Search tips
Search criteria 


Logo of neuroncolAboutAuthor GuidelinesEditorial BoardNeuro-Oncology
Neuro Oncol. 2013 April; 15(4): 442–450.
Published online 2013 February 3. doi:  10.1093/neuonc/nos323
PMCID: PMC3607265

Vascular change measured with independent component analysis of dynamic susceptibility contrast MRI predicts bevacizumab response in high-grade glioma



Standard pre- and postcontrast (T1 + C) anatomical MR imaging is proving to be insufficient for accurately monitoring bevacizumab treatment response in recurrent glioblastoma (GBM). We present a novel imaging biomarker that detects abnormal tumor vasculature exhibiting both arterial and venous perfusion characteristics. We hypothesized that a decrease in the extent of this abnormal vasculature after bevacizumab treatment would predict treatment efficacy and overall survival.


Dynamic susceptibility contrast perfusion MRI was gathered in 43 patients with high-grade glioma. Independent component analysis separated vasculature into arterial and venous components. Voxels with perfusion characteristics of both arteries and veins (ie, arterio-venous overlap [AVOL]) were measured in patients with de novo untreated GBM and patients with recurrent high-grade glioma before and after bevacizumab treatment. Treated patients were separated on the basis of an increase or decrease in AVOL volume (+/−ΔAVOL), and overall survival following bevacizumab onset was then compared between +/−ΔAVOL groups.


AVOL in untreated GBM was significantly higher than in normal vasculature (P < .001). Kaplan-Meier survival curves revealed a greater median survival (348 days) in patients with GBM with a negative ΔAVOL after bevacizumab treatment than in patients with a positive change (197 days; hazard ratio, 2.51; P < .05). Analysis of patients with combined grade III and IV glioma showed similar results, with median survivals of 399 days and 153 days, respectively (hazard ratio, 2.71; P < .01). Changes in T1+C volume and ΔrCBV after treatment were not significantly different across +/−ΔAVOL groups, and ΔAVOL was not significantly correlated with ΔT1+C or ΔrCBV.


The independent component analysis dynamic susceptibility contrast–derived biomarker AVOL adds additional information for determining bevacizumab treatment efficacy.

Keywords: angiogenesis, bevacizumab, brain tumor, DSC, glioblastoma, glioma, perfusion, ICA, independent component analysis, MRI

Glioblastoma (GBM) is one of the most deadly cancers in adults. GBM accounts for 52% of all parenchymal brain tumor cases and 20% of all intracranial tumors.1 The current standard of care, which includes surgery followed by radiation therapy and chemotherapy, is associated with a survival of ~14.6 months.2 The anti–vascular endothelial growth factor (VEGF) antibody bevacizumab was approved by the Food and Drug Administration for the treatment of patients with recurrent GBM in 2009. This neoplastic angiogenesis-targeting drug3 demonstrated an improvement in progression-free survival (PFS) from the 15% historical control rate to 42.6% and 29% in separate phase II clinical trials.4,5

Although bevacizumab is quickly becoming standard treatment for recurrent GBM, it has become increasingly clear that standard pre- and postcontrast anatomical imaging methods, which have been used to measure tumor volumes and, thus, treatment responses, are no longer adequate.6,7 Bevacizumab decreases vessel permeability,8 which results in diminished contrast agent extravasation.9 This decreases the enhancing tumor volume. Unfortunately, this does not necessarily reflect a tumor's true biological response.10 Therefore, alternative imaging methods beyond blood-brain barrier disruption are being explored as more direct indicators of tumoral responses to anti-angiogenic treatments.

In a recent article, the change in T2w signal was shown to be associated with greater survival,11 and another study showed that histogram-based changes in apparent diffusion coefficient (ADC) correlated with response and, subsequently, longer PFS.12 In addition, single-threshold and graded functional diffusion maps derived from serial changes in ADC have shown sensitivity to treatment response.13,14 Additionally, researchers have noticed that the presence of bevacizumab-induced calcification predicts overall survival (OS).15 Although these methods have proven to be effective for predicting response to treatment, they are insensitive to vascular change, which is the mechanistic target of bevacizumab treatment.

Measures of relative cerebral blood volume (rCBV) have been shown to be sensitive to tumor grade,1618 predictive of survival,19 and able to distinguish regions of contrast agent enhancement resulting from treatment effects (particularly postradiation changes) from those due to recurrent or residual tumor.20,21 More recently, MRI perfusion parameters, which include rCBV, contrast agent volume transfer coefficient (Ktrans), and vessel size index, have demonstrated the potential to reflect a condition termed vascular normalization.22,23 Although physiologic angiogenesis, such as that which occurs with wound healing, results in the formation of well-ordered, mature vessels, pathologic angiogenesis results in the formation of chaotic and immature arterioles and venules.24 Vascular normalization is thought to occur when a tumor's chaotic vasculature becomes more ordered and efficient, such as that of normal vasculature, resulting in more efficient delivery of oxygen and cytotoxic drugs to tumor.25 Imaging biomarkers sensitive to these vascular changes may provide an additional measure of biologic response.

In this study, we used independent component analysis (ICA) to measure the temporal characteristics of contrast agent perfusing through brain tumor and normal vasculature. ICA is an emerging technique in functional MRI (fMRI) data processing that takes a data-driven, multivariate approach to categorize voxel time series by examining voxels exhibiting the same temporal response patterns. After its first use in fMRI,26 ICA has been used in many fMRI and EEG applications studying brain activation. ICA applied to dynamic susceptibility contrast (DSC) MRI in patients with brain tumors has shown the ability to separate tumor from normal vasculature and distinguish perfusion patterns in GBM from those in meningioma.27

This study uses ICA to classify voxels on the basis of stage delays of contrast agent perfusion through different vessel types.28 We hypothesized that abnormal vasculature classified in both venous and arterial ICA components (ie, arterio-venous overlap [AVOL]) exists in greater proportions in contrast-enhancing tumor, compared with normal brain vasculature. We additionally hypothesized that vessels of this nature may benefit from treatment with bevacizumab and that effective treatment would be characterized by an overall decrease in the volume of AVOL. To address these hypotheses, we applied ICA to DSC-MRI data collected in untreated patients with GBM (dataset A) to determine the proportion of AVOL in ICA-classified normal and tumor vasculature. An additional mutually exclusive dataset (dataset B) was processed before and after bevacizumab treatment in patients with recurrent grades III and IV glioma to determine whether changes in AVOL are predictive of OS.


Patient Population

Forty-three patients were retrospectively analyzed and included in different aspects of this analysis. Table 1 illustrates the studies performed on 2 separate datasets. Dataset A included 11 patients with de novo GBM for whom the initial imaging study was performed before surgery and pathological diagnosis. Dataset B consisted of 32 patients with high-grade glioma who received treatment with bevacizumab. Data on these patients were collected consecutively at our institution from February 2007 through August 2011. To be included in this analysis, patients were required to have a baseline DSC imaging session before treatment and a follow-up imaging session within 3 months after treatment onset. Patients were excluded from this study if the 2 imaging sessions were acquired at different resolutions or if pathological diagnosis of a high-grade glioma was lacking at the time of treatment onset. Patients were also excluded if one of the ICA components modeled in either imaging session was dominated by motion artifacts. Twenty-three of these patients received a diagnosis of World Health Organization (WHO) grade IV GBM, and 9 had pathologically confirmed WHO grade III glioma before the initiation of bevacizumab treatment. Four of the patients were given bevacizumab up front at diagnosis as part of the bevacizumab for newly diagnosed glioblastoma trial (BINGO, RTOG 0825,, and the remaining patients received bevacizumab treatment at tumor recurrence. The mean time from baseline imaging to bevacizumab treatment initiation was 19 days (min: 0, max: 104 days), and the mean time to follow-up imaging was 38 days (min: 14, max: 72 days) after the first dose of bevacizumab treatment. All patients were continuing to undergo bevacizumab treatment at the time of the follow-up scan. Written informed consent was obtained from each patient under a research protocol approved by our Institutional Review Board.

Table 1.
Study and sample descriptions of datasets A and B


All anatomical images were analyzed retrospectively after routine clinical acquisition. DSC imaging was appended on each patient's clinical imaging session and banked in our institutional brain tumor imaging database. The images were acquired on 1 of 6 clinical MRI scanners at our institution, including three 1.5T GE, one 1.5T Siemens, one 3T GE, and one 3T Siemens (GE Healthcare, Waukesha, WI; Siemens Medical, Malvern, PA). Conventional precontrast T1-weighted images were acquired with vendor provided parameters of echo time (TE)/repetition time (TR) of 10–17 ms/500–666 ms (8.9–10 ms/500–600 ms, 3T), and matrix of 512 × 512, voxel size of 0.47 × 0.47 mm, and flip angle of 75°–90° (75°, 3T). Immediately before dynamic imaging for DSC, a 0.05–0.1 mmol/kg (preload) dose of MultiHance gadobenate dimeglumine (Gd; Bracco Diagnostics Inc., Princeton, NJ) contrast agent was administered based on patient weight, and clinical postcontrast T1-weighted imaging was obtained using the same parameters listed above for the precontrast T1w images. Three patients lacked postcontrast imaging. The preload dose reduces T1w effects resulting from agent extravasation that confounds the rCBV measurements derived from the DSC data.18,29,30 Single-shot gradient-echo (GE) echo-planar imaging was used to collect images during a second 0.05–0.15-mmol/kg bolus of Gd injected at a rate of 3 mL/s at 60 s after starting the DSC acquisition. A total of 120 DSC time points were obtained for each image slice. Typically, 13 slices of DSC data were acquired with the following parameters: thickness of 5 mm, skip 1.5 mm slice prescription, fat suppression, flip angle = 72° (70–80°, 3T), TE/TR: 30 ms/1s-1.25s, field of view of 220 × 220 mm, matrix size of 128 × 128, and voxel size of 1.72 × 1.72 × 5 mm. Fluid attenuated inversion recovery (FLAIR) imaging was also acquired with typical vendor-provided parameters, including an inversion time of 2200–2600 ms (2250–2500 3T), TE/TR: 90-150/8000–10 000 ms (127–136/9000 ms, 3T), slice thickness of 5 mm, skip of 1.5 mm, flip angle of 90° (150°, 3T) and matrix of 512 × 512, and voxel size of 0.43 × 0.43 mm.

Independent Component Analysis

Preprocessing of the DSC data included removal of the first 4 time points, after which a steady-state baseline signal was achieved, followed by motion correction using MCFLIRT (FMRIB tool library). Data were then processed using probabilistic independent component analysis,31,32 as implemented in the MELODIC software (FMRIB tool library). The algorithm was limited to modeling 3 components from each DSC acquisition. This was found to be sufficient for defining the arterial, venous, and one additional component. We empirically determined that modeling 3 components maximized the number of AVOL voxels.33 Results were then manually categorized into arterial and venous components. In general, arterial components were easy to distinguish, because the Circle of Willis was always present, as were symmetric cortical arteries. Components considered to be venous contained the choroidal veins/choroid plexus and superficial draining veins leading to the major draining sinuses.33


The MELODIC-derived statistically thresholded (mixture modeled, alternative hypothesis testing P > .5 vs. null31) arterial and venous maps were binarized. A threshold of 0.5 in alternative hypothesis testing indicated a voxel with a higher probability of being in the active class than the background noise class. The overlap of the 2 maps was calculated to locate voxels with mixed arterial and venous kinetics (Fig. 1). For the untreated GBMs in dataset A, the AVOL was assessed both in postcontrast enhancement (ie, tumor) and outside enhancement and/or FLAIR abnormality (ie, normal vasculature or nontumor). This was enabled by first coregistering the T1, T1 + C, and mean DSC image to the FLAIR image (FLIRT, FMRIB tool library). Contrast-enhancing (tumor) regions of interest (ROIs) were created by subtracting standardized T1 images from a standardized T1+C images, followed by empirical thresholding. These ROIs were then manually edited to exclude nontumor voxels, such as those in normal vessels or dura. A board-certified radiologist (S.D.R.) verified questionable ROIs. FLAIR ROIs were created by empirically thresholding each FLAIR image and manually excluding regions misclassified. The coregistered ROIs were then down-sampled from the FLAIR resolution to the DSC resolution and interpolated using a nearest-neighbor interpolation. In both tumor and nontumor, the total volume of arterial, venous, and AVOL components was determined, and percentages of each were compared.

Fig. 1.
Example of AVOL in a representative case of untreated GBM. (Top) Row 1 shows a series of T1-weighted postcontrast images, and row 2 shows the same series with the arterial (red) and venous (blue) ICA components overlaid. The overlap of the two is indicated ...

For the patients undergoing bevacizumab treatment in dataset B, the DSC data collected after initiation of bevacizumab treatment were coregistered to the DSC images collected before treatment using FLIRT (FMRIB tool library). The AVOL maps from both imaging sessions were then restricted to regions of initial FLAIR abnormality or contrast enhancement, created as previously described. This was done to ensure that regions considered to be abnormal at treatment onset remained classified as such, even after bevacizumab reduced the extent of FLAIR abnormalities at follow-up.9 The volume of AVOL from both time points was then calculated, and the baseline was compared with the follow-up by calculating a difference relative to the mean, where

equation image

RCBV and Volume of Enhancement Measurements

Voxelwise rCBV values were calculated based on methods previously published,18,29,30 using a leakage-corrected trapezoidal integration, followed by intensity standardization,34,35 as implemented in the IBNeuro software package ( The reference T1 scan acquired in the same slice prescription as the DSC data was coregistered to the FLAIR images, and the resulting transformation matrix was applied to the rCBV maps to bring the rCBV into the same space as the T1+C and FLAIR images. These ROIs were manually edited to exclude nontumor voxels, such as those in normal vessels, or dura. A board-certified radiologist (S. D. R.) verified questionable ROIs. Median standardized rCBV values were then extracted from the contrast-enhancing ROIs for each patient at each time point. A relative difference in rCBV and a relative difference in enhancing volume were then calculated similar to equation (1). These values were then compared across groups dichotomized by +/−ΔAVOL.

Statistical Comparisons

In dataset A, the percentage of venous, arterial, and AVOL were compared between tumor and nontumor vessels with use of a paired t test. Results were considered to be statistically significant for P < .05. Kaplan-Meier survival curves were generated for dataset B to compare groups using GraphPad Prism software (GraphPad Inc., La Jolla, CA). Patient OS after bevacizumab treatment initiation was grouped on the basis of either a positive or a negative change in AVOL (ΔAVOL). P <.05 was considered to be statistically significant. Two analyses were performed: (1) GBMs alone and (2) all grade III and IV gliomas. Percentage difference in enhancing volume and median rCBV in enhancement were also calculated and compared between +/−ΔAVOL groups with use of a 2-sample t test. The OS was compared between groups split by median age. Finally, a Pearson correlation was computed to determine the relationship among ΔAVOL vs ΔT1 + C, ΔAVOL vs ΔrCBV, and ΔAVOL vs OS.


Figure 1 shows AVOL in a representative patient with an untreated GBM. The arterial and venous components are shown in red and blue, respectively. The analysis of the 11 untreated GBMs is shown in Fig. 2, in which the percentage of AVOL in tumor vasculature is significantly higher than in nontumor vasculature. In the tumor, there are more venous and fewer arterial voxels, compared with normal tissue. Figure 3 shows 2 representative patients from dataset B undergoing bevacizumab treatment, one with a positive ΔAVOL and the other with negative ΔAVOL after treatment. These 2 patients had substantially different survival times of 103 days and 409 days, respectively. Kaplan-Meier curves (Fig. 4) revealed that patients with GBM with a negative ΔAVOL after treatment had an OS advantage (n = 15; 348 days) over patients with a positive ΔAVOL (n = 8; 197 days), with a hazard ratio (HR) of 2.51 (P = .0398). The OS for the entire group of 32 grade III and IV gliomas was significantly different, with a median survival for –ΔAVOL (n = 21) of 399 days and +ΔAVOL (n = 11) of 153 days, with an HR of 2.71 (P < .0091) (Fig. 3). The OS differences remained statistically significant after the removal of the 4 BINGO trial patients (P = .0327). The OS was not significantly different when dichotomized by median age, although it showed a trend toward statistical significance (P = .066), with older (age at bevacizumab onset, >57 years) patients surviving longer than for younger patients (median, 405 days vs 233 days).

Fig. 2.
Results of AVOL analysis in 11 patients with untreated GBM. Percentage of voxels within the T1 + C enhancement (tumor) classified as arterial, venous, and AVOL, compared with normal-appearing vessels outside the enhancement and FLAIR abnormalities. The ...
Fig. 3.
Change in the ΔAVOL after bevacizumab treatment. The patient at the top exhibited increased AVOL after treatment, with a survival of 103 days, and the patient at the bottom had a dramatic decrease in AVOL, with a survival of 409 days. Both patients ...
Fig. 4.
Kaplan-Meier curves of all patients dichotomized into 2 groups of negative and positive changes in AVOL (+/−ΔAVOL). (A) The overall survival (following bevacizumab onset) curves for patients with GBM (n = 23) are significantly different ...

Relative difference in volume of enhancement and rCBV measures were not significantly different when comparing these values between the +/−ΔAVOL groups, with a P value of .81 and .87, respectively (Fig. 5 A1-B3). Pearson correlation between ΔAVOL and ΔT1 + C and ΔrCBV were not significantly correlated, with R = 0.065, P = .737 and R = 0.079, P = .683, respectively (Fig. 5D and E). ΔAVOL vs OS showed an R = −0.47, P < .05 (Supplemental Figure).

Fig. 5.
Comparison of rCBV and enhancing volume after bevacizumab treatment between groups dichotomized by +/−ΔAVOL in 29 patients with grade III and IV glioma with T1 + C imaging. (A1) Median standardized rCBV pre-bevacizumab, post-bevacizumab ...


This study used ICA to characterize the DSC-MRI signal time to identify regions of AVOL in brain tumors. We found that the percentage of ICA-classified tumor vasculature composed of AVOL was significantly greater than the percentage of AVOL in nontumor vasculature (Fig. 2). We hypothesized that these regions would benefit from bevacizumab treatment and, indeed, found that patients with −ΔAVOL after treatment showed an increase in OS (Fig. 4), which was not observed with the initial survival studies performed in response to bevacizumab treatment. This finding may suggest that a subpopulation of patients may be more responsive to bevacizumab with a higher OS than previously documented. Additional rCBV and enhancing volume measures did not significantly differ across groups defined by +/−ΔAVOL (Fig. 5, A1-B3) and did not correlate with ΔAVOL (Fig. (Fig.5,5, D and E), suggesting that ΔAVOL is a unique biomarker giving information not provided by these other measures.

The origin of AVOL is unclear. It is possible that tumor-generated arterioles and venules lying within close enough proximity for their perfusion kinetics to both contribute to the same voxel would result in the classification as both arterial and venous. This is consistent with our finding that normal brain has significantly fewer voxels with AVOL characteristics, compared with tumor. Early-stage vessels have been shown to be more susceptible to anti-VEGF therapy.36 AVOL may be sensitive to these early vessels because they develop in close proximity to one another. This would be consistent with our findings that patients with –ΔAVOL after bevacizumab treatment survive longer. It is also possible that AVOL voxels contain arterio-venous shunting37 vessels with abnormal perfusion characteristics. These vessels are common in angiograms of patients with GBM, in whom the characteristic capillary blush and early draining vein stage coincide with the arterial phase. Early drainage through malformed vessels may cause the perfusion characteristics of AVOL, but it may also be the case that such vessels would exhibit more arterial perfusion patterns. Although anti-VEGF treatment is generally thought to prune tumor vessels and decrease vessel diameter, basement membrane thickness, and vascular permeability.25 It is unclear what vascular characteristics altered by treatment result in a change in AVOL classification.

One of the most controversial issues regarding the application of ICA to fMRI data is determining the number of components necessary to most accurately fit the data.38 In this study, 3 components were modeled to, in effect, force vascular and tumor voxels to fit in nominal venous or nominal arterial components (or in the case of AVOL, both). The goal of this study was to determine which voxels in the tumor showed both arterial and venous kinetics. We empirically found that defining 3 components maximized the amount of AVOL. The addition of more components has the effect of breaking the DSC signal in tumor into more anatomical subcomponents. A previous study applying ICA to DSC imaging in brain tumors modeled 6 components resulting in a tumor-specific component,27 although portions of venous anatomy were also present in the same component. The MELODIC algorithm models the largest sources of variance, which when limited to 3 components, always includes both the arterial and the venous phases of the contrast perfusion. The results of any ICA study are dependent on the number of components chosen, and optimizing this for brain tumor DSC is a potential route for future research.

One potential disadvantage of working with imaging studies after bevacizumab treatment is that locating abnormalities becomes more difficult when vasogenic edema and enhancement recedes after treatment.9 This resulted in AVOL voxels often being detected in regions initially abnormal on FLAIR, yet appearing normal after treatment. We overcame this by masking both time points by the baseline FLAIR abnormality. This ensured that AVOL voxels were within regions of baseline abnormality, even if treatment induced a dramatic decrease in abnormality.

Another disadvantage to this study was that the imaging was analyzed retrospectively, where scanner field strength and type was not controlled. A future study should determine the repeatability and reproducibility of ICA across scanner vendors and field strengths when applied to DSC imaging in brain tumor cases.

One further potential disadvantage of this study is that pathological samples were not available at the time of bevacizumab treatment. Treatment in all cases, except for the 4 patients enrolled in the BINGO trial, began at tumor recurrence. There is some inherent uncertainty with regard to the exact tumor diagnosis at recurrence, because lower WHO grade III tumors may have indeed progressed to WHO grade IV. For this reason, the data were separated and processed with GBM only and then with the addition of the WHO grade III tumors. Although both analyses yielded significant results, the addition of the grade III patients strengthened the OS analysis (Fig. 4). It is also conceivable that pseudoprogression drove the diagnosis of tumor progression, leading to bevacizumab treatment. Because tissue diagnosis was not available at treatment onset, this remains uncertain. It is unclear what the AVOL signature would be in regions of pseudoprogression and what relationship changes in AVOL after treatment have to pseudoprogression. These questions raise potential avenues for further research. Histological validation of the vasculature included in AVOL voxels should also be the topic of future studies.

The change in enhancing volume and rCBV after treatment was not significantly different when compared across the +/−ΔAVOL groups. This does not mean that these biomarkers are insensitive to treatment response. In fact, mean rCBV has been shown to be predictive of survival.19,39 This study shows that ΔAVOL adds additional unique information not realizable with these biomarkers because changes in both rCBV and enhancing volume after treatment onset are not different between the +/−ΔAVOL groups.

 In conclusion, we showed an application of ICA to DSC perfusion MRI. We found that the overlap of the arterial and venous ICA components occurred in higher proportions in ICA-classified tumor vasculature (ie, vessels within enhancement) than in normal vasculature. We also found that change in AVOL after bevacizumab treatment predicted OS, whereas relative percentage change in enhancing volume and rCBV did not differ across the +/−ΔAVOL-defined groups and did not correlate with ΔAVOL. We believe that ICA DSC has the potential to contribute an additional imaging biomarker for the clinical monitoring of recurrent GBM.


This work was supported by MCW Translational Brain Tumor Research Program, Advancing a Healthier Wisconsin, and National Institutes of Health, National Cancer Institute (R01 CA082500).


We thank Cathy Marszalkowski, for administrative support, and the patients, for participating in our imaging studies.

Conflict of interest statement: K. M. S. has had ownership interest in Imaging Biometrics LLC. S. D. R. has served on the medical advisory board for Imaging Biometrics LLC. All other authors: no conflicts.


1. Louis DN, Ohgaki H, Wiestler OD, et al. The 2007 WHO classification of tumours of the central nervous system. Acta Neuropathol. 2007;114(2):97–109. [PMC free article] [PubMed]
2. Stupp R, Mason WP, van den Bent MJ, et al. Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med. 2005;352(10):987–996. [PubMed]
3. Folkman J. Angiogenesis in cancer, vascular, rheumatoid and other disease. Nat Med. 1995;1(1):27–31. [PubMed]
4. Friedman HS, Prados MD, Wen PY, et al. Bevacizumab alone and in combination with irinotecan in recurrent glioblastoma. J Clin Oncol. 2009;27(28):4733–4740. [PubMed]
5. Kreisl TN, Kim L, Moore K, et al. Phase II trial of single-agent bevacizumab followed by bevacizumab plus irinotecan at tumor progression in recurrent glioblastoma. J Clin Oncol. 2009;27(5):740–745. [PMC free article] [PubMed]
6. Macdonald DR, Cascino TL, Schold SC, Jr., et al. Response criteria for phase II studies of supratentorial malignant glioma. J Clin Oncol. 1990;8(7):1277–1280. [PubMed]
7. Wen PY, Macdonald DR, Reardon DA, et al. Updated response assessment criteria for high-grade gliomas: response assessment in neuro-oncology working group. J Clin Oncol. 2010;28(11):1963–1972. [PubMed]
8. Hicklin DJ, Ellis LM. Role of the vascular endothelial growth factor pathway in tumor growth and angiogenesis. J Clin Oncol. 2005;23(5):1011–1027. [PubMed]
9. Pope WB, Lai A, Nghiemphu P, et al. MRI in patients with high-grade gliomas treated with bevacizumab and chemotherapy. Neurology. 2006;66(8):1258–1260. [PubMed]
10. Verhoeff JJ, van Tellingen O, Claes A, et al. Concerns about anti-angiogenic treatment in patients with glioblastoma multiforme. BMC Cancer. 2009;9:444. [PMC free article] [PubMed]
11. Ellingson BM, Cloughesy TF, Lai A, et al. Quantification of edema reduction using differential quantitative T2 (DQT2) relaxometry mapping in recurrent glioblastoma treated with bevacizumab. J Neurooncol. 2012;106(1):111–119. [PubMed]
12. Pope WB, Lai A, Mehta R, et al. Apparent diffusion coefficient histogram analysis stratifies progression-free survival in newly diagnosed bevacizumab-treated glioblastoma. AJNR Am J Neuroradiol. 2011;32(5):882–889. [PubMed]
13. Ellingson BM, Malkin MG, Rand SD, et al. Volumetric analysis of functional diffusion maps is a predictive imaging biomarker for cytotoxic and anti-angiogenic treatments in malignant gliomas. J Neurooncol. 2011;102(1):95–103. [PMC free article] [PubMed]
14. Ellingson BM, Cloughesy TF, Lai A, et al. Graded functional diffusion map-defined characteristics of apparent diffusion coefficients predict overall survival in recurrent glioblastoma treated with bevacizumab. Neuro Oncol. 2011;13(10):1151–1161. [PMC free article] [PubMed]
15. Bahr O, Hattingen E, Rieger J, Steinbach JP. Bevacizumab-induced tumor calcifications as a surrogate marker of outcome in patients with glioblastoma. Neuro Oncol. 2011;13(9):1020–1029. [PMC free article] [PubMed]
16. Maeda M, Itoh S, Kimura H, et al. Tumor vascularity in the brain: evaluation with dynamic susceptibility-contrast MR imaging. Radiology. 1993;189(1):233–238. [PubMed]
17. 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. [PubMed]
18. 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. [PubMed]
19. 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. [PubMed]
20. Sugahara T, Korogi Y, Tomiguchi S, et al. Posttherapeutic intraaxial brain tumor: the value of perfusion-sensitive contrast-enhanced MR imaging for differentiating tumor recurrence from nonneoplastic contrast-enhancing tissue. AJNR Am J Neuroradiol. 2000;21(5):901–909. [PubMed]
21. Hu LS, Baxter LC, Smith KA, et al. Relative cerebral blood volume values to differentiate high-grade glioma recurrence from posttreatment radiation effect: direct correlation between image-guided tissue histopathology and localized dynamic susceptibility-weighted contrast-enhanced perfusion MR imaging measurements. AJNR Am J Neuroradiol. 2009;30(3):552–558. [PubMed]
22. Sorensen AG, Batchelor TT, Zhang WT, et al. A “vascular normalization index” as potential mechanistic biomarker to predict survival after a single dose of cediranib in recurrent glioblastoma patients. Cancer Res. 2009;69(13):5296–5300. [PMC free article] [PubMed]
23. Quarles CC, Schmainda KM. Assessment of the morphological and functional effects of the anti-angiogenic agent SU11657 on 9L gliosarcoma vasculature using dynamic susceptibility contrast MRI. Magn Reson Med. 2007;57(4):680–687. [PubMed]
24. Bullitt E, Reardon DA, Smith JK. A review of micro- and macrovascular analyses in the assessment of tumor-associated vasculature as visualized by MR. Neuroimage. 2007;37(Suppl 1):S116–119. [PMC free article] [PubMed]
25. Jain RK. Normalization of tumor vasculature: an emerging concept in antiangiogenic therapy. Science. 2005;307(5706):58–62. [PubMed]
26. McKeown MJ, Makeig S, Brown GG, et al. Analysis of fMRI data by blind separation into independent spatial components. Hum Brain Mapp. 1998;6(3):160–188. [PubMed]
27. LaViolette PS, Cohen AD, Rand SD, et al. Independent component analysis of dynamic susceptibility contrast MRI in brain tumor: a new biomarker for measuring tumor perfusion patterns. Proc ISMRM, Montreal Quebec. 2011;789
28. Kao YH, Guo WY, Wu YT, et al. Hemodynamic segmentation of MR brain perfusion images using independent component analysis, thresholding, and Bayesian estimation. Magn Reson Med. 2003;49(5):885–894. [PubMed]
29. 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. [PubMed]
30. 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. [PubMed]
31. Beckmann CF, Smith SM. Probabilistic independent component analysis for functional magnetic resonance imaging. IEEE Trans Med Imaging. 2004;23(2):137–152. [PubMed]
32. Hyvarinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans Neural Netw. 1999;10(3):626–634. [PubMed]
33. LaViolette PS, Cohen AD, Schmainda KM. Contrast leakage in high grade glioma measured with independent component analysis of dynamic susceptibility contrast MRI. 2012;846 Proc ISMRM, Melbourne, Australia. [PMC free article] [PubMed]
34. Bedekar D, Jensen T, Schmainda KM. Standardization of relative cerebral blood volume (rCBV) image maps for ease of both inter- and intrapatient comparisons. Magn Reson Med. 2010;64(3):907–913. [PubMed]
35. Nyul LG, Udupa JK. On standardizing the MR image intensity scale. Magn Reson Med. 1999;42(6):1072–1081. [PubMed]
36. Sitohy B, Nagy JA, Jaminet SC, et al. Tumor-surrogate blood vessel subtypes exhibit differential susceptibility to anti-VEGF therapy. Cancer Res. 2011;71(22):7021–7028. [PMC free article] [PubMed]
37. Mariani L, Schroth G, Wielepp JP, et al. Intratumoral arteriovenous shunting in malignant gliomas. Neurosurgery. 2001;48(2):353–357. . discussion 357–358. [PubMed]
38. Li YO, Adali T, Calhoun VD. Estimating the number of independent components for functional magnetic resonance imaging data. Hum Brain Mapp. 2007;28(11):1251–1266. [PubMed]
39. Schmainda KM, Bedekar D, Rand SD, et al. Initial rCBV predicts response to bevacizumab in patients with high-grade gliomas. 2010;4288 Proc. Intl. Soc. Mag. Reson. Med., 18th Annual Meeting, Stockholm Sweden.

Articles from Neuro-Oncology are provided here courtesy of Society for Neuro-Oncology and Oxford University Press