV, 16 cm; matrix, 16 x 16; slice thickness, 10–20 mm; acquisition, 1 average; and scanning time, 4 minutes 20 seconds. The volume of interest (VOI) was placed on either nonangled contrast-enhanced axial T1-weighted images or FLAIR images to ensure that voxels were placed with certainty over both the contrast-enhancing lesion and the adjacent normal-appearing white matter.
The volume of the VOI varied depending on the location in the brain in which the lesion was located, with a smaller VOI being used in the posterior fossa and a larger VOI being used in the supratentorial compartment. In all patients, the VOI was placed in such a way as to include both the lesion and the adjacent normal-appearing brain. At least two MRS sequences were usually performed with the VOI placed to include the lesion and the adjacent normal white matter (defined as areas not displaying signal alteration on T2-weighted or FLAIR sequences) at two different levels.
In addition to the out-of-FOV saturation bands routinely placed for all spectroscopic examinations, in-FOV saturation bands were placed when necessary for suppression of adjacent osseous and CSF-containing structures adjacent to the tissue of interest to optimize the 2D CSI spectra. Automatic prescanning was performed twice before each spectroscopic scan to ensure adequate water suppression. The full width at half maximum (FWHM) was kept under 10 with a flip angle of approximately 125° and water saturation was between 98% and 99% to allow separation of the Cho and Cr peaks.
The spectroscopic data were retrospectively reviewed by a single neuroradiologist with 11 years' subspecialty experience and 10 years' experience with MRS. These data were transferred to a separate workstation (Sun, GE Healthcare) for offline postprocessing using FuncTool 2000 software (GE Healthcare). In the previously defined VOI, 1 x 1 x 1 cm voxels were individually placed in the lesion and in normal-appearing adjacent white matter. The various metabolic peaks were measured in each individual voxel using integrals of each peak as a measurement of intensity. The integration limits of the respective peaks were manually defined by the same neuroradiologist before computerized calculations, as has been previously described in the literature [14] (Fig. 1A, 1B, 1C). The integrals of the metabolites in the same voxels were used for calculation of the metabolite ratios. The spectra were analyzed for the signal intensity of NAA, Cho, and Cr and the presence of lactates and lipids. The following metabolic ratios were manually calculated: NAA/Cr, Cho/Cr, and Cho/NAA. For each patient, the highest NAA/Cr, Cho/Cr, and Cho/NAA ratios in any one voxel were used in subsequent analyses.
|
|
|
Patient demographics, initial pathology, MRS findings, follow-up interval, and final diagnosis based on histopathology or long-term follow-up are presented in Table 1.
|
For this retrospective study, recurrent neoplasm was defined by histopathology results or interval progression detected on routine MRI follow-up examinations performed every 3–6 months as part of standard clinical care. Radiation necrosis was defined by histopathology results or by stability or interval regression of the contrast-enhancing lesion on anatomic MRI follow-up examinations per formed as part of standard of clinical care. The frequency of follow-up imaging was at the clinician's discretion and varied between 3 and 6 months.
Data and Statistical Analysis
To determine statistically significant differences in each metabolic ratio
between tumor recurrence and radiation change, the Wilcoxon's rank sum test, a
nonparametric test for nonpaired data, was initially used. A logistic
regression model was constructed to predict tumor recurrence using the single
best metabolic ratio. The odds ratio (OR) and its confidence interval (CI)
were obtained from the final model as a measure of the association between the
predictor and tumor recurrence. The discriminatory performance of the model
was assessed by a receiver operating characteristic (ROC) curve analysis and
the area under the ROC curve was calculated for the final model. Subsequently,
the probability of tumor recurrence was estimated from the final logistic
regression model–based metabolic ratio data.
We used the bootstrap method to obtain more precise estimates of the OR and its 95% CI. We drew 10,000 bootstrap samples with replacement of the same sample size as the underlying population. Bootstrapping is the method by which a repeated number of random samples are drawn with replacement from the observed samples to obtain a more reliable estimate of the standard error of the estimated OR of the predictor variable. The precision of an estimated OR is dependent on the sampling distribution of the OR obtained from the sample drawn from the underlying population. Multiple random samples of the underlying population yield a more precise estimate of the distribution of the OR than a single sample. Bootstrapping approximates this sampling process.
All statistical analyses were performed using statistics software (Stata, version 9.0, Stata), and statistical significance was set at 0.05.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The mean interval between the completion of radiation treatment and the development of new contrast-enhancing lesions on follow-up MRI was 24.6 months (range, 1–108 months; median, 17 months). In patients classified as having tumor recurrence (n = 20), the mean time to the development of contrast-enhancing lesions was 26.1 months (range, 1–108 months; median, 18.5 months), whereas in patients classified as having radiation change (n = 13), the mean time to the development of contrast-enhancing lesions was 17.1 months (range, 3–39 months; median, 14 months).
Lesion Classification
Tumor recurrence was confirmed by biopsy, surgical resection, or autopsy in
nine lesions. Eleven lesions were also classified as tumor recurrence on
subsequent MRI (mean, 16.3 months; range, 2–24 months). Radiation injury
was confirmed by histopathology in four lesions. Nine lesions were also
classified as radiation change on conventional MRI follow-up (mean, 15.7
months; range, 9–22 months).
MRS Metabolic Ratios as Predictors of Tumor Recurrence
The patients with tumor recurrence had higher mean values of Cho/Cr and
Cho/NAA than those with radiation change but lower mean values of NAA/Cr
(Table 2). Statistically
significant differences between the two patient groups were detected using the
Wilcoxon's rank sum test across all three metabolic ratios (Cho/NAA,
p = 0.0001; Cho/Cr, p = 0.0007; NAA/Cr, p =
0.0183). The difference in 95% CIs of Cho/NAA in patients with tumor
recurrence compared with those with radiation change was greater than the
difference in 95% CIs for Cho/Cr or NAA/Cr in the same groups (Fig.
1A,
1B,
1C). Therefore, we used the
Cho/NAA ratio as the predictor of tumor recurrence in the development of the
logistic regression model.
|
Model for Predicting Tumor Recurrence
The final logistic regression model using Cho/NAA for predicting tumor
recurrence had an area under the ROC curve of 0.92
(Fig. 2), indicating excellent
discrimination between tumor recurrence and radiation change. The model
attained a sensitivity of 85%, a specificity of 69.2%, a positive predictive
value of 81%, and a negative predictive value of 75%. The
R2 for the model, a measure of the fraction of variation
in the data attributable to the predictor (Cho/NAA), was 0.45. The OR for
Cho/NAA was 12.7 (95% CI = 2.05–78.52). Therefore, for every unit
increase in the Cho/NAA ratio, the odds of tumor recurrence increased
approximately 13-fold. Bootstrapping yielded a bias-corrected 95% CI of
1.30–4.83 for the OR after 10,000 repetitions. The probability of tumor
recurrence as a function of the Cho/NAA ratio is depicted in
Figure 3. For example, all
other factors being equal, a patient with a lesion Cho/NAA ratio of 1.0 would
have an 11% probability of tumor recurrence, whereas a patient with a lesion
that has a Cho/NAA ratio of 2.7 would have a 90% probability of tumor
recurrence.
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The results of recent studies have suggested that MRS is an effective tool in discriminating between postradiation change and recurrent tumor in patients who have previously been treated for a primary intracranial neoplasm [9–12] and have quantified the alterations in MRS ratios of Cho/NAA, NAA/Cr, and Cho/Cr associated with tumor recurrence. MRS findings have also been shown to affect the clinical management of these patients [18]. In the context of an evaluation to discriminate between radiation injury and recurrent brain tumor, there is current debate regarding the optimal metabolite ratios for differentiation of these diagnostic entities. In particular, the use of both the plain metabolite ratios from data obtained in the diagnostic voxel (e.g., Cho/Cr) and the normalized metabolite ratio comparing the single metabolite levels in the normal and affected parts of the brain (e.g., Cho/normalized Cho) have been evaluated [5–12], although no consensus has been reached as to the optimal ratio.
In this study, we used the technique of calculation of plain metabolite ratios from data obtained within the diagnostic voxel. Our results confirm significant elevations of the Cho/NAA and Cho/Cr ratios with a concomitant reduction in the NAA/Cr ratio in contrast-enhancing lesions representing tumor recurrence compared with lesions representing radiation change. Further, although all three metabolic ratios significantly predicted tumor, the most significant correlation was noted with Cho/NAA.
Some authors have suggested using different cutoff values for these metabolite ratios to differentiate between recurrent tumor and radiation injury. These numeric heuristics or rules of thumb may aid the radiologist at the time of image interpretation. However, numeric thresholds may not adequately measure the true utility of a diagnostic test. Guyatt et al. [19] have noted that the clinical relevance of a positive or negative test may be estimated by the corresponding likelihood ratio. Although likelihood ratios are an effective tool to guide selection of a diagnostic test, clinical decision making at the level of the individual patient relies ultimately on the assignment of posttest probability—that is, the probability of disease given the test result.
The use of a regression model to derive posttest probabilities has been previously explored in the use of CT perfusion to predict the probability of treatment response in patients with advanced squamous cell carcinoma of the upper aerodigestive tract and to suggest how these probabilities can be used to direct clinical management [20].
As we previously noted, of the three metabolite ratios that we evaluated, Cho/NAA had the most significant association with tumor and showed the widest difference in 95% CIs between patient groups. Thus, to estimate posttest probability, we used Cho/NAA as the predictor of tumor recurrence to develop the regression model. The final logistic regression model for predicting tumor recurrence attained an area under the ROC curve of 0.92, indicating excellent discrimination between tumor and radiation change. The OR was 12.7, with a 95% CI of 2.1–78.6 (p = 0.006). The large OR point estimate suggests a substantial effect of increasing Cho/NAA on the probability of recurrent tumor. The broad 95% CI likely reflects the imprecision of the estimate. A larger sample size may increase the precision of the estimated OR.
Our result of discriminating tumor recurrence from radiation change, as shown by the area under the ROC curve of 0.92, is higher than the 0.78 described by Terakawa et al. [21] with 11C-methionine PET. It is difficult to compare our results with those of Sugahara et al. [22] who evaluated MR perfusion alone and Jain et al. [23] who evaluated first-pass CT perfusion as methods for differentiating tumor recurrence from radiation change because those investigators did not present data from ROC analyses. Law et al. [24] evaluated the use of MRS alone and in combination with MR perfusion in grading gliomas but grouped both pretreatment and posttreatment patients together. Although Law et al. did not present the numeric values for areas under the ROC curve, a visual comparison of the ROC curves reveals that our results show better discriminatory ability than their analysis of the Cho/Cr ratio alone and that our results are similar to their analysis of the Cho/Cr ratio and relative cerebral blood volume.
|
|
15% probability of tumor recurrence is deemed acceptable for excluding
tumor and 
80% probability acceptable for diagnosis of recurrence, then
patients with a Cho/NAA ratio of 1.1 could be assigned imaging follow-up
and those with a Cho/NAA ratio of > 2.3 could be immediately treated.
Patients with a Cho/NAA ratio between those values would undergo biopsy. Under
this stratification scenario, 13 patients would have undergone a biopsy: seven
patients with radiation change and six with recurrence. No patients with
recurrence would have been assigned to routine imaging follow-up and only one
patient with radiation change would have been immediately treated for tumor
recurrence. We intentionally proposed criteria for lesion classification in
the prediction model that were conservative and that may result in more
patients undergoing lesion biopsy. However, we chose the conservative criteria
to reduce the number of patients with radiation change being empirically
treated. A significant limitation of our study is that histopathologic diagnosis was available in only 13 of the 33 patients. In the remaining patients, the final classification of recurrent tumor and radiation change was based on clinical follow-up and long-term imaging follow-up. To address this limitation, we attempted to maximize the follow-up interval for these patients. As we described earlier, the mean follow-up interval for patients classified as having postradiation change without histopathologic confirmation was 15.7 months (range, 9–22 months). All but one of these patients was followed up for more than 12 months. One patient died after only 9 months of follow-up from a new intracranial mass located at a site different from the initially treated primary brain neoplasm. We think that this patient was correctly placed in the postradiation change group because on follow-up imaging less than 4 months after the initial MRS evaluation, the contrast-enhancing lesion that had prompted MRS evaluation showed marked regression. It is unlikely that a recurrent tumor that has shown new contrast enhancement would spontaneously regress without further treatment. As with other methodologically similar studies evaluating MRS in the differentiation of recurrent brain tumor from radiation change [21, 22, 24], there is some heterogeneity in the original tumor pathology in our population. However, we used rigorous inclusion and exclusion criteria to amass as homogeneous a sample as possible with most cases representing grade II–IV gliomas (Table 1).
Regarding the logistic regression model, the model could not be directly verified given the small sample size. Bootstrapping was performed to approximate verification. Given the relatively small study sample size, we incorporated only one metabolite ratio as a predictor to preserve model parsimony and to avoid overfitting. Other patient-related characteristics, such as age, tumor type, and radiation dose, or study-related characteristics, such as lesion size or other metabolite ratios, may refine the model and should be explored in a larger patient population. Future work should be targeted toward verifying this prediction model.
In conclusion, our data suggest that elevation of the Cho/NAA ratio predicts tumor recurrence in patients with radiation-treated primary brain tumors who present with new contrast-enhancing lesions on conventional MRI follow-up. Posttest probabilities of recurrent tumor can be explicitly quantified for the individual patient using a prediction model. Explicit assignment of risk of recurrence may facilitate clinical decision making. In the future, prediction models combining multiple metabolic ratios with or without clinical data may prove to be even more effective decision-making tools and result in a reduction in the number of patients subjected to unnecessary invasive procedures or treatment.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  P.C. Sundgren MR Spectroscopy in Radiation Injury AJNR Am. J. Neuroradiol., September 1, 2009; 30(8): 1469 - 1476. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
