Prospective analysis of parametric MRI biomarkers: Identification of early and distinct glioma response patterns not predicted by standard radiographic assessment

Abstract

Purpose

Currently, radiologic response of brain tumors is assessed according to the Macdonald criteria 10 weeks from the start of therapy. There exists a critical need to identify non-responding patients early in the course of their therapy for consideration of alternative treatment strategies. Our study assessed the effectiveness of the Parametric Response Map (PRM) imaging biomarker to provide for an earlier measure of patient survival prediction.

Experimental Design

Forty-five high grade glioma patients received concurrent chemoradiation. Quantitative MRI including apparent diffusion coefficient (ADC) and relative cerebral blood volume (rCBV) maps were acquired pre-treatment and 3 weeks mid-treatment on a prospective institutional-approved study. PRM, a voxel-by-voxel image analysis method, was evaluated as an early prognostic biomarker of overall survival. Clinical and conventional MR parameters were also evaluated.

Results

Multivariate analysis showed that PRM_ADC+ in combination with PRM_rCBV- obtained at week 3 had a stronger correlation to one-year and overall survival rates than any baseline clinical or treatment response imaging metric. The composite biomarker identified three distinct patient groups, non-responders (median survival (MS) of 5.5 months CI: 4.4-6.6) months, partial responders (MS of 16 months CI: 8.6-23.4) and responders (MS has not yet been reached.)

Conclusions

Inclusion of PRM_ADC+ and PRM_rCBV- into a single imaging biomarker metric provided early identification of patients resistant to standard chemoradiation. In comparison to the current standard of assessment of response at 10 weeks (MacDonald Criteria) the composite PRM biomarker potentially provides a useful opportunity for clinicians to identify patients who may benefit from alternative treatment strategies.

Introduction

The standard of care for patients with high grade gliomas is concomitant and adjuvant temozolomide and radiotherapy based on recent clinical results showing improved median and 5-year survival for patients treated with this chemoradiation regimen (1). Current clinical assessment of treatment response is determined 10 weeks post-treatment initiation based on Macdonald criteria(2-3). Limitations in accurate assessment of response has led to the development of a new standardized response criteria that accounts for transient changes in tumor volume post-therapy by the Response Assessment in Neuro-Oncology Working Group (RANO)(4). There exists a critical need for the development of individualized approaches to the treatment of this disease, which has a one year survival rate of roughly 50%. In this regard, MGMT-promoter methylation status is a potential predictive biomarker for selection of brain tumor patients who may derive the most benefit from treatment with temozolomide and radiotherapy (5). This initial success provides motivation for the development of non-invasive imaging biomarkers of response capable of predicting sensitivity to novel therapeutic interventions since not all patients are able to undergo surgical resection and provide sufficient tissue for MGMT analysis. The potential for using quantitative imaging modalities as early surrogate markers of response is compelling due to the ability to noninvasively detect treatment-associated functional and metabolic changes (6).

In this study, we investigated a multi-modality image-based response biomarker for high grade gliomas in order to more accurately predict treatment outcome early. Identification and validation of an image biomarker could provide the opportunity to select patients that may benefit from alternative treatment strategies. Quantitative magnetic resonance imaging (MRI) can provide structural and functional information regarding tumor cellularity and perfusion characteristics that compliment anatomical MRI. The value of this approach is that the microenvironment and physiology of the tumor can be monitored noninvasively, allowing for early assessment of tumor therapy response prior to treatment completion. Two techniques investigated for their diagnostic and prognostic value in brain tumor patients are diffusion weighted (DW)-MRI and dynamic susceptibility contrast (DSC)-MRI (7-8). DW-MRI, capable of measuring the random thermal (Brownian) motion of water, samples the tumor microenvironment on a subcellular level. A drop in tumor cellularity as a result of significant cell-kill following effective therapy has been associated with an increase in water diffusivity as expressed as the apparent diffusion coefficient (ADC) (9). This trend has been observed in preclinical and clinical studies supporting the notion of DW-MRI as a surrogate imaging biomarker for treatment response assessment in oncology (10-12). DSC-MRI, capable of quantifying changes in tumor hemodynamics, provides information on neovascularity and angiogenesis (13-17). At present, the relative cerebral blood volume (rCBV), as measured by DSC-MRI, has shown promise in determining tumor grade as well as prediction and early assessment of patient clinical response (18-21).

Imaging biomarkers (i.e. ADC or rCBV) of tumor response have historically been evaluated using the mean change in the whole tumor volume as the quantitative metric for detection of therapeutic efficacy. However, tumor heterogeneity is an inherent property of glioblastoma multiforme (GBM) and therefore standard methods of analysis using average values throughout an entire tumor can significantly under-estimate regional changes following therapy. In 2005 Moffat and colleagues proposed a new approach for evaluating tumor response using serial ADC maps termed the “functional diffusion map” (fDM) (22). For the first time changes in ADC were evaluated on a voxel-wise basis by spatially aligning (via image registration) serial ADC maps acquired pre- and mid-therapy. The tumor volume fraction that exhibited a significant increase in mid-treatment ADC from baseline values was found to be predictive of 10 week radiological response in a cohort of 20 glioma patients as early as three weeks into therapy. This technique has now been established in multiple clinical sites, including breast cancer, prostate cancer, metastatic cancer to the bone and head and neck cancer (23-29). For the most part, this technique has only been applied to ADC maps. In 2009, the voxel-based approach was extended to rCBV maps acquired from DSC-MRI of 44 glioma patients. Now referred to more generally as the “parametric response map” (PRM), this post-processing approach when applied to rCBV (PRM_rCBV) was predictive of survival as early as one week post-treatment initiation as well as distinguishing progression from pseudo-progression (30-31).

We report here the outcome of a prospective single-center trial of WHO Gr 3 and 4 glioma patients that was designed to evaluate the predictive accuracy of PRM_ADC and PRM_rCBV imaging biomarkers as single and composite metrics of early treatment response as assessed 3 week post-treatment initiation. We compared parametric models using PRM imaging metrics from all patients (n=45) to assess correlation with one year and overall survival.

Results

Patient Population

A total of 45 subjects with high-grade glioma were included in this prospective study (Table 1). The median survival for the population, as determined by Kaplan-Meier analysis was 14.9 months (95% confidence interval: 8.6-21.2 months) with 56% of the patient population realizing one-year survival. Patients expiring prior to one year had a median survival of 6.6 months (95% confidence interval: 4.4-8.8 months) whereas those whose survival was beyond one year had a median survival of 35.1 months (95% confidence interval: 13.7-56.5 months). Patient age, Radiation Therapy Oncology Group Recursive Partitioning Analysis (RTOG RPA) (32) and Radiological Response (RR) at 10 weeks, were found to be significant predictors of one-year survival (p=0.003, p=0.025 and p=0.001, respectively) (Table 1).

Table 1. Patient characteristics.

Variable		<1 year survival	≥1 year survival	p values
Total		20	25
Age	(years)	60.9±2.4	48.6±3.1	0.003*
Pre-treatment Tumor volume	CE-T1 (cm3)	45.1 ±8.1	31.9±4.7	0.166
Pre-treatment Tumor volume	FLAIR (cm3)	103.9±11.6	92.0±16.8	0.566
Pathology	Grade 3	1	7	0.059#
	Grade 4	19	18
Karnofsky performance status	<70	8	5	0.141#
	70+	12	20
Location	Frontal/Temporal	12	18	0.397#
	Other	8	7
Surgery	Biopsy	10	6	0.191#
	Sub-total	5	10
	Near GTR	5	9
Recursive partition analysis	1	0	3	0.025#
	2	0	0
	3	3	7
	4	5	8
	5	5	6
	6	7	1
Radiation Therapy	Dose (Gy)	68.2±2.0	70.4±1.4	0.375*
MacDonald Criteria	PR/SD	6	20	0.001#
	PD	14	5

Note: Statistical significance were assessed at p<0.05 using an unpaired Student's t-test (*) and Likelihood Ratio (#). Pre-treatment tumor volumes were determined on contrast-enhanced T l weighted images (CE-Tl) and FLAIR images.

Individually PRM_ADC and PRM_rCBV have been reported to be predictive biomarkers of clinical outcome (30, 33). The underlying mechanisms driving therapeutic-induced alterations in ADC and rCBV differ and therefore the predictive value of the outcome measures of PRM (PRM_X+ and PRM_X-) vary on the quantitative parameter used for analysis. A large volume fraction of the tumor with significantly increasing ADC (PRM_ADC+) correlated with overall survival, unlike PRM_ADC- which had no predictive value. In stark contrast, a small volume fraction of tumor with decreasing rCBV (PRM_rCBV-) correlated with a positive outcome, where no predictive value was observed for PRM_rCBV+. Color-overlays of PRM_ADC and PRM_rCBV at week 3 post-treatment initiation of a representative responder (overall survival=17months; PRM_ADC+=12%; PRM_rCBV-=4%) and non-responder (overall survival=2months; PRM_ADC+=1%; PRM_rCBV-=40%) show spatial variation in the response of these tumors with respect to water diffusivity and blood volume (Figure 1). Both lesions had high vascular irregularities as can be presumed from the immense contrast-enhancement on the T1-weighted images (Figure 1A,D). The complex changes in tumor cellularity and vasculature following a therapeutic intervention appear from the PRM overlays to be somewhat dissociated from each other. For example, the responding tumor clearly showed regions of increased ADC, denoted in red, (Fig. 1B), which is associated with a loss in tumor cellularity. These same regions had statistically no change in rCBV (green area in Fig. 1E). In contrast, a significant volume of the non-responding tumor showed no change in ADC, Figure 1C, suggesting no real loss of tumor cellularity. Unlike ADC, Figure 1F shows that a substantial fraction of tumor underwent a significant drop in rCBV post-treatment initiation, perhaps suggesting increased hypoxia in the tumor conferring radioresistance. Therefore, the combination of diffusion and perfusion imaging may provide complementary information that serves to improve sensitivity and specificity in predicting patient outcome. To evaluate a combined metric related to these physiological changes to treatment, we analyzed these parameters as a unified predictive metric of survival.

Figure 1.

Representative (A,D) contrast-enhanced images and PRM overlays applied to (B,E) ADC and (C,F) rCBV from a (A-C) responding and (D-E) non-responding patients. The responding patient, diagnosed with a glioblastoma, had an overall survival of 17 months, PRM_ADC+ of 12% and PRM_rCBV- of 4%. In contrast, the non-responding subject, also diagnosed with a glioblastoma, had an overall survival of 2 months, PRM_ADC+ of 1% and PRM_rCBV- of 40%. PRM was acquired using pretreatment and 3 week post-treatment initiation imaging data.

We first assessed the sensitivity of the individual parameter models (PRM_ADC+ and PRM_rCBV-) and composite model (PRM_ADC+/PRM_rCBV-), as well as RR for comparison, to predict 6 month progression free survival (PFS) and one year survival in our cohort of patients using a logistic regression. Univariate analysis found a strong correlation of parametric models with both 6 month PFS and one year survival (Table 2). Inclusion of all four imaging metrics (PRM_ADC+, PRM_rCBV-, PRM_ADC+/PRM_rCBV- and RR) into a single multivariate logistic regression showed that PRM_ADC+/PRM_rCBV- and RR provided the strongest statistical power for assessing 6 month PFS (p=0.012 and p=0.005, respectively) but only PRM_ADC+/PRM_rCBV- for assessing one year survival (p<0.0001). To determine the statistical power of age, dichotomized based on responders ≥50 years and non-responders <50 years, and RPA RTOG against the imaging models, a multivariate analysis was performed with age, RTOG RPA, PRM_ADC+ and PRM_rCBV- for assessing one-year survival. Both age and RTOG RPA classification were not found to contribute to the statistical model.

Table 2. Statistical analyses of parametric models.

Model	6 month PFS		One Year Survival		Overall Survival
	Univariate p values	Multivariate p values	Univariate p values	Multivariate p values	Univariate p values	Multivariate p values
PRMADC+	0.021		0.003		0.002
PRMrCBV-	0.001		0.001		0.001
PRMADc+/PRMrcBv-	<0.0001	0.012	<0.0001	<0.0001	<0.0001	<0.0001
Radiological Response	<0.0001	0.005	0.001		<0.0001

Note: All parameter models were assessed at 6 month progression free survival (PFS), one year survival and overall survival. The Likelihood ratio test from a univariate and multivariate logistic regressions was used to determine a significant fit of parametric models to 6 month PFS or one year survival. The log-rank test was used to assess correlation of individual models to overall survival. Multivariate Cox regression was performed on all parametric models presented to determine the correlation of respective models to overall survival. All multivariate analyses were performed using a forward entry algorithm. Statistical significance was assessed at p<0.05.

Stratification of the patient population by each of the four models used, which includes RR, are presented as Kaplan-Meier plots in Figure 2. The median survivals for non-responders and responders as determined by PRM_ADC+ were 8.3 months (95% confidence interval (CI): 6.2-10.4; n=25) and not reached (n=20), respectively. Using PRM_rCBV- to categorize non-responders and responders, there median survival was found to be 7.1 months (95% CI: 5.1-9.1; n=22) and 20.4 months (95% CI: 1.7-39.1; n=23), respectively. Median survivals for non-responders and responders as determined by RR were 8.1 months (95% CI: 6.0-10.2; n=19) and not reached (n=26), respectively. For the composite PRM model with the three discrete response groups median survivals for non-responders, intermediate and responders were 5.5 months (95% CI: 4.4-6.6 months n=14), 16 months (8.6-23.4 months n=19) and not reached (n=12), respectively. Using a log-rank test to assess significance, all four models were highly predictive of overall survival (Table 2). Performing this same statistical test pair-wise between response groups defined by the composite model found that non-responders had a significantly shorter survival than the intermediate and responder subpopulations (p<0.0001). Although not found to be significant based on our statistical criteria (p<0.05), the survival of responders when compared to intermediate generated a p value of 0.06. This is partly attributed to the small numbers analyzed in this study. Assessment of overall survival of all models in a unified multivariate test found PRM_ADC+/PRM_rCBV- assessed at week 3 mid-treatment to have the strongest statistical power even over RR.

Figure 2.

Kaplan-Meier survival plots for overall survival, presented as a function of (a) PRM_ADC+, (b) PRM_rCBV- and (c) PRM_ADC+/PRM_rCBV- stratification at week 3 post-treatment initiation and (d) Radiological Response stratification at week 10 post-treatment initiation. The three discrete groups, as categorized by the composite PRM model, had median survivals for non-responders, intermediate and responders of 5.5 months (95% CI: 4.4-6.6 months n=14), 16 months (8.6-23.4 months n=19) and not reached (n=12), respectively.

In the cohort of patients studied, eight were diagnosed with grade 3 gliomas. Additional analysis was performed to determine if inclusion of these patients adversely affected the results presented above. As observed in Table 1, 7 of these patients had survived longer than one year with only a single patient surviving less than a year (OS was 5.1 months). This lone patient had a tumor that exhibited PRM values of ADC and rCBV indicative of a tumor non-responsive to therapy. Based on these results, the composite PRM categorized this patient as a non-responder. In fact, out of the 8 patients with grade 3 glioma, the composite model of PRM only identified the referenced patient as a non-responder, 3 were intermediate (mean OS (SD) was 22 (6.7) months) and 4 were responders (mean OS (SD) was 35 (19) months). Finally, assessment of the composite PRM model when limited to grade 4 gliomas found the composite PRM to be significant for OS (p<0.0001) with median survivals of 6 months (n=13) for non-responders, 12.8 months (n=16) for intermediate response, and median survival not reached at greater than 30 months in the 8 patients scored as responsive. Multivariate analyses, analogous to those performed in Table 2, of patients with grade 4 gliomas (n=37) identified a strong correlation of PRM_ADC+/PRM_rCBV- and RR to 6 month PFS (p=0.016 and p=0.02, respectively) and only PRM_ADC+/PRM_rCBV-to both one year survival (p=0.002) and overall survival (p<0.0001).

Discussion

Personalized medicine in cancer care is slowly becoming a reality from the “one-size fits all” approach. Advances in molecular and histological biomarkers are now providing detailed information on tumor biology that is being used to individualize therapy (34). Pretreatment assessment using molecular and histological biomarkers is increasingly being used to dictate initial choice of therapy. There is also a critical need for the deployment of surrogate biomarkers of response to provide real-time information on the state of the tumor mid-treatment. With respect to assessment of treatment in GBM patients, modern advancements in neuroimaging techniques, such as MRI, positron emission tomography (PET), single photon emission computed tomography (SPECT) and computed tomography (CT), are providing detailed functional and quantitative information on tumor heterogeneity. More specifically, techniques utilizing MRI, with its high tissue contrast and resolution, are capable of imaging with high sensitivity changes in tumor microenvironment and hemodynamics at the cellular level allowing for early serial assessment of a tumor during treatment.

Standard management of malignant high grade gliomas continues to be concurrent temozolomide and radiation 60 Gy followed by adjuvant TMZ. Currently, randomized phase III studies are being conducted to determine the additional benefit of novel targeted therapies, including various signal transduction inhibitors encompassing monoclonal antibodies, e.g. Avastin (bevacizumab), as well as small molecules. However, these therapies are unlikely to benefit all patients and are extremely costly and associated with a significant increase in the risk of toxicity. Reliable predictive imaging biomarkers of response are important in selecting those patients most likely to benefit from alternative treatment strategies and selecting those patients as early as possible.

We and others have found quantitative neuroimaging methods are highly sensitive to analyzing treatment-induced cellularity and hemodynamic alterations within the tumor. Difficulties in analyzing these parametric maps ensue from the high spatial heterogeneity typical of gliomas. By focusing on regions of increasing or decreasing parameter value, the functional behavior of the tumor can be spatially dissected and interpreted for its prognostic value. Our group has demonstrated the improved prognostic capabilities of quantitative imaging techniques (DW- and DSC-MRI) when analyzed by PRM, which has been compared to standard histogram-based methods and radiological response (30, 33). Expanding further, we have demonstrated in the present study that a composite of ADC and rCBV as analyzed by PRM enhances the sensitivity of the biomarker for predicting those patients resistant to chemoradiation at 3 weeks post-treatment initiation.

It is important to note that administration of steroid would have no confounding effect on the results reported in this study. As indicated in the manuscript steroid doses were closely followed prior to treatment, weekly during radiation therapy, and at the time of each MRI (pre-treatment and 3 weeks during treatment). Given the close timing between the pre-treatment scan (within 5 days prior to the start of treatment) and the 3 week scan (on average <21 days from the start of treatment) there were minimal to no changes in steroid dosing over this short time period. It has been our standard clinical practice not to taper steroids during the first several weeks of radiation therapy due to concerns for possible increased edema during the early course of therapy. Therefore, steroid doses were typically not adjusted until weeks 3-6 of radiation therapy. Further, the benefit seen by composite PRM was not simply a measure of generalized steroid effect since previous results have shown that the percentage change in either ADC and rCBV would also have been impacted by steroid dosing and were not prognostic (30, 33).

Varying MRI field strengths as well as Gd-DTPA doses used in DSC-MRI for calculating CBV were also not suspected of affecting the current results. It is important to note that ADC is independent of field strength. Nevertheless, from the total population studied (n=45), 31 were imaged at 1.5T and the remaining 14 were imaged at 3T. To avoid variability between scanners, the same scanner was used for a given patient throughout the study. An additional analysis was performed to determine if there were any systematic differences in mean ADC and rCBV values pre-therapy and at week 3 mid-treatment as well as differences in PRM_ADC+ and PRM_rCBV- as a result of field strength using an unpaired Student's t-test. No significant differences in imaging parameters were observed as a result of field strength. As for varying dosing of contrast agent, our algorithm for calculating CBV mitigates the effects of variable Gd-DTPA dose by use of the arterial input function measured on each patient individually, which implicitly scales with the same dose factor. All our CBV analyses use values normalized by white matter CBV which further reduces dose dependency. To mitigate leakage bias as a result of T1 effects, we employed a pre-injection of contrast agent prior to a second bolus given during dynamic T2* imaging. In addition, we used a relatively long TR of 1.5-2 seconds to further reduce T1-weighting.

This is the first study that describes a composite model of PRM_ADC+ and PRM_rCBV- obtained at week 3 in assessing tumor response. Our data shows that the composite model is a much stronger predictive factor of overall survival compared to either metric alone. This composite metric provides a novel imaging biomarker that has significant implications for early stratification in clinical trials in high grade glioma as well as potentially other clinical sites. When included in a single multivariate statistical analysis, the composite of PRM_ADC+ and PRM_rCBV- obtained at week 3 was found to be a much stronger predictor of overall survival than either metric alone, as well as 10 week radiological response. Use of the composite imaging biomarker not only provides a more early and sensitive method for identifying populations of responders and nonresponders but also allowed for the identification of a third patient subgroup that exhibited an intermediate response pattern not previously shown by conventional response criteria. Ongoing studies are elucidating the molecular determinants of response that differentiate tumors from each of these three response groups. Such information could ultimately facilitate the design of improved therapies for GBM patients identified in this manner.

A biomarker that allows for early identification of patients resistant to standard of care also provides an opportunity to reassess treatment options in a timelier manner compared to current clinical practice (35-37). For example, determination of responsiveness at week 3 into temozolomide with concomitant radiotherapy could allow patients resistant to standard of care to be given adjuvant treatment or be moved more quickly to an aggressive salvage regimen. To obtain further support for this approach, we recently initiated a clinical trial utilizing the composite imaging biomarker at week 3 following treatment initiation to select non-responding patients for further treatment intensification with bevacizumab and dose painting IMRT. While a multicenter trial to validate these findings is warranted, the results presented herein currently support a future paradigm shift in the current clinical management of GBM to a more individualized approach. In GBM, this would also provide an opportunity to develop new trials designed to improve patient outcome by selecting patients likely to benefit from more aggressive or experimental treatment regimens.

Materials and Methods

Patient Population

Patients newly diagnosed with histological confirmed grade III and IV gliomas were enrolled on a protocol of intra-treatment MRI. This cohort of patients has been used previously in published work investigating the prognostic value of PRM applied to the individual parameters ADC and rCBV (30, 33). Surgical resection was performed on all patients and the extent was evaluated by post-operative MRI obtained within 72 hours of resection. Radiotherapy was delivered using 3D-conformal therapy or intensity modulated radiation therapy. Standard techniques were utilized with a 2.0-2.5 cm margin on either the enhancing region on gadolinium-enhanced scans or the abnormal signal on T2-weighted scans to 46-50 Gy with the central gross tumor treated to a final median dose of 70 Gy in 6-7 weeks (38). Twenty-three of these individuals were treated on a phase 2 protocol of high-dose (>60 Gy) radiation therapy concurrent with temozolomide 75 mg/m² daily followed by adjuvant temozolomide 150-200 mg/m² daily for 5 days of a 28 day cycle. Remaining patients received radiation alone (n=22). Chemotherapy was commonly administered in both groups depending upon clinical circumstances (Table 1). None of these patients, accrued from 2000-2006, received any anti-angiogenic therapy during their primary treatment although some patients did receive Avastin at time of progression. Informed consent was obtained, and images and medical record use was approved by the University of Michigan Institutional Review Board.

MRI Examination

Patients were evaluated pre-therapy, three and 10 weeks after the start of radiation with follow-up scans every 2 to 3 months. DW- and DSC-MRI as well as standard MRI (fluid attenuation inversion recovery, T2-weighted and gadolinium-enhanced T1-weighted MRI) were performed on either a 1.5T Signa (General Electric Medical Systems) (n=31 subjects) or 3T Achieva (Philips Medical Systems) (n=14 subjects) system. Radiological response at 10 weeks was based on changes in tumor volume on T1-weighted contrast-enhanced MRI and steroid doses and was classified as complete response (CR), partial response (PR), stable disease (SD) and progressive disease (PD)(2). Steroid doses were recorded before each scan, weekly during radiotherapy, and at each follow-up (39).

Quantitative MRI

DW-MRI utilized a single-shot, spin-echo, diffusion-sensitized, echo-planar imaging (EPI) acquisition sequence. On the 1.5T system, 24 6-mm axial-oblique sections were acquired using a 22 cm-field of view (FOV) and 128 matrix with “b-factor” = 0 and 1000 s/mm² along three orthogonal directions (repetition time (TR) = 1000ms; echo time (TE) = 71 to 100 ms, and number of averages (NS) = 1). On the 3T system, at least 28 4-mm axial-oblique sections were acquired through the brain using a 24-cm FOV and 128 matrix (TR = 2636 ms; TE = 46ms; NS=1 for b=0, and NS=2 for b=1000 s/mm²). Parallel imaging (sensitivity encoding factor = 3) was used on 3T to reduce spatial distortion. The diffusion weighted images for the three orthogonal directions were used to calculate an apparent diffusion coefficient (ADC) map (33).

For DSC-MRI, 14 to 20 slices were acquired by a gradient-echo echo-planar imaging pulse sequence (TR=1.5 to 2s, TE=50 to 60ms, field of view 220×220 mm², matrix 128×128, flip angle 60°, and 4 to 6mm thickness and 0mm gap). Gadolinium-diethylenetriamine pentaacetic acid (Gd-DTPA; Bayer HealthCare Pharmaceuticals) was injected intravenously with a dose of 0.05 to 0.1ml kg^-1 as a bolus using a power injector at a rate of 2 mL s^-1, followed immediately by 15 cc of saline flush at the same rate. Subsequently a Gd-enhanced T₁-weighted imaging was acquired. Cerebral blood volume (CBV) map was generated from DSC images as described previously (40). To assess differences in tumor blood volume during chemoradiation therapy (cRT) and between subjects, CBV maps were normalized to values within white matter regions contralateral to the tumor to generate relative CBV maps (rCBV). For simplicity in notation, relative blood volumes for both brain and tumor have been denoted by “rCBV” throughout this article. The following guidelines were used to define the white matter regions of interest for normalization: (1) contralateral to tumor, (2) received <30 Gy accumulated dose, and 3) as large as possible but avoiding regions with susceptibility artifacts and partial volume averaging.

Image Analysis

Pre- and 3 week mid-treatment ADC and rCBV maps were co-registered to Gd-enhanced T1-weighted images acquired before cRT using mutual information as an objective function and simplex as an optimizer (41). The co-registration of different and similar weighted MRI scans for the same subject is totally automatic and assumes a rigid-body geometry relationship, i.e. rotation and translation, between head scans. Following co-registration, brain tumors, manually contoured by a neuroradiologist, were defined within the enhancing regions of the tumor on the Gd-enhanced T1-weighted images. Only patients with contrast-enhancing tumors ≥ 4cc were included in this study. For simplicity, tumor will refer to only the contrast-enhancing volume on a T1-weighted post-contrast image. Shrinkage or growth of the tumor during the time between scans may have occurred; therefore, only voxels that are contrast enhancing on both the pre-RT and mid-RT tumor volumes were included. Subsequent to co-registration, the resultant maps consist of spatially aligned voxels with ADC and rCBV values prior and 3 weeks following treatment initiation.

The Parametric Response Map of any parameter (PRM_X, where X denotes any parametric map such as ADC and rCBV) was determined by first calculating the difference between X (ΔX = mid X – baseline X) for each voxel within the tumor pre-therapy and at week 3 post-treatment initiation. Voxels yielding ΔX greater than a predetermined threshold (set to 1.2 and 55 for rCBV and ADC, respectively; details described previously (22, 30, 33)), were designated red (i.e. ΔrCBV > 1.2; ΔADC > 55). Blue voxels represent volumes whose parameter value decreased by more than the threshold (i.e. ΔrCBV < -1.2; ΔADC < -55) and the green voxels represent voxels within the tumor that were unchanged (i.e. |ΔrCBV| ≤ 1.2; |ΔADC| ≤ 5.5). The volume fractions within the tumor determined from PRM_X were PRM_X+ for increasing X, PRM_X-for decreasing X, and PRM_X0 for unchanged X.

Data and Statistical Analyses

The patient population was stratified based on previously reported cutoff values of 4.7% and 6.8% for PRM_ADC and PRM_rCBV, respectively (30, 33). The composite model of these two dichotomized metrics was determined by categorizing patients as responders, non-responders or intermediate. Patients were designated non-responding if both single metrics predict non-responding (PRM_ADC+<4.7% and PRM_rCBV->6.8%), whereas patients were designated responders when both single metrics predict responder (PRM_ADC+≥4.7% and PRM_rCBV-≤6.8%). For the cases where single metrics were in disagreement patients were designated intermediate (PRM_ADC+≥4.7% and PRM_rCBV->6.8% or PRM_ADC+<4.7% and PRM_rCBV-≤6.8%). A univariate and multivariate logistic regression with forward entry was used to assess the predictive value of each metric for 6 month progression free survival (PFS) and one-year survival. Assessment of overall survival for the metrics was determined using a Kaplan-Meier analysis with log-rank test and a multivariate Cox regression with forward entry. Secondary analysis was performed to assess the predictive value of patient characteristics to one year survival using either an unpaired Student's t-test or likelihood ratio test. All statistical computations were performed with a statistical software package (SPSS Software Products), and results were declared statistically significant at the two-sided 5% comparison-wise significance level (p < 0.05).