Evaluation of the Diagnostic Performance of F18-Fluorodeoxyglucose-Positron Emission Tomography, Dynamic Susceptibility Contrast Perfusion, and Apparent Diffusion Coefficient in Differentiation between Recurrence of a High-grade Glioma and Radiation Necrosis

Introduction

The most challenging question in neuro-oncological practice is to differentiate between recurrence of a high-grade glioma and radiation necrosis.[1] This differentiation is vital as the prognosis and further management differs between recurrent tumor and radiation necrosis.[2,3] Many studies have focused on this very research question using various diagnostic techniques. Histopathological diagnosis remains the gold standard in resolving this dilemma of recurrence v/s radiation necrosis is concerned. It may not be a first choice to establish the diagnosis as invasive biopsy has its inherent risks and may further compromise the quality of life of these patients who already have been subjected to aggressive therapies.[4] In the current practice, the advanced magnetic resonance imaging (MRI) techniques including magnetic resonance spectroscopy, diffusion-weighted imaging (DWI), and perfusion-weighted imaging (PWI) along with nuclear medicine imaging techniques such as positron emission tomography (PET) and single-photon emission computed tomography (SPECT) remain the mainstay in the differential diagnosis of tumor recurrence and radiation necrosis.[5-8] With the advent of hybrid imaging modalities such as PET/MRI, we have a unique opportunity to capture the pathophysiology and metabolic characteristics of various brain pathologies in a single session.[9] Dynamic susceptibility contrast (DSC) MRI perfusion is a technique which captures the susceptibility changes occurring in the microvasculature during the first pass of a gadolinium-based contrast agent through the vascular bed.[10] As neoangiogenesis is one of the hallmarks of a high-grade glioma and radiation necrosis is essentially avascular, it is prudent to hypothesize that tumor recurrence should show increased perfusion, as quantified by the relative cerebral blood volume (rCBV), while radiation necrosis should be hypoperfused as compared to normal brain parenchyma.[11] As far as PET is concerned, the recurrent tumor is composed of metabolically active viable cells which show the uptake of the radiopharmaceutical F18-fluorodeoxyglucose (F18-FDG), whereas radiation necrosis may not show any uptake.[12] Diffusion-weighted MRI is expected to show a more restricted diffusion in cases of tumor recurrence as compared with radiation necrosis in view of more cellularity in the former; however, the reported literature is inconsistent in reporting the utility of apparent diffusion coefficient in differentiation of radiation necrosis from recurrence.[4] These techniques are far from being perfect in differentiation of recurrence from radiation necrosis and many false positives and false negatives may be found while interpreting the results of these techniques.[11,12] Although F18-FDG-PET is considered valuable in differentiating radiation necrosis from recurrence, it has now been shown that amino acid PET may provide a more comprehensive clinical information.[12,13]

In this study, we attempted to compare the diagnostic performance of F18-FDG-PET, diffusion-weighted MRI, and DSC MR perfusion in differentiating recurrence in high-grade gliomas from radiation necrosis.

Materials and Methods

A retrospective analysis of patients undergoing hybrid PET/MRI scan from January 2015 to October 2019 was done. Written informed consent was obtained from all the patients before the PET/MRI scan.

Analysis of positron emission tomography data

The PET data were analyzed in the syngo.via workstation (Siemens Healthineers, Erlangen, Germany). The fused PET/MR images were visually assessed for optimal quality. The maximum standardized uptake value (SUV_max) of the lesion and the contralateral normal-appearing cerebral cortex were determined using 3D region of interest (ROI) measurements, as shown in Figure 1. The tumor-to-brain ratio (TBR) was calculated for all the patients as a ratio between the SUV_max of the tumor and the SUV_max of the contralateral cerebral cortex. Early and delayed TBR and early and delayed SUV_max were recorded. The average delay between injection of the radiotracer and early image acquisition was 45 min (range: 35 min to 90 min) and the average interval between early and delayed image acquisition was 90 min (range: 60 min to 120 min).

Close

Figure 1 Method for calculation of TBR. TBR is calculated as the ratio of SUV_max of the lesion and the SUV_max of the normal-appearing cerebral parenchyma, TBR: Tumor-to-brain ratio, SUV_max: Maximum standardized uptake value

Export to PPT

Analysis of dynamic susceptibility contrast perfusion magnetic resonance imaging

DSC perfusion data were analyzed using the Philips IntelliSpace Portal 12.0 (Koninklijke Philips NV, Amsterdam, The Netherlands). Voxel-wise signal-to-time curves were inspected visually for assessment of optimal quality of perfusion data. Leakage correction was used for quantitative analysis of the DSC perfusion data, as depicted in Figure 2. Using postcontrast 3D magnetization-prepared rapid gradient echo (MPRAGE) as an underlay, the corrected rCBV was obtained by drawing an elliptical ROI with an area between 10 mm² and 20 mm² within the enhancing component of the lesion. A control mirror ROI was drawn in the contralateral cerebral parenchyma to determine the corrected rCBV in that location. Areas corresponding to hemorrhage, vessels, and cystic components were avoided while drawing the ROIs. A normalized rCBV ratio was then calculated by the software as the ratio between the corrected rCBV of the lesion and the corrected rCBV of the contralateral cerebral parenchyma.

Close

Figure 2 Method used for calculation of normalized rCBV ratios on Philips IntelliSpace Portal 6.0. ROI 1 is drawn within the enhancing component of the lesion as seen on the fused perfusion color map and postcontrast T1W MPRAGE images. ROI 2 is a mirror ROI with the same area as that of ROI 1 and is drawn on the contralateral normal-appearing cerebral parenchyma. The software uses leakage correction to derive the corrected rCBV values and the ratio of corrected rCBV values between ROI 1 and ROI 2. An overshoot of the signal intensity-time curve is noted which indicates effects of contrast leakage and T1 relaxation, rCBV: Relative cerebral blood volume, MPRAGE: Magnetization prepared rapid gradient echo, ROI: Region of interest

Export to PPT

Analysis of diffusion-weighted imaging and apparent diffusion coefficient ratio

The apparent diffusion coefficient (ADC) ratio was calculated using the Philips IntelliSpace Portal 12.0 (Koninklijke Philips NV, Amsterdam, The Netherlands). For determination of the ROI, PET images were taken as a guide and an ROI with an area between 10 mm² and 20 mm² was drawn within the area of uptake on PET and a mirror ROI was drawn within the contralateral normal-appearing brain parenchyma [Figure 3]. The ADC ratio was then calculated as the ratio between the ADC value of the lesion and the contralateral normal-appearing brain parenchyma.

Close

Figure 3 Grading of contrast enhancement on postcontrast T1W MPRAGE images. Grade 1 is shown in (a), Grade 2 in (b), and Grade 3 in (c), MPRAGE: Magnetization prepared rapid gradient echo

Export to PPT

Assessment of the contrast enhancement of the lesion

Postcontrast 3D MPRAGE sequence was used for a visual assessment of the degree of enhancement of the lesion. Using an ordinal grading scale, mild enhancement was given a Grade 1, moderate enhancement was given a Grade 2, and avid enhancement was given a Grade 3, as shown in Figure 4.

Close

Figure 4 ROC curves for early TBR (a), Delayed TBR (b), Early SUV_max (c), Delayed SUV_max (d), Normalized rCBV ratio (e), Normalized ADC ratio (f) and Contrast enhancement (g). The highest AUC is noted for delayed TBR (0.097). The lowest AUC is noted for normalized ADC ratio (0.353), ROC: Receiver operator characteristic, TBR: Tumor-to-brain ratio, SUV_max: Maximum standardized uptake value, rCBV: Relative cerebral blood volume, ADC: Apparent diffusion coefficient, AUC: Area under the curve

Export to PPT

Results

Fifty patients were included in the final analysis, 32 of them being men (64%). The average age of the patients was 46 ± 13.8 years (range, 10–64 years). The average time interval between the initial treatment and PET/MRI scan was 19.2 months. Glioblastoma was the most common primary tumor (n = 28, 56%) diagnosed on the initial histopathology. Anaplastic oligodendroglioma (n = 10, 20%), anaplastic oligoastrocytoma (n = 4, 8%), anaplastic ependymoma (n = 3, 6%), and anaplastic astrocytoma (n = 4, 8%) were the other histopathological types of primary tumors diagnosed among the patients. In one case, further classification could not be done on histopathology and it was labeled as anaplastic glioma without further classification. Following PET/MRI study, the final diagnosis was established by surgical resection/biopsy followed by histopathological examination in 22 cases (44%). In the remaining 28 cases (56%), the diagnosis was established on follow-up imaging as per the criteria described earlier. Thirty-six patients (72%) were diagnosed with recurrence and 14 patients (28%) were diagnosed as having radiation injury. Out of the 36 patients with a final diagnosis of recurrence, histopathological confirmation of the diagnosis was possible in 20 cases. In the remaining 16 cases, the diagnosis of recurrence was based on follow-up imaging criteria. In all the 14 patients with a diagnosis of radiation injury, the final diagnosis was made on the basis of follow-up imaging criteria.

Fifty lesions were analyzed for contrast enhancement, early and delayed TBR, early and delayed SUV_max on fused PET/MR images, normalized ADC ratios derived from DWI, and normalized rCBV ratios derived from DSC perfusion. All the lesions showed enhancement on postcontrast T1W MPRAGE sequence. Forty-one lesions (82%) showed Grade 3 enhancement. Grade 2 enhancement was noted in five lesions and Grade 1 enhancement was seen in four lesions. The ROC analysis of the various variables is summarized in Figure 4 and Table 1. The early and delayed TBR, early and delayed SUV_max, normalized ADC ratio, and normalized rCBV ratio were found to be higher among the recurrent lesions as compared to the radiation injury, as represented in Table 2, Figures 5 and 6. A cutoff value of 1.21 for normalized rCBV ratio was found to yield a sensitivity of 100%, a specificity of 60%, and an accuracy of 88.9% for the differentiation between radiation necrosis and recurrence. Among the FDG-PET metrics, a cutoff value of 0.8 for early TBR had a sensitivity of 100% and specificity of 80% for differentiating recurrence from radiation necrosis with an accuracy of 94.4%, which was found to be the highest among all the metrics analyzed in this study. A Grade 3 enhancement had a sensitivity of 100% and a specificity of 60% in making a differentiation between recurrence and radiation necrosis with an accuracy of 88.9%. Normalized ADC ratio with a cutoff of 1.66 had a sensitivity of 38.5% and a specificity of 80% for differentiating radiation necrosis from recurrence with an accuracy of 50%, which was the lowest among all the other metrics. The area under ROC curve was the highest for late TBR (0.907) and the lowest for normalized ADC ratio (0.35). Performance analysis after combining various metrics revealed that a combination of early TBR (cutoff value of 0.8), late TBR (cutoff value of 0.93), and rCBV ratio (cutoff value of 1.21) showed a sensitivity of 100%, specificity of 92.3%, positive predictive value of 88.9%, negative predictive value (NPV) of 93.7% and an accuracy of 96.6% in discrimination between radiation necrosis and recurrence of tumor. This performance measured was better than individual performances of each of the parameters as well as any other combination of parameters.

Close

Figure 5 A 47-year-old woman who had undergone surgical resection of right frontal glioblastoma followed by chemoradiation, presented with headache and seizures 21 months after the initial treatment. A well-defined lesion is seen in the right frontal lobe with Grade 3 contrast enhancement on axial postcontrast T1W MPRAGE (a), Fused 18F-FDG-PET/MRI image (b) shows avid uptake of the radiopharmaceutical with TBR of 2.82. DSC perfusion color map (c) and quantitative analysis of the perfusion parameters using leakage correction in Philips IntelliSpace Portal 6.0 (d) show elevated corrected rCBV within the lesion with normalized rCBV ratio of 7.93. The diagnosis of recurrence of glioblastoma was confirmed after surgical biopsy and histopathology, MPRAGE: Magnetization prepared rapid gradient echo, 18F-FDG-PET: F18-Fluorodeoxyglucose-positron emission tomography, MRI: Magnetic resonance imaging, TBR: Tumor-to-brain ratio, DSC: Dynamic susceptibility contrast, rCBV: Relative cerebral blood volume

Export to PPT

Close

Figure 6 44-year-old female was treated for anaplastic oligo-astrocytoma with surgical resection of the tumour followed by chemoradiation. She presented 28 months later with headache and vomiting. Axial post-contrast T1W MPRAGE image (a) shows patchy areas of post-contrast enhancement adjacent to the post-operative cavity. There was no obvious uptake in the lesion on 18F-FDG PET on visual assessment (b) and the TBR was 0.85. DSC perfusion color maps and (c) quantitative data from the analysis in Philips intellispace portal 6.0 shows no evidence of elevated perfusion within the lesion with (d) normalized rCBV value of 0.42. Follow-up MRI at 6 months and 12 months after PET/MRI scan (not shown) showed gradual reduction in the enhancing component of the lesion and a final diagnosis of radiation necrosis was made

Export to PPT

Discussion

High-grade glioma is a term which encompasses WHO Grade III and Grade IV gliomas. Among the high-grade gliomas, glioblastomas are considered the most aggressive and account for approximately 15%–20% of all primary intracranial tumors.[14] Recurrence is common in high-grade gliomas, and it has been shown that the mean time since initial treatment to recurrence is approximately 32–36 weeks in glioblastoma.[15] Recurrence occurs due to continuous neoplastic activity within a margin of 2–3 cm from the primary neoplasm.[15] The standard of care management of high-grade gliomas includes surgery, radiotherapy, and chemotherapy.[16] With increasing use of multimodality approaches to glioma management, the risk of radiation necrosis posttreatment of gliomas has increased.[1] Radiation necrosis is considered a late complication of brain irradiation and usually occurs months to years after the radiation exposure.[1] Among the various risk factors for development of radiation necrosis in gliomas, total radiation dose, fraction size, and chemotherapy are thought to be important.[17]

Radiation necrosis may have an appearance on conventional imaging which may be indistinguishable from tumor recurrence.[18,19] Although various patterns of contrast enhancement, such as “soap” bubble appearance and “Swiss cheese” appearance, have been described for radiation necrosis, these are not specific and may be seen in recurrent tumors as well.[18,19] In a study by Mullins et al. evaluating the conventional MRI features for distinguishing radiation necrosis from recurrence in gliomas, involvement of the corpus callosum along with various combinations of other findings, such as multiple enhancing foci, crossing of midline, and subependymal spread, were suggestive of progression of glioma.[20]

The performance of DSC MR perfusion has been evaluated in various studies. Sugahara et al. evaluated the value of DSC perfusion in differentiating tumor recurrence from nonneoplastic enhancing lesions in 20 patients. They included various types of neoplasms in their study including astrocytoma (Grades II and IV), ganglioglioma, germinoma, and primitive neuroectodermal tumor. They found that a normalized rCBV ratio of more than 2.6 was suggestive of recurrence whereas that of <0.6 was suggestive of nonneoplastic enhancing tissue. They further concluded that between normalized rCBV ratio of 2.6 and 0.6, ²⁰¹Tl-SPECT could help in differentiating the two.[21] In a retrospective study on 57 patients with glioblastoma, Barajas et al. found that the mean, maximum, and minimum relative peak height and rCBV values were significantly higher in recurrence of glioblastoma as compared to the radiation necrosis. They also showed that the mean, maximum, and minimum percentage signal recovery was significantly lower in recurrent glioblastoma as compared to radiation necrosis.[6] In a prospective study on 42 tissue specimens in 13 patients with high-grade gliomas, Hu et al. reported a threshold value of 0.71 for rCBV to differentiate between recurrence and radiation necrosis with a sensitivity of 91.7% and specificity of 100%.[22] Solimana et al. reported a threshold value of 1.8 for relative peak height and 1.22 for rCBV to differentiate recurrent glioma from radiation necrosis on DSC perfusion on the basis of a study on 20 patients with Grade II to Grade IV gliomas treated with surgery and radiotherapy with or without chemotherapy.[23] There is a lot of variation in the literature regarding the thresholds of perfusion metrics for differentiating radiation necrosis and recurrence. In our study, we found that a threshold of normalized rCBV ratio of 1.21 could differentiate recurrence from radiation necrosis with a sensitivity of 100% and specificity of 60%. The variation may be accounted for by the differences in the patient characteristics, the histopathological types of tumors, the imaging protocol used, the method of deriving the perfusion parameters, etc., which can vary from institution to institution and from one scanner to another.

F18-FDG-PET has been used as a tool for differentiating radiation necrosis and recurrence for many years. Di Chiro et al. reported a 100% sensitivity and specificity of 18F-FDG-PET in differentiating radiation necrosis from recurrence.[24] They had a histopathological diagnosis in all their cases. Another study by Valk et al. showed a sensitivity and specificity of 88% and 81%, respectively, of 18F-FDG for the diagnosis of radiation necrosis versus recurrence.[25] However, there are few studies which have demonstrated a lower sensitivity and specificity of 18F-FDG-PET because of the physiological tracer uptake of the brain which is the major disadvantage. For instance, Ricci et al., in their study on 31 patients with histopathological confirmation of diagnosis, found a sensitivity of 73% and specificity of 56% for differentiating radiation necrosis from recurrence.[26] Chao et al. reported an overall sensitivity of 75% and a specificity of 80%. They used MRI and PET co-registration in their study and found that for brain metastasis, there was an improvement of sensitivity from 65% (PET alone) to 86% (PET with MRI co-registration).[27] An inflammatory response in radiation necrosis may be one of the reasons for false-positive results when evaluating 18F-FDG in such cases.[28] Recently, amino acid PET has been used to differentiate radiation necrosis from recurrence. Hotta et al. used radiomics approach in 11C-methionine PET for differentiation of recurrence from radiation necrosis. They found that radiomics and tumor-to-normal ratio had a sensitivity of 90.1% and 60.6% and specificity of 93.9% and 72.7%, respectively.[29]

There have been only a few studies in the literature which have compared two or more modalities simultaneously for differentiation of radiation necrosis and recurrence. Kumar et al. compared DSC perfusion and 18F-FDG-PET in 28 cases (23 primary brain tumors and 5 metastatic lesions). They found that DSC perfusion had an accuracy of 94.5% as compared to 85.1% of 18F-FDG-PET for differential diagnosis of recurrence and radiation necrosis.[30] Tomura et al. studied 18 metastatic brain lesions in 15 patients who had undergone Gamma Knife radiosurgery. They compared 11C-methionine-PET, 18F-FDG-PET, MR permeability, and ADC for differentiating recurrence from radiation necrosis. They concluded that 11-C-methionine PET may be superior to MR-permeability scan and ADC in this regard.[31] A recent study by Qiao et al. compared 11C-methionine-PET/CT and DSC perfusion MRI for the purpose of differentiating radiation necrosis from recurrence in 42 patients (33 – recurrence and 9 – radiation injury). 11C-methionine PET/CT showed a sensitivity of 0.909 and specificity of 0.556, while DSC perfusion showed sensitivity and specificity of 0.667 and 0.778, respectively. The optimal cutoffs for tumor-to-background SUV_max and mean rCBV were 1.85 and 1.83, respectively. They concluded that both 11C-methionine PET/CT and PWI were equally accurate in differentiation of recurrence from radiation injury in high-grade gliomas.[32] Another study by Hojjati et al. compared FDG-PET/MRI, FDG-PET/CT, and DSC perfusion MRI in differentiating radiation necrosis from recurrence in glioblastoma. For PET/MRI, a relative mean ≥1.31 yielded an AUC of 0.94 with a 100% sensitivity and NPV. For DSC perfusion, CBV_max ≥3.32 yielded AUC of 0.94 with sensitivity and NPV of 100%. Joint model of relative mean (PET/MRI) and CBV (DSC perfusion) resulted in AUC of 1.0. They concluded that a combination of DSC perfusion with PET/MRI parameters had the best diagnostic value in differentiation of radiation necrosis from recurrence in glioblastoma.[33]

Diffusion-weighted MRI is sensitive to the Brownian motion of protons inside a particular voxel. ADC, derived from DWI, is a measure of diffusion of protons within a voxel (measured in units mm²/s).[34] A high cellularity confers restriction of anisotropic motion of water molecules, thereby lowering the ADC.[34] Intuitively, tumor recurrence should depict a lower ADC than radiation necrosis. However, the literature is inconsistent when it comes to reporting the ADC values in recurrence and radiation necrosis. Sundgren et al. reported a higher ADC value in recurrence as compared to radiation-induced changes.[35] Other studies by Hien et al.,[36] Xu et al.,[37] and Zeng et al.[38] reported a lower ADC value in recurrence. In a recent study by Zakhari et al.,[39] central diffusion restriction within a lesion was found to be indicative of radiation necrosis rather than recurrence. An interplay of vasogenic edema, necrosis, extracellular space enlargement, and gliosis could confound the otherwise seemingly straightforward and intuitive interpretation of ADC in differentiating radiation necrosis from recurrence.

In our study, we simultaneously compared TBR (early and delayed) and SUV_max (early and delayed) derived from 18F-FDG, normalized ADC ratio derived from DWI, and normalized rCBV ratios derived from DSC perfusion. The rationale behind evaluating early and delayed TBR and SUV_max is the improved sensitivity and specificity of detection of malignant lesions and their differentiation from benign lesions, as demonstrated by Kubota et al.,[40] Basu et al.,[41] and Suga et al.[42] Indeed, we found early TBR to be the most accurate in differentiating radiation necrosis from recurrence. rCBV ratios had a fairly good sensitivity, specificity, and accuracy in this regard as well. Contrast enhancement, graded visually, also had a comparable diagnostic performance with DSC perfusion, both having a high sensitivity (100%) but a low specificity (60%). The other metrics derived from F18-FDG-PET had a lower accuracy when compared with early TBR, DSC perfusion, and enhancement. The normalized ADC ratio had a very low sensitivity (38.5%) and accuracy (50%) in differentiating radiation necrosis from recurrence. This low diagnostic performance of ADC could be explained by the factors discussed earlier, such as vasogenic edema, necrotic component, and expansion of extracellular space and presence of gliotic scar tissue postradiotherapy which may complicate the ADC calculation. Our results indicate that it may be more useful to combine early TBR, delayed TBR, and rCBV ratio, in order to differentiate radiation necrosis from recurrence of glioma, since the accuracy, sensitivity, specificity, positive predictive value, and NPV of a combination of these parameters yielded a better diagnostic performance than individual parameters.

Various challenges are faced once PET/MRI is used as a diagnostic modality in evaluation of this problem of differentiating radiation necrosis from recurrence. Attenuation correction in PET/MRI is a problem since MRI is based on proton density and tissue relaxivity characteristics and does not provide direct information regarding electron density in contrast to PET/CT.[43] Various techniques have been attempted to resolve the issue of attenuation correction in PET/MRI such as pseudo-CT generation, zero TE sequence, atlas-based approaches, and Dixon-based approaches and new methods based on machine learning algorithms are being developed.[43] As far as DSC perfusion is concerned, contrast leakage from intravascular to extravascular compartment is one of the pitfalls in analysis of perfusion data.[10] Due to contrast leakage, T1 effects may overcome the DSC perfusion effects and may underestimate the rCBV.[44] Blood–brain barrier disruption is a characteristic feature of high-grade gliomas, and T1 effects may be predominant while computing rCBV using DSC perfusion.[44,45] In this situation, leakage correction algorithm may improve the accuracy of quantification of the rCBV values using DSC perfusion.[10,46] Blood–brain barrier disruption is also a feature of radiation necrosis.[1,19] We used leakage correction algorithm for analysis of DSC perfusion data as contrast leakage is a concern in both high-grade gliomas and radiation necrosis.

We acknowledge that a relative sample size may be considered a limitation of this study. Another limitation could be the nonavailability of histopathological confirmation of the diagnosis in all the cases.

Conclusion

Eighteen- FDG-PET and DSC perfusion can reliably differentiate tumor recurrence from radiation necrosis, with early TBR showing the highest accuracy for the differentiation at a cutoff value of 0.8. ADC demonstrates a low sensitivity, specificity, and accuracy in differentiating radiation necrosis from recurrence. A combination of early TBR, delayed TBR, and rCBV is the most useful approach with best diagnostic performance in differentiating radiation necrosis from recurrence of glioma.