Abstract
BACKGROUND AND PURPOSE: Differentiating glioma recurrence from treatment-induced necrosis can be a challenge on conventional imaging. This study aimed to assess the diagnostic performance of each functional MR imaging and PET parameter derived by using simultaneous FDG-PET/MR imaging individually and in combination in the evaluation of suspected glioma recurrence.
MATERIALS AND METHODS: Thirty-five treated glioma patients with 41 enhancing lesions (World Health Organization grade II = 9, III = 13, IV = 19) on MR imaging after an operation followed by radiation therapy and/or chemotherapy formed part of this study. Using PET/MR imaging, we calculated the normalized mean relative CBV, mean ADC, Cho/Cr, and maximum and mean target-to-background ratios. Statistical analysis was performed to determine the diagnostic performance of each parameter by receiver operating characteristic analysis individually and in combination with multivariate receiver operating characteristic analysis for the detection of glioma recurrence. Histopathology or clinicoradiologic follow-up was considered the criterion standard.
RESULTS: Of 35 patients, 25 (30 lesions) were classified as having a recurrence and 10 (11 lesions) patients as having treatment-induced necrosis. Parameters like rCBVmean (mean relative CBV), ADCmean, Cho/Cr, and maximum and mean target-to-background ratios were statistically significant in the detection of recurrent lesions with an accuracy of 77.5%, 78.0%, 90.9%, 87.8%, and 87.8%, respectively. On multivariate receiver operating characteristic analysis, the combination of all 3 MR imaging parameters resulted in an area under the curve of 0.913 ± 0.053. Furthermore, an area under the curve of 0.935 ± 0.046 was obtained when MR imaging parameters (ADCmean and Cho/Cr) were combined with the PET parameter (mean target-to-background ratio), demonstrating an increase in diagnostic accuracy.
CONCLUSIONS: Simultaneous PET/MR imaging with FDG offers correlative and synergistic multiparametric assessment of glioma recurrence with increased accuracy and clinical utility.
ABBREVIATIONS:
- AUC
- area under the curve
- CE
- contrast-enhanced
- max
- maximum
- rCBV
- relative cerebral blood volume
- ROC
- receiver operating characteristic
- SUV
- standardized uptake value
- TBR
- target-to-background ratio
Glioma is currently managed by surgical resection followed by radiation therapy and/or chemotherapy, depending on the aggressiveness of the tumor. Most of these tumors, despite treatment, recur or progress during the course of the disease. Furthermore, treatment with chemoradiation therapy is associated with necrosis.1 Up to 30% of patients with glioblastoma on chemoradiation therapy develop treatment-related effects, which mimic tumor recurrence, and both present as a new enhancing lesion on contrast-enhanced (CE) MR imaging.2 This effect gets further complicated because recurrences most often occur within or adjacent to the primary tumor site.3 Moreover, recurrent glioma and radiation necrosis may coexist, further obscuring this differentiation.4 Hence, differentiation of tumor recurrence from treatment-induced necrosis, which remains a challenge on conventional CE-MR imaging,2,5is important because their management and prognosis are completely different.6
Advanced neuroimaging with MR imaging and PET has been proposed in the past for assessment of glioma recurrence.7,8 MR imaging, besides providing superior tissue contrast, also provides functional information through perfusion, diffusion, and spectroscopy. PET imaging, on the other hand, provides additional information about tumor metabolism. FDG has been extensively used in the differentiation of glioma recurrence from radiation necrosis with variable results.9 High physiologic uptake of FDG in the normal brain and its uptake in inflammatory cells often result in poor tumor-to-background differentiation and make it a moderate test for characterization of tumor recurrence.10 This outcome has resulted in research into several more accurate PET tracers. Amino acid–based radio tracers such as 11C-methionine, 18F-fluoro-ethyl-tyrosine, and 18F-fluoro-l-thymidine are known to offer better imaging characteristics, however, are not easily available. 11C with a very short half-life needs an in-house cyclotron for production. 18FDG, being a commonly available tracer with low cost suitable for wider application, formed the basis of our work. The role of CT in the diagnosis of glioma recurrence is limited, with CE-MR imaging remaining the mainstay in neuroimaging.11 In an attempt to improve diagnostic accuracy, PET/CT has been used along with CE-MR imaging with limited success.12 With an aim to complement information for this purpose, multiparametric assessment with various functional MR imaging parameters such as choline/creatine, choline/N-acetylaspartate, relative cerebral blood volume (rCBV), and apparent diffusion coefficient have been used in the past,13,14 whereas a few studies have tried coregistration of MR imaging with PET parameters such as standardized uptake value (SUV) and target-to-background ratio (TBR) performed separately on different occasions to improve diagnostic accuracy.12,15,16
The advent of simultaneous PET/MR imaging has made it possible to assess all the parameters together in the same physical space in a single examination. We believe that simultaneous acquisition may help in overcoming some of the limitations of individual techniques and bring a synergistic effect in improved differentiation of recurrence from treatment-induced necrosis.
The present study aimed to assess the diagnostic performance of each of these parameters derived by using simultaneous FDG-PET/MR imaging individually and in combination by receiver operating characteristic (ROC) analysis for the evaluation of suspected glioma recurrence.
Materials and Methods
Patients
In this retrospective study between March 2013 and September 2015, 41 consecutive patients who underwent simultaneous FDG-PET/MR imaging for glioma recurrence evaluation were chosen after obtaining prior approval of the institutional review board and signed informed consent. Inclusion criteria were histopathologic-proved glioma, previous treatment with an operation and radiation therapy with or without chemotherapy, and a high index of clinical suspicion for recurrence along with contrast enhancement on MR imaging. Exclusion criteria were nonglial primary brain tumors, proved malignancy of other sites, pregnancy, being younger than 18 years of age, standard contraindications to MR imaging, and loss of the patient to follow-up. Finally, 35 patients (6 women, 29 men; mean age, 50 ± 12 years) with 41 lesions (World Health Organization grade II = 9, III = 13, IV = 19) formed part of this study. Twenty-nine patients received chemoradiation therapy, whereas 6 patients received only radiation therapy. The interval between radiation therapy and FDG-PET/MR imaging ranged from 7 to 96 months, with a solitary case with a maximum of 233 months. The patient characteristics are available in Table 1.
Lesion Diagnosis
A combination of clinical and imaging follow-up and histopathology was considered the criterion standard. Patients with a disease-related adverse event, progressive disease on imaging, and/or a biopsy positive for viable tumor tissue were positive for recurrence. Patients who were stable or did not show any adverse events clinically and did not progress on imaging were considered positive for treatment-induced necrosis.
Instrumentation: Simultaneous PET/MR Imaging
Simultaneous FDG-PET/MR imaging was performed on a Biograph mMR scanner (Siemens, Erlangen, Germany). This system consists of a modified 3T system (Magnetom Verio; Siemens) with a fully functional PET system, equipped with avalanche photodiode technology. The MR imaging scanner features a high-performance gradient system (45 mT/m) with a slew rate of 200 T/m/s and is equipped with total imaging matrix coil technology. The PET scanner has a spatial resolution of 4.3 mm at 1 and 5.0 mm at 10 cm from the transverse FOV; its sensitivity is 1.47% at the center of the FOV and 1.38% at 10 cm.
Imaging Protocol
The patients fasted for 6 hours before intravenous FDG injection of 352.12 ± 64.26 MBq. PET/MR imaging started 45–60 minutes after injection for 25–30 minutes of PET acquisition, during which various MR imaging sequences were performed.
A simultaneous brain PET/MR imaging protocol was composed of a transversal T1WI ultrashort TE sequence (TR/TE1/TE2, 11.94/0.07/22.46 ms) for attenuation correction and other MR imaging sequences for complete diagnostic evaluation of the brain, which included an axial FLAIR sequence (TR/TE, 7000/94 ms; TI, 2215.2 seconds; section thickness, 5 mm); a T2-weighted turbo spin-echo sequence (TR/TE, 4300/100 ms; section thickness, 5 mm); DWI (TR/TE, 4600/101 ms; b = 0, 400, 1000 s/mm2); PWI/perfusion EPI (TR/TE, 2550/31 ms); and 3D-encoded MPRAGE (TR/TE/TI, 1500/2.33 ms/900 seconds; spatial resolution, 1.2 × 1 × 1 mm) in the sagittal plane. In each case, 3D multivoxel 1H-MR spectroscopy (TR/TE, 1510/135 ms; FOV, 120 mm; matrix, 8 × 8 × 8 cm; acquisition, 1 average; scanning time, 5 minutes 41 seconds) was performed.
After the scan, all coincident data were sorted into a 2D-PET sinogram, which was subsequently reconstructed into transaxial sections, with an iterative 3D-ordered-subset expectation maximization algorithm with 3 iterations and 21 subsets, Gaussian smoothing of 4 mm in full width at half maximum, and a zoom of 1. The voxel size of brain PET images was 1.39 × 1.39 × 2.03 mm. MR imaging, PET, and PET/MR imaging scans were reviewed at a syngo.via platform (Siemens) by using the mMR general workflow.
PET and MR Image Quality Control
All PET/MR imaging studies were performed by using standard imaging protocol for both PET and MR imaging, adhering to the principle of the Quantitative Imaging Biomarkers Alliance protocol and standards (https://www.rsna.org/qiba/): in case of PET, tracer dose, time delay after injection, acquisition time; and, in case of MR imaging, consistency of imaging sequences. Distortion-correction-enabled diffusion-weighed images were acquired to minimize error in the ADC calculation. For PET images, emission data were corrected for randoms, dead time, and scatter, and attenuation correction was performed by using an MR imaging–based ultrashort TE sequence.
Image Analysis
Quantitative analysis of FDG-PET/MR imaging scans was performed by a radiologist and a nuclear medicine physician in consensus, with >10 years' experience in diagnostic radiology or nuclear medicine, who were blinded to the clinical information, histopathologic data, clinicoradiologic follow-up information, and final diagnosis for each lesion.
ROI Selection
First, by subtraction of precontrast from postgadolinium T1WIs, we achieved a software-based postcontrast subtraction series wherever possible to accurately isolate areas of enhancing tumor.17 We calculated ADC on-line by using system software, applying a monoexponential model. The most representative spectra observed within/around the lesion showing maximum PET uptake was taken for calculation of Cho/Cr.18
A freehand 2D ROI was drawn manually over the most representative section with maximum enhancement and was stored. This freehand ROI was duplicated by using the copy-paste function over the FDG-PET image, software generated ADC and CBV maps to find the respective maximum standardized uptake value (SUVmax), mean standardized uptake value (SUVmean), mean ADC (ADCmean), and mean CBV (CBVmean) (Figs 1 and 2). Maximum target-to-background ratio (TBRmax), mean target-to-background ratio (TBRmean), and rCBVmean were calculated by dividing SUVmax, SUVmean, and CBVmean obtained from the lesion ROI by the SUVmean, SUVmean, and CBVmean, respectively, from a separate ROI drawn for normalization on normal contralateral white matter.
Statistical Analysis
Data were checked for normality by using the Shapiro-Wilk test before statistical analysis. Individual PET/MR imaging parameters were evaluated for the difference between glioma recurrence and treatment-induced necrosis by using a 2-tailed independent Student t test or Mann-Whitney U test. Optimal-threshold value, area under the curve (AUC), sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were calculated by using receiver operating characteristic analysis for each of the PET/MR imaging parameters for their ability to detect recurrence. Multivariate analysis of variance was performed to check whether a combination of parameters was better than individual parameters. A multivariate ROC analysis was performed to analyze the increment offered by a combination of various parameters by conducting a logistical regression analysis to generate a combined ROC curve of different combinations of parameters. All statistical analyses were performed with the SPSS software package (Version 17.0; IBM, Armonk, New York). For all statistical tests, a P value < .05 was considered a significant difference.
Results
Patient Information
Thirty-five patients enrolled in this retrospective study had 41 lesions for evaluation. Five patients had multicentric lesions. Histopathology after repeat operation showed viable tumor in 16 patients (21 lesions) and evidence of treatment-induced necrosis in 2 patients (2 lesions).
Eight patients (9 lesions) without histopathologic evaluation were classified as having treatment-induced necrosis based on a stable clinical state on extended follow-up periods, absence of any new neurologic symptoms, and no remarkable increase of the lesion size and/or metabolic activity observed on follow-up PET/MR imaging up to a mean period of 11.8 ± 4.5 months (range, 7–24 months). Nine patients (9 lesions) were finally classified as having recurrent brain tumors due to the development of neurologic symptoms and a progressive increase in size on CE-MR imaging (1 patient) or PET/MR imaging (8 patients) during follow-up.
Scan Information
All 35 patients underwent simultaneous FDG-PET/MR imaging. One patient did not have perfusion MR imaging; in 4 patients (5 lesions), the MR spectroscopy study was not contributory because of noisy spectra, and it was not acquired in 2 patients (3 multicentric lesions) and hence was not included in the analysis.
Among considered parameters, rCBVmean, ADCmean, Cho/Cr, TBRmax, and TBRmean were found significant (P < .05) and used for further combined analysis (Table 2 and Fig 3). The tumor size ranged from 0.92 to 9.5 cm2.
Comparison of Advanced MR Imaging Parameters
PWI: rCBVmean.
In this study, rCBVmean was not significantly higher in patients with recurrence than in patients with treatment-induced necrosis (2.41 ± 1.02 versus 1.82 ± 1.12, P = .082; Fig 3A). However, the diagnostic accuracy of rCBVmean for correct identification of glioma recurrence reached 77.5% with rCBVmean ≥ 1.71 (AUC = 0.680 ± 0.111, P = .008), which was found to be significant (Table 2 and Fig 4A).
DWI: ADCmean.
The ADCmean was significantly lower in patients with recurrence than in those with treatment-induced necrosis (1283.13 ± 210.96 versus 1558.55 ± 313.32 × 10−6mm2/s, P = .015; Fig 3B). The diagnostic accuracy of ADCmean for correct identification of glioma recurrence reached 78% with ADCmean ≤ 1507 × 10−6mm2/s (AUC = 0.752 ± 0.015, P = .013; Table 2 and Fig 4A).
MR Spectroscopy: Cho/Cr.
Cho/Cr was significantly higher in patients with glioma recurrence than in those with treatment-induced necrosis (3.47 ± 2.20 versus 1.63 ± 0.64, P = .002; Fig 3C). The diagnostic accuracy of Cho/Cr for correct identification of glioma recurrence reached 90.9% by using Cho/Cr ≥ 1.405 (AUC = 0.861 ± 0.08, P < .001; Table 2 and Fig 4A).
Comparison of FDG Uptake Indices (TBRmax and TBRmean)
The TBRmax and TBRmean were significantly higher in patients with glioma recurrence than in those with treatment-induced necrosis (TBRmax, 2.50 ± 0.97 versus 1.53 ± 0.59, P = .001 and TBRmean, 1.70 ± 0.61 versus 0.98 ± 0.36, P < .001; Fig 3D). The diagnostic accuracy of TBR values for the correct identification of recurrence of brain gliomas reached 87.8% with TBRmax ≥ 1.58 (AUC = 0.827 ± 0.078, P < .001; Table 2 and Fig. 4C) and also with TBRmean ≥ 1.18 (AUC = 0.888 ± 0.059, P < .001; Table 2 and Fig 4C), though with a superior AUC.
Comparison of Combined Parameters
On multivariate analysis of variance, a statistically significant difference between glioma recurrence and treatment-induced necrosis (F5,26 = 5.871, P = .001; Wilks λ = 0.470, partial η2 = 0.530) was observed.
Combinational Analysis
Various combinations of several parameters were performed, and multivariate ROC analysis was performed as depicted in Table 3 (Fig 4B, -D).
Combined Analysis of MR Imaging Parameters: Cho/Cr, rCBVmean, and ADCmean
Individually, Cho/Cr achieved a maximum AUC of 0.861 ± 0.080 in the detection of glioma recurrence. On multivariate ROC analysis, the addition of ADCmean to either rCBVmean or Cho/Cr resulted in an improved diagnostic capability of both rCBVmean and Cho/Cr as noted by an increment in the AUC of the combinations. The combination of all 3 MR imaging parameters resulted in a combined AUC of 0.913 ± 0.053 (Table 3 and Fig 4B).
Combined Analysis of MR Imaging and FDG Parameters: TBRmax, TBRmean, Cho/Cr, rCBVmean, and ADCmean
Individually, TBRmean has the maximum AUC (0.888 ± 0.059) of all the MR imaging and PET parameters. Among intervariable evaluation with multivariate analysis of variance for the predictability of diagnosis, ADCmean (P = .001) and TBRmean (P = .001) were found to be the most significant variables in predicting glioma recurrence. On multivariate ROC analysis, the maximum AUC of 0.935 ± 0.046 was achieved with a combination of ADCmean, Cho/Cr, and TBRmean (Fig 4D). Moreover, a combination of either TBRmax or TBRmean with ADCmean and/or Cho/Cr improved the AUC value significantly beyond that of individual combining parameters (Table 3).
Summary of Results
Among all individual parameters, Cho/Cr in MR imaging and TBRmean in PET are the most significant discriminators for the prediction of recurrence. Among MR imaging parameters alone, rCBVmean should be used in conjunction with Cho/Cr and ADCmean for differentiating recurrence from treatment-induced necrosis. The maximum AUC is achieved by combining TBRmean, ADCmean, and Cho/Cr.
Discussion
Treatment-induced necrosis is a common treatment-related morbidity in the management of gliomas, and the rate of radiation necrosis increases with incorporation of temozolomide into high-grade glioma management.19 The differentiation of glioma recurrence or progression from radiation injury can be a radiologic dilemma, irrespective of the imaging technique used.20
It is known that functional imaging methods complement anatomic information and yield different aspects of pathophysiology. The combination of PET and CE-MR imaging information with retrospective fusion has been reported to increase diagnostic accuracy over any single technique.12,15,16 In an integrated PET/MR system, PET and MR imaging information are not only combined but an additional value is expected owing to co-interpretation of both the signals.21
In our study, the results derived from individual PET and functional MR imaging parameters from simultaneously acquired FDG-PET/MR imaging are in agreement with the published literature.
With an optimized threshold of 1507 × 10−6mm2/s for ADCmean, an AUC of 0.752 ± 0.015 was obtained, which is close to the reported threshold of 1490 × 10−6mm2/s and AUC of 0.779.22 We achieved an accuracy of 77.5% and an AUC of 0.68 ± 0.111 with an optimized threshold of 1.709 for rCBVmean. Our calculated sensitivity and specificity of 82.8% and 63.6% are similar to 86% and 70%, respectively, reported with a comparatively lower threshold of 1.3.23 However, wide variation has been reported in rCBV thresholds to detect recurrence,24 and reasons ascribed include rapid extravasation of gadolinium-based contrast agent and vascular leak in the irradiated bed, especially with chemoradiation therapy.25 With an optimized threshold of 1.405 for Cho/Cr, we achieved an accuracy and AUC of 90.9% and 0.861 ± 0.080, respectively, which are in concordance with the reported values of 93.1% and 0.913, respectively, with 1.54 as a threshold.26 FDG-PET has been used to detect glioma recurrence with a varied range of reported sensitivities from 43% to 95% and specificities from 50% to 100%.9 Despite known limitations of FDG in the detection of glioma recurrence of any histology, a recent meta-analysis of FDG found a respectable pooled AUC of 0.866 ± 0.034 for FDG-PET.9 We achieved an AUC of 0.888 ± 0.059 by using an optimized threshold of 1.18 for TBRmean in the detection of glioma recurrence that is in agreement with the reported AUC of 0.898 by Nozawa et al16 in the differentiation of recurrent/residual high-grade glioma from posttreatment changes or low-grade glioma, however, by using TBRmax with a threshold of 1.8.
Because there is no single technique that can solve this clinical dilemma, researchers have attempted to combine functional MR imaging parameters to achieve an improved diagnostic accuracy beyond any single parameter.13,27
Moreover, in the past, retrospective fusion of FDG-PET with CE-MR imaging has aided in improved differentiation of glioma recurrence qualitatively, compared with FDG-PET or MR imaging alone.12,15,16 In a group of 30 patients with high-grade gliomas, Estrada et al12 reported an accuracy of 73% for FDG-PET, which improved to 93% with visual qualitative analysis of fused FDG-PET with CE-MR imaging.
Matsusue et al,27 with a semi-quantitative multiparametric scoring system for ADC ratio, rCBV, Cho/Cr, and Cho/NAA in 15 patients, reported an accuracy of 93.3% in the detection of glioma recurrence. We achieved an AUC of 0.913 ± 0.053 by combining ADCmean, rCBVmean, and Cho/Cr by multivariate ROC analysis.
Our quantitative analysis of combined parameters also supports this body of evidence. We observed that combinations of individual parameters demonstrate superior results in comparison with individual parameters. We achieved an AUC of 0.913 ± 0.053 when all MR imaging parameters were combined, which further improved to 0.932 ± 0.046 when TBRmax or TBRmean was added, clearly highlighting the utility of combinational analysis. However, the highest AUC of 0.935 ± 0.046 was achieved by combining ADCmean, Cho/Cr, and TBRmean.
We found that a simultaneous PET/MR imaging study offers advantages such as being logistically simple, implying, thereby, that it is performed as a single examination to obtain PET and MR imaging datasets; reduced examination time, thereby improving patient compliance; and correlative image reading and interpretation, which could help in localizing the most suspicious area on the PET image in voxel selection for MR spectroscopy. Furthermore, as observed by Mong et al,28 we also found that PET uptake could help in accurate classification of suspicious ADC lesions. In the present study, we used an ultrashort TE sequence for MR imaging–based attenuation correction for brain PET, which is reported to result in accurate quantification of PET parameters,29 more importantly offering comparable accuracy of CT-based attenuation correction.
Our study has several limitations. It is a retrospective study with a small group and a relatively heterogeneous class of patients to perform robust statistical analysis. The present study only provides within-sample prediction characteristics and is not adequately sized to provide a decent out-of-sample prediction. Although desirable, histologic diagnosis in all the lesions could not be obtained due to ethical concerns associated with high morbidity and sampling error.30
Hence, further large-scale prospective multicentric trials are required with simultaneous PET/MR imaging and FDG, a commonly available radiotracer, to validate our results.
Conclusions
Simultaneous PET/MR imaging with FDG offers correlative and synergistic multiparametric assessment of glioma recurrence with increased accuracy and clinical utility.
Acknowledgments
We thank Dr Aru Singh for technical help in the statistical analysis.
Footnotes
All procedures performed in studies involving human participants were in accordance with the ethics standards of the institutional and/or national research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable ethics standards.
REFERENCES
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
- 12.
- 13.
- 14.
- 15.
- 16.
- 17.
- 18.
- 19.
- 20.
- 21.
- 22.
- 23.
- 24.
- 25.
- 26.
- 27.
- 28.
- 29.
- 30.
- Received August 27, 2016.
- Accepted after revision December 21, 2016.
- © 2017 by American Journal of Neuroradiology