HHS Public Access Author manuscript Author Manuscript
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01. Published in final edited form as: AJNR Am J Neuroradiol. 2015 December ; 36(12): 2242–2249. doi:10.3174/ajnr.A4451.
Impact of software modeling on the accuracy of perfusion MRI in glioma Leland S. Hu, MD, Zachary Kelm, BS, Panagiotis Korfiatis, BS, Amylou C Dueck, PhD, Christian Elrod, BS, Benjamin M Ellingson, PhD, Timothy J. Kaufmann, MD, Jennifer M. Eschbacher, MD, John P. Karis, MD, Kris Smith, MD, Peter Nakaji, MD, Debra Brinkman, PhD, Deanna Pafundi, PhD, Leslie C. Baxter, PhD, and Bradley J Erickson, MD, PhD
Author Manuscript
Abstract Purpose—To determine whether differences in modeling implementation will impact the correction of leakage effects (from blood brain barrier disruption) and relative cerebral blood volume (rCBV) calculations as measured on T2*-weighted dynamic susceptibility-weighted contrast-enhanced (DSC)-MRI at 3T field strength.
Author Manuscript
Materials and Methods—This HIPAA-compliant study included 52 glioma patients undergoing DSC-MRI. Thirty-six patients underwent both non Preload Dose (PLD) and PLDcorrected DSC acquisitions, with sixteen patients undergoing PLD-corrected acquisitions only. For each acquisition, we generated two sets of rCBV metrics using two separate, widely published, FDA-approved commercial software packages: IB Neuro (IBN) and NordicICE (NICE). We calculated 4 rCBV metrics within tumor volumes: mean rCBV, mode rCBV, percentage of voxels with rCBV > 1.75 (%>1.75), and percentage of voxels with rCBV > 1.0 (Fractional Tumor Burden or FTB). We determined Pearson (r) and Spearman (ρ) correlations between non-PLD- and PLD-corrected metrics. In a subset of recurrent glioblastoma patients (n=25), we determined Receiver Operator Characteristic (ROC) Areas-Under-Curve (AUC) for FTB accuracy to predict the tissue diagnosis of tumor recurrence versus post-treatment effect (PTRE). We also determined correlations between rCBV and microvessel area (MVA) from stereotactic biopsies (n=29) in twelve patients.
Author Manuscript
Results—Using IBN, rCBV metrics correlated highly between non-PLD- and PLD-corrected conditions for FTB (r=0.96, ρ=0.94), %>1.75 (r=0.93, ρ=0.91), mean (r=0.87, ρ=0.86) and mode (r=0.78, ρ=0.76). These correlations dropped substantially with NICE. Using FTB, IBN was more accurate than NICE in diagnosing tumor vs PTRE (AUC=0.85 vs 0.67) (p60 mg/min/1.72m2. Perfusion MRI (pMRI) data acquisition
Author Manuscript
Each 3T exam was performed on one of two MRI magnets (Sigma HDx; GE Healthcare, Milwaukee, Wisconsin or Magnetom Skyra; Siemens Healthcare, Erlangen, Germany). All patients underwent initial preload dose (PLD) administration that allowed the acquisition of “PLD-corrected” DSC-pMRI data, which were all acquired via a second GBCA injection (0.05-mmol/kg) (gadodiamide or gadobenate dimeglumine) using previously described methods.8,19 In all patients, the PLD amount totaled 0.1 mmol/kg, administered either via single bolus injection or two separate (0.05 mmol/kg) bolus injections, depending on the departmental protocol at the time of imaging. In a subset of patients, we acquired “NonPLD-corrected” DSC pMRI data during the initial PLD bolus injection, using either 0.05 mmol/kg or 0.1 mmol/kg GBCA injections, depending on the clinical perfusion MRI protocol employed at the time of acquisition. We performed a separate subanalysis to determine the impact of different injection doses as shown in Online Appendix/Table 2.
Author Manuscript
All DSC data (gradient-echo echo-planar imaging with TR/ TE/flip angle=1500–2000 ms/20 ms/60°; FOV=24×24cm; matrix=128×128; 5mm section; no gap) were acquired over 3 minutes with the bolus injection occurring at the 1-minute mark after the start of the DSC sequence. All GBCA injections were via power injector at 3–5cc/sec, followed by 20cc normal saline flush. The final GBCA dose for all patients (irrespective of the method of PLD administration) was 0.15 mmol/kg (body weight). Perfusion MRI (pMRI) data analysis After transferring all MRI data to an off-line workstation and removing baseline points collected during the first five seconds, we generated whole-brain relative cerebral blood volume (rCBV) maps using two commonly published commercial software packages:
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 4
Author Manuscript Author Manuscript
nordicICE (NICE) (v.2.3.13, NordicNeuroLab, Bergen, Norway) and IB Neuro (IBN) (v.1.1, Imaging Biometrics, Elm Grove, Wisconsin), both approved by the Food and Drug Administration (FDA). For NICE, we used all available default options and included leakage correction in all cases. Default options consisted of 1) automatic pre-bolus baseline selection to define pre-bolus baseline and integration intervals; and 2) subsequent noise threshold adjustment to maximize brain tissue used for CBV calculation. We did not employ spatial or temporal smoothing for either software package to help maintain data integrity and limit potential confounding factors. We performed rCBV calculations with gamma variate fitting (gvf) prior to leakage correction or without gvf. For IBN, we used all default options including leakage correction: 1) automated detection of brain tissue mask for voxles used in CBV calculation; 2) automated detection of contrast arrival within brain mask voxels to define pre-bolus baseline and integration intervals; 3) leakage correction based on Boxerman et al.17 For rCBV generated with either NICE or IBN, we co-registered the rCBV maps with stereotactic anatomical images using registration methods implemented in ITK, the Insight Segmentation and Registration Toolkit (www.itk.org) within IB Suite (v.1.0.454 Imaging Biometrics, Elm Grove, Wisconsin), as previously described.17,18,29,30
Author Manuscript
We normalized all rCBV maps to mean CBV from two 3×3 voxel-sized square regions of interest (ROIs) within contralateral frontal and parietal normal appearing white matter (NAWM).8,19 To reduce variability, we used identical NAWM ROIs for both software package analyses to generate all rCBV metrics. We calculated multiple previously published rCBV metrics including: 1) volume fraction of tumor voxels above the rCBV threshold of 1.75 (%>1.75); 2) volume fraction of tumor voxels above the rCBV threshold of 1.0, also known as perfusion-MRI Fractional Tumor Burden (FTB); 3) histogram mean rCBV and 4) histogram mode rCBV for all tumor voxels. We chose the thresholds 1.0 and 1.75 because of previous studies reporting the biological significance of these values.6,8,30 Based on the rCBV maps generated from NICE and IBN packages, we calculated volume fraction metrics using IB Suite and histogram metrics using custom code written in Matlab (v.R2012a, MathWorks, Natick, Massachusetts). To reduce variability, we also used identical segmented enhancing tumor volumes for both software analyses and all rCBV metrics (as described below). Conventional MRI acquisition and analysis
Author Manuscript
For each exam, we acquired routine conventional contrast-enhanced MRI that included preand post-contrast T1W Spoiled Gradient-Echo (SPGR-IR prepped) stereotactic (i.e., volumetric) MRI data sets (TI/TR/TE=300/6.8/2.8msec; matrix=320×224; FOV=26cm; slice thickness=2mm). Tumor volumes were defined as abnormal enhancing tissue by an experienced neuroradiologist (XXX). In non-enhancing glioma, we defined tumor volumes using T2W stereotactic MRI (TR/TE=4500/82msec; matrix=256×256; FOV=26cm; slice thickness=2mm). Stereotactic biopsy, image co-registration, and histologic microvessel analysis Our cohort included a subset of patients in which neurosurgeons collected an average of 2–3 tissue specimens from each tumor using stereotactic surgical localization, following the smallest possible diameter craniotomies to minimize brain shift. Biopsies were performed
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 5
Author Manuscript Author Manuscript
without knowledge of rCBV analyses. Similar to previous studies, biopsy locations and neuronavigational coordinates were recorded and coregistered with MRI to enable localized rCBV measurement (3×3 voxel sized ROIs) at corresponding biopsy sites.11,31 Multiple biopsy targets in the same patient were separated by a minimum of 2 cm. The neurosurgeon visually validated stereotactic imaging locations with corresponding intracranial anatomic landmarks, such as vascular structures. Stereotactic biopsy samples were sectioned (10μm thickness), CD-34 stained, and submitted for quantification of total microvessel area (MVA) using previously published methods.31–34 Corresponding sections were also stained with hematoxylin-eosin per standard protocol. For each CD-34 stained slide, we measured total microvessel area (MVA) as previously described.31,32,35 Raw data from seven of these patients were studied previously (XXX). The current study differs in that: 1) we employed commercial software packages and modeling correction to measure rCBV; 2) we determined test performance differences between packages; and 3) we compared PLD- against non-PLD conditions. Quantification of histologic tumor fraction (HTF) in recurrent GBM Our cohort included a subset of 25 patients with recurrent Glioblastoma Multiforme (GBM), previously treated with Stupp protocol.36 We enrolled each of these patients at the time of recurrence, at which time they underwent pre-operative MRI (including pMRI) for surgical debulking of newly developed or enlarging lesions suspicious for recurrence identified on surveillance CE-MRI.
Author Manuscript
Following debulking, we fixed all surgical tissue specimens in 10% formalin, embedded in paraffin, sectioned (10-micron), and H&E stained per standard diagnostic protocol at our institution. Two neuropathologists quantified GBM and/or PTRE elements for all specimens without knowledge of DSC-MRI, by simultaneously estimating histological fractional volume of tumor relative to nonneoplastic treatment-related features, as previously described.8,30,37,38 Features of tumor recurrence38 and PTRE37,39 were quantified and used to determine the histologic tumor fraction (HTF) from surgical resection material to diagnose either tumor progression (HTF≥50%) or PTRE (HTF1.75; r=0.93, ρ=0.91); correlations were also high for mean rCBV (r=0.87, ρ=0.86) and mode rCBV (r=0.78, ρ=0.76). With NICE modeling, these correlations dropped substantially (Figure 1) for thresholding metrics (FTB; r=0.79, ρ=0.72) (%>1.75; r=0.55, ρ=0.61), mean rCBV (r=0.11, ρ=0.42) and mode rCBV (r=0.44, ρ=0.65). By omitting gamma variate fitting, correlations for mean rCBV using NICE mildly increased although the other metrics remained largely unchanged (Table 1). Upon visual inspection of thresholding maps, nonPLD and PLD-corrected voxels showed greater spatial correspondence when using IBN compared with NICE (Figure 2). Table 1 summarizes correlations for all conditions. Of these 36 patients, 10 received PLD via 2 separate half-dose injections. To assess potential effects of heterogeneity in PLD administration, we performed a sub-analysis (n=26) excluding these 10 subjects, which showed correlations consistent with the original analysis (Online Appendix/Table 2). The type of modeling implementation impacts rCBV’s accuracy to diagnose tumor vs pseudoprogression/radiation necrosis
Author Manuscript
In a subset of recurrent GBM patients (n=25) undergoing surgical debulking for suspected tumor recurrence, we used ROC analysis to determine the accuracy of FTB, as measured by IBN or NICE, to diagnosed tumor vs PTRE (i.e., pseudoprogression, radiation necrosis). We used histologic tumor fraction (HTF) from surgical resection to categorize each subject’s diagnosis as either tumor recurrence (HTF≥50%) or PTRE (HTF 1.75 (%>1.75). The thresholding metrics (FTB, %>1.75) correlate most strongly between PLDand non-PLD corrected conditions.
Author Manuscript AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 14
Author Manuscript Author Manuscript
Figure 2.
(A–E): Representative case in a 39 y/o with recurrent high-grade Ganglioglioma shows an enhancing mass (A). Color overlay %>1.75 thresholding maps (B–E) depict orange voxels with high rCBV>1.75, compared with intermediate yellow voxels (rCBV 1.0–1.75) and low green voxels (rCBV
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01. Published in final edited form as: AJNR Am J Neuroradiol. 2015 December ; 36(12): 2242–2249. doi:10.3174/ajnr.A4451.
Impact of software modeling on the accuracy of perfusion MRI in glioma Leland S. Hu, MD, Zachary Kelm, BS, Panagiotis Korfiatis, BS, Amylou C Dueck, PhD, Christian Elrod, BS, Benjamin M Ellingson, PhD, Timothy J. Kaufmann, MD, Jennifer M. Eschbacher, MD, John P. Karis, MD, Kris Smith, MD, Peter Nakaji, MD, Debra Brinkman, PhD, Deanna Pafundi, PhD, Leslie C. Baxter, PhD, and Bradley J Erickson, MD, PhD
Author Manuscript
Abstract Purpose—To determine whether differences in modeling implementation will impact the correction of leakage effects (from blood brain barrier disruption) and relative cerebral blood volume (rCBV) calculations as measured on T2*-weighted dynamic susceptibility-weighted contrast-enhanced (DSC)-MRI at 3T field strength.
Author Manuscript
Materials and Methods—This HIPAA-compliant study included 52 glioma patients undergoing DSC-MRI. Thirty-six patients underwent both non Preload Dose (PLD) and PLDcorrected DSC acquisitions, with sixteen patients undergoing PLD-corrected acquisitions only. For each acquisition, we generated two sets of rCBV metrics using two separate, widely published, FDA-approved commercial software packages: IB Neuro (IBN) and NordicICE (NICE). We calculated 4 rCBV metrics within tumor volumes: mean rCBV, mode rCBV, percentage of voxels with rCBV > 1.75 (%>1.75), and percentage of voxels with rCBV > 1.0 (Fractional Tumor Burden or FTB). We determined Pearson (r) and Spearman (ρ) correlations between non-PLD- and PLD-corrected metrics. In a subset of recurrent glioblastoma patients (n=25), we determined Receiver Operator Characteristic (ROC) Areas-Under-Curve (AUC) for FTB accuracy to predict the tissue diagnosis of tumor recurrence versus post-treatment effect (PTRE). We also determined correlations between rCBV and microvessel area (MVA) from stereotactic biopsies (n=29) in twelve patients.
Author Manuscript
Results—Using IBN, rCBV metrics correlated highly between non-PLD- and PLD-corrected conditions for FTB (r=0.96, ρ=0.94), %>1.75 (r=0.93, ρ=0.91), mean (r=0.87, ρ=0.86) and mode (r=0.78, ρ=0.76). These correlations dropped substantially with NICE. Using FTB, IBN was more accurate than NICE in diagnosing tumor vs PTRE (AUC=0.85 vs 0.67) (p60 mg/min/1.72m2. Perfusion MRI (pMRI) data acquisition
Author Manuscript
Each 3T exam was performed on one of two MRI magnets (Sigma HDx; GE Healthcare, Milwaukee, Wisconsin or Magnetom Skyra; Siemens Healthcare, Erlangen, Germany). All patients underwent initial preload dose (PLD) administration that allowed the acquisition of “PLD-corrected” DSC-pMRI data, which were all acquired via a second GBCA injection (0.05-mmol/kg) (gadodiamide or gadobenate dimeglumine) using previously described methods.8,19 In all patients, the PLD amount totaled 0.1 mmol/kg, administered either via single bolus injection or two separate (0.05 mmol/kg) bolus injections, depending on the departmental protocol at the time of imaging. In a subset of patients, we acquired “NonPLD-corrected” DSC pMRI data during the initial PLD bolus injection, using either 0.05 mmol/kg or 0.1 mmol/kg GBCA injections, depending on the clinical perfusion MRI protocol employed at the time of acquisition. We performed a separate subanalysis to determine the impact of different injection doses as shown in Online Appendix/Table 2.
Author Manuscript
All DSC data (gradient-echo echo-planar imaging with TR/ TE/flip angle=1500–2000 ms/20 ms/60°; FOV=24×24cm; matrix=128×128; 5mm section; no gap) were acquired over 3 minutes with the bolus injection occurring at the 1-minute mark after the start of the DSC sequence. All GBCA injections were via power injector at 3–5cc/sec, followed by 20cc normal saline flush. The final GBCA dose for all patients (irrespective of the method of PLD administration) was 0.15 mmol/kg (body weight). Perfusion MRI (pMRI) data analysis After transferring all MRI data to an off-line workstation and removing baseline points collected during the first five seconds, we generated whole-brain relative cerebral blood volume (rCBV) maps using two commonly published commercial software packages:
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 4
Author Manuscript Author Manuscript
nordicICE (NICE) (v.2.3.13, NordicNeuroLab, Bergen, Norway) and IB Neuro (IBN) (v.1.1, Imaging Biometrics, Elm Grove, Wisconsin), both approved by the Food and Drug Administration (FDA). For NICE, we used all available default options and included leakage correction in all cases. Default options consisted of 1) automatic pre-bolus baseline selection to define pre-bolus baseline and integration intervals; and 2) subsequent noise threshold adjustment to maximize brain tissue used for CBV calculation. We did not employ spatial or temporal smoothing for either software package to help maintain data integrity and limit potential confounding factors. We performed rCBV calculations with gamma variate fitting (gvf) prior to leakage correction or without gvf. For IBN, we used all default options including leakage correction: 1) automated detection of brain tissue mask for voxles used in CBV calculation; 2) automated detection of contrast arrival within brain mask voxels to define pre-bolus baseline and integration intervals; 3) leakage correction based on Boxerman et al.17 For rCBV generated with either NICE or IBN, we co-registered the rCBV maps with stereotactic anatomical images using registration methods implemented in ITK, the Insight Segmentation and Registration Toolkit (www.itk.org) within IB Suite (v.1.0.454 Imaging Biometrics, Elm Grove, Wisconsin), as previously described.17,18,29,30
Author Manuscript
We normalized all rCBV maps to mean CBV from two 3×3 voxel-sized square regions of interest (ROIs) within contralateral frontal and parietal normal appearing white matter (NAWM).8,19 To reduce variability, we used identical NAWM ROIs for both software package analyses to generate all rCBV metrics. We calculated multiple previously published rCBV metrics including: 1) volume fraction of tumor voxels above the rCBV threshold of 1.75 (%>1.75); 2) volume fraction of tumor voxels above the rCBV threshold of 1.0, also known as perfusion-MRI Fractional Tumor Burden (FTB); 3) histogram mean rCBV and 4) histogram mode rCBV for all tumor voxels. We chose the thresholds 1.0 and 1.75 because of previous studies reporting the biological significance of these values.6,8,30 Based on the rCBV maps generated from NICE and IBN packages, we calculated volume fraction metrics using IB Suite and histogram metrics using custom code written in Matlab (v.R2012a, MathWorks, Natick, Massachusetts). To reduce variability, we also used identical segmented enhancing tumor volumes for both software analyses and all rCBV metrics (as described below). Conventional MRI acquisition and analysis
Author Manuscript
For each exam, we acquired routine conventional contrast-enhanced MRI that included preand post-contrast T1W Spoiled Gradient-Echo (SPGR-IR prepped) stereotactic (i.e., volumetric) MRI data sets (TI/TR/TE=300/6.8/2.8msec; matrix=320×224; FOV=26cm; slice thickness=2mm). Tumor volumes were defined as abnormal enhancing tissue by an experienced neuroradiologist (XXX). In non-enhancing glioma, we defined tumor volumes using T2W stereotactic MRI (TR/TE=4500/82msec; matrix=256×256; FOV=26cm; slice thickness=2mm). Stereotactic biopsy, image co-registration, and histologic microvessel analysis Our cohort included a subset of patients in which neurosurgeons collected an average of 2–3 tissue specimens from each tumor using stereotactic surgical localization, following the smallest possible diameter craniotomies to minimize brain shift. Biopsies were performed
AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 5
Author Manuscript Author Manuscript
without knowledge of rCBV analyses. Similar to previous studies, biopsy locations and neuronavigational coordinates were recorded and coregistered with MRI to enable localized rCBV measurement (3×3 voxel sized ROIs) at corresponding biopsy sites.11,31 Multiple biopsy targets in the same patient were separated by a minimum of 2 cm. The neurosurgeon visually validated stereotactic imaging locations with corresponding intracranial anatomic landmarks, such as vascular structures. Stereotactic biopsy samples were sectioned (10μm thickness), CD-34 stained, and submitted for quantification of total microvessel area (MVA) using previously published methods.31–34 Corresponding sections were also stained with hematoxylin-eosin per standard protocol. For each CD-34 stained slide, we measured total microvessel area (MVA) as previously described.31,32,35 Raw data from seven of these patients were studied previously (XXX). The current study differs in that: 1) we employed commercial software packages and modeling correction to measure rCBV; 2) we determined test performance differences between packages; and 3) we compared PLD- against non-PLD conditions. Quantification of histologic tumor fraction (HTF) in recurrent GBM Our cohort included a subset of 25 patients with recurrent Glioblastoma Multiforme (GBM), previously treated with Stupp protocol.36 We enrolled each of these patients at the time of recurrence, at which time they underwent pre-operative MRI (including pMRI) for surgical debulking of newly developed or enlarging lesions suspicious for recurrence identified on surveillance CE-MRI.
Author Manuscript
Following debulking, we fixed all surgical tissue specimens in 10% formalin, embedded in paraffin, sectioned (10-micron), and H&E stained per standard diagnostic protocol at our institution. Two neuropathologists quantified GBM and/or PTRE elements for all specimens without knowledge of DSC-MRI, by simultaneously estimating histological fractional volume of tumor relative to nonneoplastic treatment-related features, as previously described.8,30,37,38 Features of tumor recurrence38 and PTRE37,39 were quantified and used to determine the histologic tumor fraction (HTF) from surgical resection material to diagnose either tumor progression (HTF≥50%) or PTRE (HTF1.75; r=0.93, ρ=0.91); correlations were also high for mean rCBV (r=0.87, ρ=0.86) and mode rCBV (r=0.78, ρ=0.76). With NICE modeling, these correlations dropped substantially (Figure 1) for thresholding metrics (FTB; r=0.79, ρ=0.72) (%>1.75; r=0.55, ρ=0.61), mean rCBV (r=0.11, ρ=0.42) and mode rCBV (r=0.44, ρ=0.65). By omitting gamma variate fitting, correlations for mean rCBV using NICE mildly increased although the other metrics remained largely unchanged (Table 1). Upon visual inspection of thresholding maps, nonPLD and PLD-corrected voxels showed greater spatial correspondence when using IBN compared with NICE (Figure 2). Table 1 summarizes correlations for all conditions. Of these 36 patients, 10 received PLD via 2 separate half-dose injections. To assess potential effects of heterogeneity in PLD administration, we performed a sub-analysis (n=26) excluding these 10 subjects, which showed correlations consistent with the original analysis (Online Appendix/Table 2). The type of modeling implementation impacts rCBV’s accuracy to diagnose tumor vs pseudoprogression/radiation necrosis
Author Manuscript
In a subset of recurrent GBM patients (n=25) undergoing surgical debulking for suspected tumor recurrence, we used ROC analysis to determine the accuracy of FTB, as measured by IBN or NICE, to diagnosed tumor vs PTRE (i.e., pseudoprogression, radiation necrosis). We used histologic tumor fraction (HTF) from surgical resection to categorize each subject’s diagnosis as either tumor recurrence (HTF≥50%) or PTRE (HTF 1.75 (%>1.75). The thresholding metrics (FTB, %>1.75) correlate most strongly between PLDand non-PLD corrected conditions.
Author Manuscript AJNR Am J Neuroradiol. Author manuscript; available in PMC 2016 June 01.
Hu et al.
Page 14
Author Manuscript Author Manuscript
Figure 2.
(A–E): Representative case in a 39 y/o with recurrent high-grade Ganglioglioma shows an enhancing mass (A). Color overlay %>1.75 thresholding maps (B–E) depict orange voxels with high rCBV>1.75, compared with intermediate yellow voxels (rCBV 1.0–1.75) and low green voxels (rCBV
