Diagnostic Accuracy of Arterial Spin-Labeling, Dynamic Contrast-Enhanced, and DSC Perfusion Imaging in the Diagnosis of Recurrent High-Grade Gliomas: A Prospective Study

Study Population

This study was approved by the local ethics board (REB#20160425-01H, The Ottawa Hospital). Written informed consent was obtained from each patient enrolled in this study. Patients who presented with a newly enhancing lesion suspicious for a recurrent glioma between December 2017 and November 2021 at our hospital were prospectively enrolled. We included adult patients (18 years of age or older) who developed a new parenchymal enhancing lesion on follow-up MR imaging after receiving standard treatment. Treatment was based on the Stupp Protocol and consisted of surgical resection followed by standard radiation treatment with concomitant temozolomide. Patients were excluded for the following reasons: 1) a small enhancing lesion (measuring <1 cm in maximal length); 2) patients who did not have a re-resection and had incomplete radiologic or clinical follow-up following the research MR imaging; 3) delay between MR imaging and the operation (>65 days); and 4) no perfusion imaging or the lesion not covered on perfusion imaging.

Imaging Acquisition

Each patient was scanned on a 3T MR imaging scanner (Discovery MR750W; GE Healthcare). The conventional MR imaging protocol used axial T1 precontrast (TR = 8.1 ms, TE = 3 ms, TI = 450 ms, flip angle = 90°, voxel size = 0.44 × 0.44 × 1 mm); axial T2 FLAIR (TR = 1000 ms, TE = 92 ms, TI = 2651 ms, flip angle =160°, voxel size = 0.86 × 0.86 × 3 mm); and axial T1 postcontrast (TR = 13 ms, TE =2.9 ms, flip angle = 12°, voxel size = 1 × 1 × 1 mm). Axial DWI was acquired (multiple b-values from 10 to 1000 s/mm², TR = 5000 ms, TE = 98.9 ms, voxel size = 0.94 × 0.94 × 4.5 mm). ADC maps were generated in-line (on the scanner). Before gadolinium injection, ASL was performed using pseudocontinuous pulse ASL (pCASL) with a 3D stack of fast spin-echo readout (TR = 4588 ms, TE = 10.3 ms, flip angle = 111°, postlabeling delay = 2025 ms, voxel size = 1.8 × 1.8 × 5 mm, number of slices = 16, number of excitations = 3, duration = 4 minutes 13 seconds). Five background suppression (inversion) pulses were used.

For both DCE MR imaging and DSC MR imaging, we used intravenous injections of Gadovist 1.0 (Bayer Schering Pharma). For DCE MR imaging, a fixed dose of 0.05 mmol/kg of Gadovist was injected at 2 mL/s. For DSC MR imaging, a second dose of gadolinium (0.05 mmol/kg) was injected 10 minutes after the first injection at a rate of 4 mL/s. DCE MR imaging was performed using a 3D spoiled gradient-recalled sequence (TR = 5.3 ms, TE = 1.1 ms, flip angle = 25°, voxel size = 1.9 × 2.5 × 5 mm, 80 phases, duration = 4 minutes 35 seconds). For DSC MR imaging, we used a T2*-weighted EPI gradient recalled-echo sequence (TR = 1275 ms, TE = 45 ms, flip angle = 90°, voxel size = 1.8 × 1.8 × 5 mm, 50 phases, duration = 2 minutes 8 seconds). Two T1-mapping sequences were performed before and after contrast injection for all patients except for the last 10: saturation method by using adaptive recovery times for cardiac T1 mapping (SMART₁Map) (TR = 4.1 ms, TE = 1.6 ms, flip angle = 20°, voxel size =1.9 × 2.5 × 5 mm, duration = 2 minutes 53 seconds) and modified Look-Locker inversion recovery (MOLLI) (TR = 4.5 ms, TE = 1.6 ms, flip angle = 20°, voxel size = 1.9 × 2.5× 5 mm, duration = 2 minutes 53 seconds).⁸

Postprocessing of ASL Images

The perfusion-weighted images were produced in-line and were quantified into CBF values via a single compartment model by using the following equation:

For this equation, T1b is the T1 of blood and is assumed to be 1.6 seconds. The partial saturation of the reference image (PD) is corrected for using a T1t of 1.2 seconds (typical of gray matter). ST is saturation time and is set to 2 seconds. The partition coefficient λ is set to the whole-brain average, 0.9. The efficiency, ε, is a combination of both inversion efficiency (0.8) and background suppression efficiency (0.75), resulting in an overall efficiency of 0.6. PLD is the postlabeling delay. LT is the labeling duration set to 1.5 seconds. PW is the perfusion-weighted or the raw difference image. SF is the scaling factor of the PW sequence. The CBF is reported in milliliters/100 g/minute. NEX is the number of excitations for PW images.

Postprocessing of DSC Images

DSC images were processed using singular value decomposition and deconvolution as implemented in a commercially available software package (Olea Sphere 3.0; Olea Medical). MR signal intensity was converted to a T2* relaxation rate. An automated algorithm selected the most suitable pixels for the vascular input function. The signal intensity (SI) was converted to relative change in R2* using the standard expression: delR2(t) = –ln[S(t) / S_o] / TE, where S is the SI at time t; S_o, the baseline SI; and TE, the echo time. Correction for leakage in CBV calculations was done using a pre-injection of contrast agent from the DCE acquisition and linear fitting to estimate the T1 contamination caused by extravasation of contrast agent. Both uncorrected and corrected CBV maps were generated.

Postprocessing of DCE Images

Three methods were used to process the DCE images: 1) without T1 correction, 2) T1 correction with a SMART₁Map acquisition, and 3) T1 correction with MOLLI acquisitions. The extended Tofts model was used in all cases. For all patients, DCE images were processed directly in the software (Olea Sphere 3.0) to generate maps of plasma volume (Vp) and volume transfer constant (K^trans). The SI was converted to percentage change in signal intensity (relSI) using the expression: relSI(t) = 100 × [S(t)-S_o] / S_o, where S is the SI at time t and S_o is the baseline SI. The vascular input function was selected from a small ROI placed in the superior sagittal sinus directly from the DCE images. Signal conversion was set as SI to relSI(%). For patients who also had T1 mapping, the SI-versus-time curve was converted to gadolinium-versus-time using the T1 maps obtained before and after contrast from either the SMART₁Map or MOLLI acquisitions. This bookend T1 mapping has been described elsewhere.⁹ Vp and K^trans were then calculated for both T1-mapping techniques using the same Olea Sphere software.

Image Interpretation

A senior neuroradiologist trained a senior neuroradiology fellow on ROI placement and verified the technique. For each patient, the fellow, who was blinded to the outcome, drew an ROI over the largest solid component of the newly enhancing lesion using axial T1-weighted postcontrast images avoiding adjacent vessels and cystic, necrotic, or hemorrhagic areas. This ROI was coregistered on all the other parametric maps. For the uncorrected CBV and corrected CBV maps, an additional ROI was placed in the contralateral unaffected white matter to obtain a ratio (normalized value). The mean value for each ROI on each parametric map was recorded.

Reference Standard

For patients who underwent resection of the newly enhancing lesions, a neuropathologist blinded to the imaging results assessed the presence and percentage of viable tumor and/or radiation necrosis in the surgical specimens. Visual assessment was performed under × 4 high-power-field microscopy to determine the approximate percentage area of radiation necrosis or tumor recurrence relative to the whole-tissue area on all available slides. The following criteria were used for the diagnosis of radiation necrosis: the presence of coagulative necrosis and hyalinized vessels. For the diagnosis of recurrent tumor, we used the following criteria: the presence of increased cellularity and nuclear pleomorphism. In lesions in which there was a mixture of viable tumor and radiation necrosis, a lesion was categorized as tumor recurrence if the percentage of viable tumor was higher than the percentage of radiation necrosis.

For patients who did not undergo an operation, clinical and radiologic follow-up following the research MR imaging was used to classify the lesion as radiation necrosis or tumor recurrence. Two neuroradiologists, blinded to the perfusion analysis, assessed the change in the morphology and size of the enhancing lesion on the follow-up MR imaging using conventional and diffusion-weighted sequences. The presence of a centrally restricted diffusion pattern in an enhancing lesion was found to be associated with a treatment-related effect.¹⁰ Thus, our criteria for tumor recurrence were the following: 1) presence of a nodular or solid component that progressively increases in size with time; and 2) absence of restricted diffusion in the central area of necrosis on visual assessment of the b = 1000 images and the ADC map. Criteria for radiation necrosis were the following: 1) a peripheral rim of enhancement that remains stable or decreases in size, or 2) the presence of a central area of restricted diffusion on visual assessment of the b = 1000 images and the ADC map. The 2 neuroradiologists agreed on the classification of 12 of 14 lesions. Consensus reading was obtained for the remaining 2 lesions.

Statistical Analysis

For each lesion, the spatial coefficient of variation for each perfusion parameter was calculated as the SD of the parameter divided by the mean value of the parameter within the tumor ROI.¹¹

For certain continuous clinical variables (such as age and duration between MR imaging and the operation) and for each perfusion parameter, differences between the 2 patient groups were assessed via Mann-Whitney U tests. While P values < α = .05 were considered statistically significant, a stepwise Holm Bonferroni procedure was applied to minimize the potential for type I errors due to multiple comparisons.¹² Associations between variables were investigated using the Spearman coefficients of rank correlation. Associations with coefficients of >0.8 were considered very strong; between 0.60 and 0.79, strong; between 0.40 and 0.59, moderate; and <0.39, weak. Receiver operator characteristic (ROC) curves were constructed for each variable and the area under each ROC was evaluated. Both the standard error of the area under each ROC curve and comparisons between areas under each ROC curve were evaluated using the method of DeLong et al.¹³ All data were analyzed using MedCalc (Version 12; MedCalc Software).