Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

DOI: 10.1148/radiol.2473070571
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Supplemental Table
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Emblem, K. E.
Right arrow Articles by Bjornerud, A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Emblem, K. E.
Right arrow Articles by Bjornerud, A.
(Radiology 2008;247:808-817.)
© RSNA, 2008

Neuroradiology

Glioma Grading by Using Histogram Analysis of Blood Volume Heterogeneity from MR-derived Cerebral Blood Volume Maps1

Kyrre E. Emblem, MSc, Baard Nedregaard, MD, Terje Nome, MD, Paulina Due-Tonnessen, MD, John K. Hald, MD, PhD, David Scheie, MD, Olivera Casar Borota, MD, Milada Cvancarova, MSc, and Atle Bjornerud, PhD

1 From the Departments of Medical Physics (K.E.E., A.B.), Neuroradiology (B.N., T.N., P.D., J.K.H.), Pathology (D.S., O.C.B.), and Biostatistics (M.C.), and the Intervention Center (K.E.E.), Rikshospitalet-Radiumhospitalet Medical Centre, Sognsvannsveien 20, N-0027 Oslo, Norway; and Department of Physics, University of Oslo, Oslo, Norway (A.B.). Received March 28, 2007; revision requested May 23; revision received July 2; accepted August 1; final version accepted November 13. Address correspondence to K.E.E. (e-mail: kyrre.eeg.emblem{at}rikshospitalet.no).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Purpose: To retrospectively compare the diagnostic accuracy of an alternative method used to grade gliomas that is based on histogram analysis of normalized cerebral blood volume (CBV) values from the entire tumor volume (obtained with the histogram method) with that of the hot-spot method, with histologic analysis as the reference standard.

Materials and Methods: The medical ethics committee approved this study, and all patients provided informed consent. Fifty-three patients (24 female, 29 male; mean age, 48 years; age range, 14–76 years) with histologically confirmed gliomas were examined with dynamic contrast material–enhanced 1.5-T magnetic resonance (MR) imaging. CBV maps were created and normalized to unaffected white matter (normalized CBV maps). Four neuroradiologists independently measured the distribution of whole-tumor normalized CBVs and analyzed this distribution by classifying the values into area-normalized bins. Glioma grading was performed by assessing the normalized peak height of the histogram distributions. Logistic regression analysis and interobserver agreement were used to compare the proposed method with a hot-spot method in which only the maximum normalized CBV was used.

Results: For the histogram method, diagnostic accuracy was independent of the observer. Interobserver agreement was almost perfect for the histogram method ({kappa} = 0.923) and moderate for the hot-spot method ({kappa} = 0.559). For all observers, sensitivity was higher with the histogram method (90%) than with the hot-spot method (55%–76%).

Conclusion: Glioma grading based on histogram analysis of normalized CBV heterogeneity is an alternative to the established hot-spot method, as it offers increased diagnostic accuracy and interobserver agreement.

Supplemental material: http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1

© RSNA, 2008


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Magnetic resonance (MR) imaging is the imaging method of choice for characterization of brain tumors prior to treatment. Although conventional contrast material–enhanced MR imaging may indicate the degree of tumor malignancy, studies have shown that the degree of contrast enhancement is not a reliable indicator of the tumor grade (1,2). Consequently, authors have suggested that contrast-enhanced dynamic perfusion imaging can improve the accuracy of MR-based glioma grading (3). Perfusion MR imaging involves the use of first-pass bolus-tracking analysis to derive relative cerebral blood volume (CBV) maps, and studies have shown that the maximal relative CBV of gliomas correlates with the glioma grade (4–7).

Differentiation of high-grade gliomas (HGGs) and low-grade gliomas (LGGs) with MR-derived relative CBV maps is based on measurement of the ratio between the most elevated relative CBV area within the glioma (ie, the hot-spot method) and the relative CBV of unaffected tissue. This value is often referred to as the normalized CBV, and HGGs tend to have higher normalized CBVs than do LGGs (7). It should be noted, however, that this approach has some inherent limitations. First, the selection of a glioma hot spot is highly user dependent because differentiation between vessels and the tumor region of true blood volume elevation can be challenging and a source of error. Second, since only a few image pixels are typically used to determine the relative CBV hot spot, the method is inherently sensitive to image noise and other sources of spurious pixel values (eg, spikes introduced by the algorithms used to generate the normalized CBV maps). Third, unaffected white matter relative CBV is generally used to derive the normalized CBV. This is based on the assumption that most gliomas are located in the white matter. However, incorrect selection of reference relative CBVs might result in either under- or overestimation of normalized CBVs. Fourth, oligodendrogliomas tend to have high normalized CBVs regardless of the glioma grade (1). As a result, cutoff normalized CBVs between HGG and LGG might be harder to establish if oligodendrogliomas are included.

In view of these facts, the purpose of our study was to retrospectively compare the diagnostic accuracy of an alternative method used to grade gliomas that is based on histogram analysis of normalized CBVs from the entire tumor volume (obtained with the histogram method) with that of the hot-spot method, with histologic analysis as the reference standard.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Patient Selection
One author (A.B.) is a consultant for Nordic Imaging Lab (Bergen, Norway). Authors without a financial interest controlled data and information that could have caused a conflict of interest. The regional medical ethics committee approved this study, and patients were included only if they provided written informed consent. Between June 2005 and March 2007, primary glioma was diagnosed at histologic analysis in 75 patients after MR perfusion imaging and subsequent surgery. Fifty-three of these patients (24 female, 29 male; mean age, 48 years; age range, 14–76 years) allowed us to use their data in our study. Two experienced neuropathologists (D.S., O.C.B.) performed histologic evaluation based on examination of tissue obtained via resection (n = 42) or stereotactic image-guided biopsy (n = 11), with use of the World Health Organization classification of central nervous system tumors (8).

Observers
Four experienced neuroradiologists (B.N., T.N., P.D., J.K.H.) with 4–5 years of experience with brain perfusion MR imaging independently performed all measurements. Patient-related information was removed from all images, and observers were blinded to the histopathologic diagnosis.

MR Imaging and Postprocessing
Imaging had been performed at 1.5 T (Sonata, Symphony, or Avanto; Siemens, Erlangen, Germany) with an eight-channel (Sonata or Symphony imagers) or 12-channel (Avanto imager) head coil. The imaging protocol included an axial T2-weighted fast spin-echo sequence (repetition time msec/echo time msec, 4000/104) and an axial T1-weighted spin-echo sequence (500/7.7) performed before and after intravenous administration of gadobutrol (Gadovist; Schering, Berlin, Germany). The voxel size was 0.45 x 0.45 x 5 mm, with 19 sections acquired with both sequences.

Dynamic contrast-enhanced perfusion MR imaging was performed with gradient-echo echo-planar imaging during contrast agent administration. The imaging parameters were as follows: 1430/46 and 1345 Hz/pixel bandwidth for acquisition of 12 axial sections and 1720/48 and 1500 Hz/pixel bandwidth for acquisition of 14 axial sections. We also used a 230 x 230-mm field of view, 1.80 x 1.80 x 5-mm voxel size, and 1.5-mm intersection gap in these examinations. For each section, 50 images were obtained at intervals equal to the repetition time. After eight to 10 time points, 0.2 mmol of gadobutrol per kilogram of body weight was injected at a rate of 5 mL/sec and immediately followed by a 20-mL bolus of saline (B. Braun Melsungen, Melsungen, Germany) injected at a rate of 5 mL/sec.

The images were postprocessed with a dedicated software package (Nordic ICE; Nordic Imaging Lab). The relative CBV (measured in milliliters per 100 g) maps were generated by using established tracer kinetic models applied to the first-pass data (9,10). To reduce the effects of recirculation, the {Delta}R2* (change in 1/T2*) curves were fitted to a gamma-variate function, which is an approximation of the first-pass response as it would appear in the absence of recirculation or leakage. Although potentially more rigorous correction methods exist (11), the gamma-variate approach was used to conform to the reference method described later in this article. Normalized CBV maps were calculated on a pixel-by-pixel basis by dividing every relative CBV value by a contralateral unaffected white matter relative CBV value defined by a neuroradiologist (B.N.) (12). The normalized CBV maps were displayed as color overlays on the structural images. Coregistration between the conventional MR images and the normalized CBV maps was performed on the basis of geometric information stored in the respective data sets (13).

Image Analysis
The four observers performed image analysis independently over a 3-month period. Two observers (B.N., T.N.) reviewed conventional MR findings for each patient. As described in previous studies (11,14), regions of interest that contained the complete tumor were drawn in each section according to the combined overlay and underlay information, with care taken to avoid areas of necrosis, cysts, or nontumor macrovessels evident on the postcontrast T1-weighted images (Fig 1). High-signal-intensity areas thought to represent tumor tissue on the T2-weighted images were used to define the outermost tumor margin, and areas of contrast enhancement seen on the postcontrast T1-weighted images were always included. The observers recorded the time needed to perform the analysis and evaluated how difficult the methods were to perform (easy, intermediate, or difficult).


Figure 1A
View larger version (121K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1a: MR images of grade II diffuse astrocytoma in patient 44 (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show how normalized CBV overlay maps are used to identify vessels within the tumor region. (a) Axial normalized CBV map. (b) Coregistered normalized CBV map overlaid on an axial T2-weighted fast spin-echo image (4000/104). (c) Axial T2-weighted fast spin-echo image (4000/104). (d) Axial postcontrast T1-weighted spin-echo image (500/7.7). In b, the arrow indicates a potential hot-spot area, as seen on the normalized CBV map. However, the underlying vessel-like structure identified in both c and d might indicate that this is not a hot spot.

 

Figure 1B
View larger version (128K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1b: MR images of grade II diffuse astrocytoma in patient 44 (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show how normalized CBV overlay maps are used to identify vessels within the tumor region. (a) Axial normalized CBV map. (b) Coregistered normalized CBV map overlaid on an axial T2-weighted fast spin-echo image (4000/104). (c) Axial T2-weighted fast spin-echo image (4000/104). (d) Axial postcontrast T1-weighted spin-echo image (500/7.7). In b, the arrow indicates a potential hot-spot area, as seen on the normalized CBV map. However, the underlying vessel-like structure identified in both c and d might indicate that this is not a hot spot.

 

Figure 1C
View larger version (122K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1c: MR images of grade II diffuse astrocytoma in patient 44 (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show how normalized CBV overlay maps are used to identify vessels within the tumor region. (a) Axial normalized CBV map. (b) Coregistered normalized CBV map overlaid on an axial T2-weighted fast spin-echo image (4000/104). (c) Axial T2-weighted fast spin-echo image (4000/104). (d) Axial postcontrast T1-weighted spin-echo image (500/7.7). In b, the arrow indicates a potential hot-spot area, as seen on the normalized CBV map. However, the underlying vessel-like structure identified in both c and d might indicate that this is not a hot spot.

 

Figure 1D
View larger version (104K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1d: MR images of grade II diffuse astrocytoma in patient 44 (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show how normalized CBV overlay maps are used to identify vessels within the tumor region. (a) Axial normalized CBV map. (b) Coregistered normalized CBV map overlaid on an axial T2-weighted fast spin-echo image (4000/104). (c) Axial T2-weighted fast spin-echo image (4000/104). (d) Axial postcontrast T1-weighted spin-echo image (500/7.7). In b, the arrow indicates a potential hot-spot area, as seen on the normalized CBV map. However, the underlying vessel-like structure identified in both c and d might indicate that this is not a hot spot.

 
Histogram analysis was performed as follows: Histograms were generated by classifying the normalized CBVs in each region of interest into a predefined number of bins (one to 1000 bins). The area under the resulting histogram curve was normalized to the value of one. The range of normalized CBVs along the x-axis was kept constant (between zero and 20). Glioma malignancy was assessed by measuring the maximum normalized peak height of distribution (ie, the relative frequency of normalized CBVs in a given histogram bin), with the hypothesis that normalized CBV heterogeneity is related to tumor malignancy and is inversely proportional to the peak height of the normalized CBV distribution. In the reference-standard hot-spot method (12), which was shown to have the highest intra- and interobserver reproducibility among a number of reported hot-spot methods, each observer selected a minimum of four regions of interest that were believed to represent high normalized CBV regions, and the maximum value was used. The size of the tumor regions of interest remained constant (radius, 1.8 mm). In the case of multiple lesions, the largest lesion was chosen. For the hot-spot method, mean normalized CBV values and standard deviations were recorded for LGG and HGG. For the histogram method, mean histogram peak heights and standard deviations were recorded.

Statistical Analysis
The diagnostic accuracy of the two methods was evaluated by using binary logistic regression to derive sensitivity, specificity, and positive and negative predictive values for LGG versus HGG. A glioma classified as an HGG or an LGG with both observer data and histologic analysis was considered a true-positive finding or a true-negative finding, respectively. As described previously, optimal cutoff values between LGG and HGG for each observer were obtained by minimizing the number of glioma grade misclassifications and maximizing the average sensitivity and specificity (7,15,19). To compare our results with the results of previous studies, the diagnostic accuracy of the hot-spot method was also calculated by using a previously published cutoff normalized CBV of 1.75 (15). The sensitivity and specificity for each observer were compared by using the McNemar test and a pairwise comparison of the area under the receiver operating characteristic curve (Az). The number of histogram bins that yielded the highest Az was derived by using an in-house developed Matlab routine (R2006a; MathWorks, Natick, Mass) that was used to calculate the Az values for all bin numbers between one and 1000.

Mann-Whitney tests were used to assess the ability to differentiate (a) between grade II oligodendroglial tumors (oligodendrogliomas or oligoastrocytomas) and grade II diffuse astrocytomas and (b) between grade III gliomas and grade IV gliomas with the two methods. Mann-Whitney tests were also used to assess whether excluding grade II oligodendroglial tumors affected glioma grading. To account for multiple-comparison effects, a significance level of P = .01 was used. This value was obtained by dividing a default P value of .05 by the number of observers and applying the Bonferroni correction.

Interobserver reproducibility between the four observers was assessed by using Fleiss {kappa} statistics based on whether the observers graded a glioma as HGG or LGG. A {kappa} value of less than zero indicated poor agreement; a {kappa} value of 0.00–0.20, slight agreement; a {kappa} value of 0.21–0.40, fair agreement; a {kappa} value of 0.41–0.60, moderate agreement; a {kappa} value of 0.61–0.80, substantial agreement; and a {kappa} value of 0.81–1.00, almost perfect agreement (16). Statistical analysis was performed by using SPSS13 (SPSS, Chicago, Ill).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
Of the 53 gliomas investigated, 24 were histologically confirmed to be LGGs (World Health Organization grade I or II) and 29 were histologically confirmed to be HGGs (World Health Organization grade III or IV) (Fig 2) (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Signs of necrosis were seen on conventional MR images in six patients with LGGs and 17 with HGGs (one with grade III glioma and 16 with grade IV glioma). On average, the four observers reported examination times of 7 minutes per patient when using the hot-spot method and 11 minutes per patient when using the histogram method. All observers reported that the two methods were equally difficult to perform (intermediate difficulty).


Figure 2
View larger version (10K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2: Flowchart shows 75 eligible patients received a histologic diagnosis of primary glioma after MR perfusion imaging and subsequent surgery over a 21-month period (June 2005 to March 2007). Only patients who agreed to participate in the study were included in the analysis.

 
Glioma Grading
Optimal normalized CBV cutoff values between LGG and HGG ranged from 3.75 to 5.58 mL/100 g (Table 1). Optimal histogram peak values between LGG and HGG ranged from 0.10 to 0.12 mL/100 g. One observer did not observe a significant difference between LGG and HGG with the hot-spot method. Excluding 13 grade II oligodendroglial tumors from glioma grading led to reduced P values for all observers when they used the hot-spot method, whereas P values obtained with the histogram method remained unchanged. One observer (observer 1) was able to differentiate (P < .002) between grade III (n = 10) and grade IV (n = 19) gliomas with the histogram method (Fig 3). Neither method enabled us to differentiate between grade II oligodendroglial tumors (n = 13) and grade II diffuse astrocytomas (n = 8).


View this table:
[in this window]
[in a new window]

  		Table 1. Normalized CBV Values and Optimal Cutoffs between LGG and HGG for Hot-Spot and Histogram Methods

 

Figure 3A
View larger version (113K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3a: (a–d) Normalized CBV maps overlaid on axial T2-weighted fast spin-echo MR images (4000/104) in patients with (a) grade I pilocytic astrocytoma, (b) grade II diffuse astrocytoma, (c) grade III anaplastic astrocytoma, and (d) grade IV glioblastoma. The patients are subjects 28, 2, 18 and 7, respectively (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Note the middle cerebral artery encased by the tumor volume in b. (e) The corresponding histogram signatures derived from the total tumor volume of these patients is shown. The histogram signatures for each patient were derived by using the mean normalized CBVs (nCBV) obtained by all four observers. Note the low maximum peak height and wide distribution in c and d compared with that in a and b.

 

Figure 3B
View larger version (114K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3b: (a–d) Normalized CBV maps overlaid on axial T2-weighted fast spin-echo MR images (4000/104) in patients with (a) grade I pilocytic astrocytoma, (b) grade II diffuse astrocytoma, (c) grade III anaplastic astrocytoma, and (d) grade IV glioblastoma. The patients are subjects 28, 2, 18 and 7, respectively (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Note the middle cerebral artery encased by the tumor volume in b. (e) The corresponding histogram signatures derived from the total tumor volume of these patients is shown. The histogram signatures for each patient were derived by using the mean normalized CBVs (nCBV) obtained by all four observers. Note the low maximum peak height and wide distribution in c and d compared with that in a and b.

 

Figure 3C
View larger version (113K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3c: (a–d) Normalized CBV maps overlaid on axial T2-weighted fast spin-echo MR images (4000/104) in patients with (a) grade I pilocytic astrocytoma, (b) grade II diffuse astrocytoma, (c) grade III anaplastic astrocytoma, and (d) grade IV glioblastoma. The patients are subjects 28, 2, 18 and 7, respectively (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Note the middle cerebral artery encased by the tumor volume in b. (e) The corresponding histogram signatures derived from the total tumor volume of these patients is shown. The histogram signatures for each patient were derived by using the mean normalized CBVs (nCBV) obtained by all four observers. Note the low maximum peak height and wide distribution in c and d compared with that in a and b.

 

Figure 3D
View larger version (115K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3d: (a–d) Normalized CBV maps overlaid on axial T2-weighted fast spin-echo MR images (4000/104) in patients with (a) grade I pilocytic astrocytoma, (b) grade II diffuse astrocytoma, (c) grade III anaplastic astrocytoma, and (d) grade IV glioblastoma. The patients are subjects 28, 2, 18 and 7, respectively (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Note the middle cerebral artery encased by the tumor volume in b. (e) The corresponding histogram signatures derived from the total tumor volume of these patients is shown. The histogram signatures for each patient were derived by using the mean normalized CBVs (nCBV) obtained by all four observers. Note the low maximum peak height and wide distribution in c and d compared with that in a and b.

 

Figure 3E
View larger version (25K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 3e: (a–d) Normalized CBV maps overlaid on axial T2-weighted fast spin-echo MR images (4000/104) in patients with (a) grade I pilocytic astrocytoma, (b) grade II diffuse astrocytoma, (c) grade III anaplastic astrocytoma, and (d) grade IV glioblastoma. The patients are subjects 28, 2, 18 and 7, respectively (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1). Note the middle cerebral artery encased by the tumor volume in b. (e) The corresponding histogram signatures derived from the total tumor volume of these patients is shown. The histogram signatures for each patient were derived by using the mean normalized CBVs (nCBV) obtained by all four observers. Note the low maximum peak height and wide distribution in c and d compared with that in a and b.

 
For the histogram method, diagnostic accuracy was observer independent (Table 2). For all observers, sensitivity (90%, 26 of 29 HGG patients) and negative predictive value (87%, 20 of 23 patients) increased with use of the histogram method compared with the sensitivity (55%–76%, 16–22 of 29 patients) and the negative predictive value (54%–74%, 15 of 28 patients to 20 of 27 patients) obtained with the hot-spot method. For observer 2 (Table 2), sensitivity values obtained with the histogram method were significantly different from those obtained with the hot-spot method (McNemar test, P = .002). Specificity was 83% (20 of 24 patients with LGG) with the histogram method and 63%–88% (13–20 of 24 patients with LGG) with the hot-spot method. The McNemar tests did not reveal a significant difference between specificity values obtained with the histogram method and those obtained with the hot-spot method. The positive predictive value was 87% (26 of 30 patients) with the histogram method and 64%–88% (16 of 25 patients to 22 of 26 patients) with the hot-spot method. Compared with the optimal hot-spot cutoff value, the 1.75 mL/100 g cutoff value resulted in improved sensitivity (97%–100%, 28–29 of 29 patients with HGG) in all observers and improved negative predictive value in three of four observers (67%–100%, between two and seven of seven patients). Both the specificity (8%–29%, between two and seven of 24 patients with LGG) and the positive predictive value (56%–63%, between 28 of 50 and 29 of 46 patients) were reduced in all observers.


View this table:
[in this window]
[in a new window]

  		Table 2. Diagnostic Accuracy of Hot-Spot and Histogram Methods

 
The mean Az values (± standard errors) were larger for all observers when they used the histogram method (range, 0.905 ± 0.041 to 0.914 ± 0.039) than when they used the hot-spot method (range, 0.698 ± 0.072 to 0.867 ± 0.055). For one observer, Az was significantly higher when the histogram method was used than when the hot-spot method was used (P < .001). Averaged over the four observers, the maximum Az was found at 108 histogram bins (mean Az = 0.909 ± 0.004) (Figs 4, 5).


Figure 4
View larger version (12K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 4: Curve shows the mean Az for all four observers. The histogram method yielded larger mean Az values than did the hot-spot method (0.801 ± 0.063, straight line), regardless of the bin number. Averaged over the four observers, the maximum Az value was found at 108 histogram bins (0.909 ± 0.004).

 

Figure 5A
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5a: Receiver operating characteristic curves for the histogram method obtained by using 108 histogram bins (solid line) and the hot-spot method (dotted line) for observers (a) 1, (b) 2, (c) 3, and (d) 4. For all observers, the mean Az values obtained with the histogram method were larger than those obtained with the hot-spot method. In a, the mean Az obtained with the histogram method was 0.905 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.822 ± 0.062. In b, the mean Az obtained with the histogram method was 0.909 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.698 ± 0.072. In c, the mean Az obtained with the histogram method was 0.909 ± 0.040, whereas the mean Az obtained with the hot-spot method was 0.818 ± 0.061. In d, the mean Az obtained with the histogram method was 0.914 ± 0.039, whereas the mean Az obtained with the hot-spot method was 0.867 ± 0.055.

 

Figure 5B
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5b: Receiver operating characteristic curves for the histogram method obtained by using 108 histogram bins (solid line) and the hot-spot method (dotted line) for observers (a) 1, (b) 2, (c) 3, and (d) 4. For all observers, the mean Az values obtained with the histogram method were larger than those obtained with the hot-spot method. In a, the mean Az obtained with the histogram method was 0.905 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.822 ± 0.062. In b, the mean Az obtained with the histogram method was 0.909 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.698 ± 0.072. In c, the mean Az obtained with the histogram method was 0.909 ± 0.040, whereas the mean Az obtained with the hot-spot method was 0.818 ± 0.061. In d, the mean Az obtained with the histogram method was 0.914 ± 0.039, whereas the mean Az obtained with the hot-spot method was 0.867 ± 0.055.

 

Figure 5C
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5c: Receiver operating characteristic curves for the histogram method obtained by using 108 histogram bins (solid line) and the hot-spot method (dotted line) for observers (a) 1, (b) 2, (c) 3, and (d) 4. For all observers, the mean Az values obtained with the histogram method were larger than those obtained with the hot-spot method. In a, the mean Az obtained with the histogram method was 0.905 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.822 ± 0.062. In b, the mean Az obtained with the histogram method was 0.909 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.698 ± 0.072. In c, the mean Az obtained with the histogram method was 0.909 ± 0.040, whereas the mean Az obtained with the hot-spot method was 0.818 ± 0.061. In d, the mean Az obtained with the histogram method was 0.914 ± 0.039, whereas the mean Az obtained with the hot-spot method was 0.867 ± 0.055.

 

Figure 5D
View larger version (9K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 5d: Receiver operating characteristic curves for the histogram method obtained by using 108 histogram bins (solid line) and the hot-spot method (dotted line) for observers (a) 1, (b) 2, (c) 3, and (d) 4. For all observers, the mean Az values obtained with the histogram method were larger than those obtained with the hot-spot method. In a, the mean Az obtained with the histogram method was 0.905 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.822 ± 0.062. In b, the mean Az obtained with the histogram method was 0.909 ± 0.041, whereas the mean Az obtained with the hot-spot method was 0.698 ± 0.072. In c, the mean Az obtained with the histogram method was 0.909 ± 0.040, whereas the mean Az obtained with the hot-spot method was 0.818 ± 0.061. In d, the mean Az obtained with the histogram method was 0.914 ± 0.039, whereas the mean Az obtained with the hot-spot method was 0.867 ± 0.055.

 
Interobserver Reproducibility
For the hot-spot method, there was moderate interobserver agreement between the four observers with use of the optimal cutoff value ({kappa} = .559) and the 1.75 mL/100 g cutoff value ({kappa} = .459). Although the size and shape of the resulting tumor regions of interest varied between the four observers (Fig 6), the interobserver agreement between the four observers was almost perfect ({kappa} = .923) when the histogram method was used.


Figure 6A
View larger version (105K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 6a: MR images obtained in a patient with grade II oligodendroglioma (patient 9) (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show the manual glioma volume delineation for two observers (red and white regions of interest). (a) Coregistered relative CBV map overlaid on a T2-weighted fast spin-echo image (4000/104). (b) T2-weighted fast spin-echo image (4000/104). Although the variation between observers is evident, resulting histograms correctly depicted LGG in both cases. Almost perfect interobserver agreement ({kappa} = .923) obtained with the histogram method suggests that the variations between observers caused by imperfect tumor delineation are relatively unimportant, given the large number of data points included in the histogram.

 

Figure 6B
View larger version (68K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 6b: MR images obtained in a patient with grade II oligodendroglioma (patient 9) (Table E1, http://radiology.rsnajnls.org/cgi/content/full/247/3/808/DC1) show the manual glioma volume delineation for two observers (red and white regions of interest). (a) Coregistered relative CBV map overlaid on a T2-weighted fast spin-echo image (4000/104). (b) T2-weighted fast spin-echo image (4000/104). Although the variation between observers is evident, resulting histograms correctly depicted LGG in both cases. Almost perfect interobserver agreement ({kappa} = .923) obtained with the histogram method suggests that the variations between observers caused by imperfect tumor delineation are relatively unimportant, given the large number of data points included in the histogram.

 

   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 
In our study, we evaluated an alternative method with which to differentiate HGG from LGG on the basis of the normalized CBV heterogeneity of the entire tumor volume. Our results suggest that the histogram method has higher interobserver agreement and yields higher sensitivity and negative predictive values and equal specificity when compared with the hot-spot method. The influence of increased diagnostic accuracy on clinical outcome is difficult to establish and was not investigated in our study. However, high diagnostic accuracy combined with high interobserver reproducibility are critical criteria for any diagnostic test (17). One potential advantage of the histogram method is that the results are independent of the choice of reference tissue, as long as the reference is kept the same throughout the cohort. For example, the effect of changing reference tissue from white to gray matter is simply a shift in the position of the peak distribution bin; the actual peak value does not change. However, the hot-spot method is critically dependent on correct selection of reference tissue since determination of normalized CBV is based solely on this parameter. Arguably, the objective of an optimal grading method should be identification of the most malignant part of the tumor, which should favor the hot-spot method. However, our results suggest that observers are not able to consistently identify the most malignant tumor region with current hot-spot methods and that there is a consequent loss in sensitivity or specificity depending on the cutoff value used.

The results we obtained with the hot-spot method are consistent with previously published data (6,15,18,19). However, the optimal cutoff values between HGG and LGG in our study (3.75–5.58 mL/100 g) were higher than those in previous studies (1.5–1.98 mL/100 g) (7,15,19). In a study in which the hot-spot method was used in 160 patients, authors reported a sensitivity of 95.0% and a specificity of 57.5% when they used a CBV cutoff of 1.75 mL/100 g (15). When we applied a normalized CBV cutoff of 1.75 mL/100 g to our data, we obtained similar values but with wide confidence intervals. Hence, our results suggest that to obtain maximum diagnostic accuracy, the choice of an optimal cutoff value should be based on perfusion data generated at a given site. One further observation in the current work was the trade-off between high sensitivity and high specificity with both methods investigated. Lowering the cutoff values increased sensitivity at the cost of reduced specificity. It could be argued that a low false-negative rate is more important than a low false-positive rate because of the serious consequences of false-negative findings. However, both types of errors are potentially critical, given the different treatment strategies for LGG and HGG (20).

The only difference between grade III gliomas and grade IV gliomas was seen by one observer who used the histogram method. In previous studies in which the hot-spot method was used in 26 (6) and 120 (15) patients with HGGs, no difference between these groups was reported. Furthermore, as described previously (21), necrosis was a specific marker for distinguishing grade III gliomas from grade IV gliomas, but it was not a sensitive one.

We were unable to differentiate between grade II astrocytomas and grade II oligodendroglial tumors with either method. Excluding grade II oligodendroglial tumors affected glioma grading only when the hot-spot method was used; this finding suggests that the diagnostic accuracy of the hot-spot method is more dependent on the number of patients than is the histogram method. When we compared the use of different histogram bin numbers, the maximum Az was found at 108 bins. The reduced diagnostic accuracy at lower bin numbers can be explained by the large range of relative CBVs contained in each of the resulting bins, which tends to mask small hypervascular regions in patients with HGGs. Also, the reduced diagnostic accuracy at higher bin numbers may be explained by increasing noise in the resulting histogram, since each bin contains fewer pixel averages.

In our study, we used gradient-echo echo-planar imaging rather than spin-echo echo-planar imaging because a higher temporal resolution can be achieved with this sequence. Also, previous studies have shown a stronger correlation between tumor grade and CBV with use of gradient-echo techniques (14,22,23). Gradient-echo echo-planar imaging sequences have also been shown to be more sensitive to macrovascular structures, aiding in the differentiation between infiltrating vessels and true tumor CBV elevation (24).

Our study had limitations. It would have been preferable to include more patients to strengthen the statistical power. However, the number of patients included in our study (n = 53) is similar to that in other studies on glioma grading (3,5–7,12–14,18,19,23). Furthermore, only one observer obtained a significant difference between the sensitivity values of the two methods. However, the mean Az values were higher for all observers when they used the histogram method compared with when they used the hot-spot method. Therefore, our data suggest that the histogram method has higher diagnostic accuracy. A further limitation of both methods is the need for an optimal coregistration between normalized CBV maps and conventional MR images. Hence, the increasing availability of intramodal image coregistration methods in clinical image software will be of benefit for the clinical utility of both methods. An obvious challenge with the histogram method is that of identifying the appropriate tumor region. Optimal operational definition of tumor volume is complicated because gliomas are infiltrating tumors with indistinct borders beyond the radiologic margins (25,26). However, the high interobserver agreement of the histogram method suggests that variations between observers caused by imperfect tumor delineation are relatively unimportant, given the large number of data points included in the histogram. The observers reported that they spent more time per patient for the histogram method than for the hot-spot method because the tumor volume had to be identified in every section. However, with both methods, the observer had to exclude vessels that infiltrated the glioma region, and, consequently, the methods were considered equally difficult to perform. On the basis of these observations, there was a clear need for more user-independent and automated methods with which to identify total tumor volume and regions representing unaffected reference tissue. Studies have shown that cluster analysis techniques can be used to quantify and classify similar tissue components on MR images (27), and such methods are currently being implemented as part of our histogram analysis software.

Histogram analysis was performed by assessing the peak height of the normalized histogram distribution of normalized CBVs in the tumor. This approach was chosen because the resulting height was directly determined on the basis of the underlying heterogeneity of the normalized CBV distribution. It is hypothesized that histogram-based analysis can be further improved with parametric analysis of the histogram shape rather than just the peak value.

In conclusion, our results suggest that the proposed histogram method is a diagnostically accurate and reproducible method with which to grade gliomas on the basis of MR-derived blood volume maps. Compared with the hot-spot method, the histogram method had higher interobserver agreement, sensitivity, and negative predictive value and equal specificity. Future developments in cluster methods for automated segmentation of tumor volume may further enhance the clinical utility of this method.


   ADVANCES IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
IMPLICATION FOR PATIENT CARE
 References
 


   IMPLICATION FOR PATIENT CARE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
References
 


   FOOTNOTES
 

Abbreviations: Az = area under the receiver operating characteristic curve • CBV = cerebral blood volume • HGG = high-grade glioma • LGG = low-grade glioma

Author contributions: Guarantor of integrity of entire study, K.E.E.; 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, K.E.E., B.N., T.N., P.D., J.K.H., D.S., O.C.B., A.B.; clinical studies, K.E.E., B.N., T.N., P.D., J.K.H., D.S., O.C.B., A.B.; statistical analysis, K.E.E., M.C., A.B.; and manuscript editing, all authors

See Materials and Methods for pertinent disclosures.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 IMPLICATION FOR PATIENT CARE
 References
 

  1. Lev MH, Rosen BR. Clinical applications of intracranial perfusion MR imaging. Neuroimaging Clin N Am 1999;9(2):309–331.[Medline]
  2. Law M, Oh S, Babb JS, et al. Low-grade gliomas: dynamic susceptibility-weighted contrast-enhanced perfusion MR imaging—prediction of patient clinical response. Radiology 2006;238(2):658–667.[Abstract/Free Full Text]
  3. Covarrubias DJ, Rosen BR, Lev MH. Dynamic magnetic resonance perfusion imaging of brain tumors. Oncologist 2004;9(5):528–537.[Abstract/Free Full Text]
  4. Edelman RR, Mattle HP, Atkinson DJ, et al. Cerebral blood flow: assessment with dynamic contrast-enhanced T2*-weighted MR imaging at 1.5 T. Radiology 1990;176(1):211–220.[Abstract/Free Full Text]
  5. Aronen HJ, Gazit IE, Louis DN, et al. Cerebral blood volume maps of gliomas: comparison with tumor grade and histologic findings. Radiology 1994;191(1):41–51.[Abstract/Free Full Text]
  6. Knopp EA, Cha S, Johnson G, et al. Glial neoplasms: dynamic contrast-enhanced T2*-weighted MR imaging. Radiology 1999;211(3):791–798.[Abstract/Free Full Text]
  7. Lev MH, Ozsunar Y, Henson JW, et al. Glial tumor grading and outcome prediction using dynamic spin-echo MR susceptibility mapping compared with conventional contrast-enhanced MR: confounding effect of elevated rCBV of oligodendrogliomas. AJNR Am J Neuroradiol 2004;25(2):214–221.[Abstract/Free Full Text]
  8. Kleihues P, Cavenee WK. Astrocytic tumors and oligodendroglial tumors and mixed gliomas. In: Kleihues P, Cavenee WK, eds. The WHO classification of tumors of the nervous system. Lyon, France: International Agency for Research on Cancer, 2000; 9–70.
  9. Rosen BR, Belliveau JW, Vevea JM, Brady TJ. Perfusion imaging with NMR contrast agents. Magn Reson Med 1990;14(2):249–265.[Medline]
  10. Ostergaard L, Weisskoff RM, Chesler DA, Gyldensted C, Rosen BR. High resolution measurement of cerebral blood flow using intravascular tracer bolus passages. I. Mathematical approach and statistical analysis. Magn Reson Med 1996;36(5):715–725.[Medline]
  11. Boxerman JL, Schmainda KM, Weisskoff RM. Relative cerebral blood volume maps corrected for contrast agent extravasation significantly correlate with glioma tumor grade, whereas uncorrected maps do not. AJNR Am J Neuroradiol 2006;27(4):859–867.[Abstract/Free Full Text]
  12. Wetzel SG, Cha S, Johnson G, et al. Relative cerebral blood volume measurements in intracranial mass lesions: interobserver and intraobserver reproducibility study. Radiology 2002;224(3):797–803.[Abstract/Free Full Text]
  13. Bjornerud A. The ICE software package: direct co-registration of anatomical and functional datasets using DICOM image geometry information. Proc Hum Brain Mapping 2003;19(2):1018p.
  14. Schmainda KM, Rand SD, Joseph AM, et al. Characterization of a first-pass gradient-echo spin-echo method to predict brain tumor grade and angiogenesis. AJNR Am J Neuroradiol 2004;25(9):1524–1532.[Abstract/Free Full Text]
  15. Law M, Yang S, Wang H, et al. Glioma grading: sensitivity, specificity, and predictive values of perfusion MR imaging and proton MR spectroscopic imaging compared with conventional MR imaging. AJNR Am J Neuroradiol 2003;24(10):1989–1998.[Abstract/Free Full Text]
  16. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977;33(1):159–174.[CrossRef][Medline]
  17. Fryback DG, Thornbury JR. The efficacy of diagnostic imaging. Med Decis Making 1991;11(2):88–94.[Abstract/Free Full Text]
  18. Sugahara T, Korogi Y, Kochi M, et al. Correlation of MR imaging-determined cerebral blood volume maps with histologic and angiographic determination of vascularity of gliomas. AJR Am J Roentgenol 1998;171(6):1479–1486.[Abstract/Free Full Text]
  19. Hakyemez B, Erdogan C, Ercan I, Ergin N, Uysal S, Atahan S. High-grade and low-grade gliomas: differentiation by using perfusion MR imaging. Clin Radiol 2005;60(4):493–502.[CrossRef][Medline]
  20. Claus EB, Black PM. Survival rates and patterns of care for patients diagnosed with supratentorial low-grade gliomas: data from the SEER program, 1973–2001. Cancer 2006;106(6):1358–1363.[CrossRef][Medline]
  21. Nelson JS, Tsukada Y, Schoenfeld D, Fulling K, Lamarche J, Peress N. Necrosis as a prognostic criterion in malignant supratentorial, astrocytic gliomas. Cancer 1983;52(3):550–554.[CrossRef][Medline]
  22. Donahue KM, Krouwer HG, Rand SD, et al. Utility of simultaneously acquired gradient-echo and spin-echo cerebral blood volume and morphology maps in brain tumor patients. Magn Reson Med 2000;43(6):845–853.[CrossRef][Medline]
  23. Sugahara T, Korogi Y, Kochi M, Ushio Y, Takahashi M. Perfusion-sensitive MR imaging of gliomas: comparison between gradient-echo and spin-echo echo-planar imaging techniques. AJNR Am J Neuroradiol 2001;22(7):1306–1315.[Abstract/Free Full Text]
  24. Aronen HJ, Perkio J. Dynamic susceptibility contrast MRI of gliomas. Neuroimaging Clin N Am 2002;12(4):501–523.[CrossRef][Medline]
  25. Price SJ, Jena R, Burnet NG, et al. Improved delineation of glioma margins and regions of infiltration with the use of diffusion tensor imaging: an image-guided biopsy study. AJNR Am J Neuroradiol 2006;27(9):1969–1974.[Abstract/Free Full Text]
  26. Grier JT, Batchelor T. Low-grade gliomas in adults. Oncologist 2006;11(6):681–693.[Abstract/Free Full Text]
  27. Itskovich VV, Samber DD, Mani V, et al. Quantification of human atherosclerotic plaques using spatially enhanced cluster analysis of multicontrast-weighted magnetic resonance images. Magn Reson Med 2004;52(3):515–523.[CrossRef][Medline]



This article has been cited by other articles:


Home page
 Am. J. Neuroradiol.Home page
K.E. Emblem, D. Scheie, P. Due-Tonnessen, B. Nedregaard, T. Nome, J.K. Hald, K. Beiske, T.R. Meling, and A. Bjornerud
Histogram Analysis of MR Imaging-Derived Cerebral Blood Volume Maps: Combined Glioma Grading and Identification of Low-Grade Oligodendroglial Subtypes
AJNR Am. J. Neuroradiol., October 1, 2008; 29(9): 1664 - 1670.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Supplemental Table
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Emblem, K. E.
Right arrow Articles by Bjornerud, A.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Emblem, K. E.
Right arrow Articles by Bjornerud, A.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE