Diagnostic accuracy of 2-hydroxyglutarate magnetic resonance spectroscopy in newly diagnosed brain mass and suspected recurrent gliomas

Diagnostic accuracy of 2-hydroxyglutarate magnetic resonance spectroscopy in newly diagnosed... Abstract Background Isocitrate dehydrogenase (IDH) mutations result in abnormal accumulation of 2-hydroxyglutarate (2HG) in gliomas that can be detected by MRS. We examined the diagnostic accuracy of 2HG single-voxel spectroscopy (SVS) and chemical shift imaging (CSI) in both newly diagnosed and posttreatment settings. Methods Long echo time (97 ms) SVS and CSI were acquired in 85 subjects, including a discovery cohort of 39 patients who had postoperative residual or recurrent glioma with confirmed IDH-mutation status and 6 normal volunteers, a prospective preoperative validation cohort of 24 patients with newly diagnosed brain mass, and a prospective recurrent-lesion validation cohort of 16 previously treated IDH-mutant glioma patients with suspected tumor recurrence. The optimal thresholds for both methods in diagnosing IDH status were determined by receiver operating characteristic analysis in the discovery cohort and then applied to the 2 validation cohorts to assess the diagnostic performance. Results The optimal 2HG/creatine thresholds of SVS and 75th percentile CSI for IDH mutations were 0.11 and 0.23, respectively. When applied to the validation sets, the sensitivity, specificity, and accuracy in distinguishing IDH-mutant gliomas in the preoperative cohort were 85.71%, 100.00%, and 94.12% for SVS, and 100.00%, 69.23%, and 81.82% for CSI, respectively. In the recurrent-lesion cohort, the sensitivity, specificity, and accuracy for discriminating IDH-positive recurrent gliomas were 40.00%, 62.50%, and 53.85% for SVS, and 66.67%, 100.00%, and 86.67% for CSI, respectively. Conclusions 2HG MRS provides diagnostic utility for IDH-mutant gliomas both preoperatively and at time of suspected tumor recurrence. SVS has a better diagnostic performance for untreated IDH-mutant gliomas, whereas CSI demonstrates greater performance in identifying recurrent tumors. chemical shift imaging, diagnostic performance, 2-hydroxyglutarate, isocitrate dehydrogenase mutations, single voxel spectroscopy Mutations of isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2) are found in about 80% of lower-grade gliomas (World Health Organization [WHO] grades II and III) and about 85% of secondary glioblastomas.1–4 Mutant IDHs gain the neomorphic ability to catalyze the nicotinamide adenine dinucleotide phosphate–dependent reduction of α-ketoglutarate to 2-hydroxyglutarate (2HG), resulting in accumulation of oncometabolite 2HG in IDH-positive tumor cells.5 Compared with IDH-wildtype tumors or normal brain tissue, 2HG level is increased by hundreds of times in IDH-mutant gliomas,5,6 making 2HG measurement a potential diagnostic marker in distinguishing IDH-mutant gliomas from other brain mass. Magnetic resonance spectroscopy (MRS) is a non-invasive technique which can examine the concentration of metabolites in brain tumors. Recent advances in MRS technique have demonstrated that the signal of the oncometabolite 2HG can be measured in vivo.7–9 Choi et al showed that the single-voxel point-resolved spectroscopy (PRESS) sequence with an optimized long echo time (TE) of 97 ms generated a well-defined narrow 2HG peak at 2.25 ppm, leading to improved differentiation between 2HG and the adjacent glutamate, glutamine, and γ-aminobutyric acid (GABA) signals.9 Also, 2HG MRS has been used in longitudinal follow-up in 76 patients serially.10 The sensitivity of the 97 ms TE single-voxel spectroscopy (SVS) for diagnosing IDH-mutant gliomas was reported to range from 8% to 91%,10,11 which was reflective of patients over the full spectrum of preoperation, postoperation, posttreatment, and recurrence. It is likely that the diagnostic performance may be dependent on the clinical status of the patients at the time of the scan. Compared with short spatial coverage of SVS, chemical shift imaging (CSI) can provide metabolic information for spatially heterogeneous tumor, and has been applied to detect 2HG concentrations in gliomas.8 This approach also has an advantage for the detection of 2HG in a posttreatment setting when nontumor regions are often indistinguishable on standard MRI. In this study, we optimize the threshold in a retrospective cohort of patients for each method in differentiating IDH-mutant gliomas from non-IDH-mutant controls, and then validate them in 2 prospective cohorts of (i) preoperative patients presenting with unknown brain mass, and (ii) postoperative patients with known IDH-mutation status presenting with suspected recurrence to determine the diagnostic value of 2HG MRS under these 2 clinically relevant settings. Materials and Methods Patient Population This single-institution study was approved by the Brigham and Women’s Hospital institutional review board and conducted in compliance with the Health Insurance Portability and Accountability Act. We obtained written informed consent for each subject. The patient cohort consisted of 3 groups: a discovery cohort, a prospective preoperative validation cohort, and a prospective recurrent-lesion validation cohort. The details of the 3 cohorts were specified: 1) The discovery cohort consisted of normal volunteers and subjects with postoperative residual or recurrent glioma retrospectively identified from patients who had MRS using the following inclusion criteria: (i) a 97 ms TE SVS and/or a 97 ms TE CSI acquisition was available and of acceptable spectral quality as detailed in the MRS processing section; (ii) residual or recurrent tumor visible on brain MRI larger than 1.5 × 1.5 × 1.5 cm (axial, sagittal, coronal) to ensure sufficient tissue to detect 2HG signal; (iii) the IDH-mutation status was known, based on immunohistochemistry or gene sequencing. 2) The preoperative validation cohort included prospectively enrolled patients with newly diagnosed brain mass, and MRS exams were obtained as part of standard preoperative evaluation. 3) The recurrent-lesion validation cohort included prospectively enrolled patients with previously treated IDH-mutant glioma presenting with suspected tumor recurrence. The presence of recurrent tumor was confirmed by surgery or 6-month radiological follow-up according to Response Assessment in Neuro-Oncology (RANO) criteria.12,13 From April 2014 to November 2017, 2HG MRS was performed in 96 subjects including a discovery cohort consisting of 42 patients and 6 normal volunteers; a preoperative validation cohort of 32 patients; and a recurrent-lesion validation cohort of 16 patients. Determination of IDH-Mutation Status IDH mutations were determined using immunohistochemistry,14,15 mass spectrometry–based mutation genotyping (OncoMap, Sequenom),15,16 or multiplex exome sequencing (OncoPanel, Illumina),17,18 depending on which genotyping technologies were available at the time of diagnosis. All sequencing assays were performed by the Molecular Diagnostics Division of the Brigham and Women’s Hospital Center for Advanced Molecular Diagnostics, a laboratory environment certified by the Clinical Laboratory Improvement Amendments, without knowledge of the results of the MRS. MRI and MRS Protocol All MRI and MRS exams were performed on one clinical 3.0T MRI scanner (Siemens TIM Skyra) with a 32-channel head coil. Prior to spectroscopy, sagittal 3D fluid attenuated inversion recovery images (FLAIR, repetition time [TR]/inversion time [TI]/TE = 9000 ms/2500 ms/81 ms, field of view = 20 × 20 cm, matrix = 224 × 320) were acquired and reconstructed in the axial and coronal planes with 2 mm slice resolution for accurate localization of the voxel. The MRS protocol included a long-TE (97 ms) single-voxel PRESS sequence and a long-TE (97 ms) semi-LASER (localization by adiabatic selective refocusing) CSI sequence. For the 97 ms TE single-voxel PRESS sequence, the acquisition parameters were: volume of interest = 20 × 20 × 20 mm3, TR/TE = 2 s/97 ms, 128 averages, 833 ms dwell time, 1024 points, and total time = 4:26 minutes. The region-of-interest (ROI) was chosen under the direction of a neuroradiologist (R.Y.H.) to include as much of the lesion as possible while avoiding the surrounding tissue. The voxel for normal volunteers was positioned in the centrum semiovale. Localized shimming was performed by adjustment of first- and second-order shim gradients using the automatic 3-dimensional B0 field mapping technique (Siemens) followed by manual adjustment of the above-mentioned shim gradients to achieve a magnitude peak width of water at half-maximum of 14 Hz or less. Manual shimming was utilized in our study to ensure consistent data quality among all patients so that shimming would not be a variable. After frequency adjustment, water-selective suppression was optimized by using the water suppression enhanced through T1 effects (WET) technique. For the 97 ms TE semi-LASER CSI sequence, the acquisition parameters included: field of view = 160 × 160 × 15 mm3, matrix = 16 × 16 for a voxel resolution = 10 × 10 × 15 mm3, TR/TE = 1700 ms/97 ms, 3 averages. Acceleration was enabled using a weighted distribution, resulting in a total time of 6:53 minutes. The ROI was positioned at the same level of the SVS voxel and was adjusted to cover as much of the lesion as possible, as well as the bilateral normal tissue while avoiding the side of the skull or other areas of susceptibility. Four saturation bands were placed along the margin of the ROI with optimum orientation to minimize lipid contamination from subcutaneous fat. Localized shimming was performed in the same way as was done for the single-voxel PRESS sequence to achieve a magnitude peak width of water at half-maximum of 25 Hz or less. Water suppression was accomplished with the WET technique. MRS Data Processing LCModel v6.2 software was used for both the SVS and CSI spectral fitting, using simulated spectra of 20 metabolites including 2HG as a customized basis set.9 For the CSI, the ROI was reconstructed on a VD17B scanner (Siemens), and the voxels contained within the abnormal hyperintense area on FLAIR images were determined by a neuroradiologist (R.Y.H.). Those spectra were then selected for processing using LCModel. Ratios of 2HG to creatine (2HG/Cr) were obtained and included in the further analysis. Signal-to-noise (SNR) ratio and full width at half maximum (FWHM) were used to assess the quality of the data. Spectra with an SNR <5 or FWHM of Cr peak >0.143 ppm were excluded for both the SVS and the spectra of the selected CSI voxels due to poor quality. For CSI, the 75th percentile 2HG/Cr values of the selected voxels were then calculated for each subject. Postprocessing was done without knowledge of patient IDH status. Statistical Analysis We first performed phantom experiments to establish that 2HG could be differentiated from other metabolites and quantified accurately by both the SVS and CSI methods. This was done using a phantom at pH 7.0 with different metabolites composed of 2HG (4 mM), glutamate, glutamine, GABA, myoinositol, glycine, lactate, N-acetylaspartate, Cr (4 mM), and choline. The SVS and CSI sequences were repeated 6 times. The means, standard deviations, and coefficients of variance for both the 2HG/Cr ratios from SVS and the 75th percentile 2HG/Cr values of CSI were calculated. Reproducibility of both the SVS and the CSI measurements was then assessed by test-retest analysis, in which a series of patients were scanned twice with the voxels placed in the same location. The patients either underwent an initial scan, got out of the scanner for 5 minutes, and then were repositioned and rescanned10 or were scanned twice within the same session based on the time restraints of clinical practices at the time of scanning. In the latter case, the 3D FLAIR sequence and the MRS sequences were repeated by another experienced technologist using the procedure described above. Scatter plot and correlation coefficient between test and retest measurements were presented. Using receiver operating characteristic (ROC) analysis, the sensitivity and specificity of 2HG/Cr ratios for diagnosing IDH-mutant gliomas in the discovery cohort were calculated for SVS and 75th percentile values of CSI. The optimal cutoff values for SVS and 75th percentile values of CSI were respectively chosen based on the analysis of the discovery cohort to optimize specificity with sensitivity above 50%. We then applied the cutoff values to both validation groups to examine the diagnostic performance of 2HG/Cr ratios for IDH-mutant gliomas and IDH-mutant glioma recurrence. The 75th percentile 2HG/Cr values of CSI were prepared with Microsoft Excel 2011, and all other analyses were performed using SPSS v21 (IBM). In addition, we performed post-hoc sample size calculations to check if the sample sizes that were used in the current studies were enough to detect the difference of diagnostic performance between the SVS method and the CSI method with type I error rate of 0.05 and power of 0.7. We applied the sample size calculation method proposed by Beam et al.19 θ1 and θ2 denote the diagnostic performance measures (eg, sensitivity, specificity, or diagnostic accuracy) for SVS and CSI, respectively. We would like to test H0: θ1−θ2=0, versus Ha: θ1−θ2=Δ . We assumed each group had equal sample size n. The formula to calculate n is shown below: n=[zα/2V0+zβVa]2Δ2, where V0=Var(θ̂1−θ̂2|H0)=φa+Δ2, Va=Var(θ̂1−θ̂2|Ha)=φa. Statistical software could provide estimates of variances for each method Var(θ̂1) and Var(θ̂1), but not the estimate of the variance of the difference Var(θ̂1−θ̂2). Note that θ̂1 and θ̂2 were dependent, since they were calculated based on the same set of subjects using different methods. To estimate V0=Var(θ̂1−θ̂2|H0), we performed a permutation analysis. Specifically, we generated 10000 permuted datasets. In each permuted dataset, we permuted the values of true positives and true negatives and kept 2HG/Cr value unchanged. We then detected if a subject was a positive or negative based on the cutoff values generated above. Next, we recalculated sensitivity, specificity, and accuracy. These values were obtained under the null hypothesis H0: θ1−θ2=0. We could calculate 10000 different θ̂1−θ̂2 based on the 10000 permuted datasets. Hence, we could estimate the variance of θ̂1−θ̂2 under H0: V0=Var(θ̂1−θ̂2|H0). To estimate Va=Var(θ̂1−θ̂2|Ha), we performed a bootstrapping analysis. Specifically, we generated 10000 bootstrapping datasets. In each bootstrapping dataset, we resampled subjects with replacement so that some subjects would be sampled multiple times and some subjects would not be sampled to a bootstrapping sample. We then detected if a subject was a positive or negative based on the cutoff values. Next, we recalculated sensitivity, specificity, and accuracy. These values were obtained under the alternative hypothesis Ha: θ1−θ2=0. We could calculate 10000 different θ̂1−θ̂2 based on the 10000 bootstrapping datasets. Hence, we could estimate the variance of θ̂1−θ̂2 under Va=Var(θ̂1−θ̂2|Ha). All power analyses were performed using R v3.3.2. Results Patient Cohorts Seven patients in the preoperative cohort were excluded, since no histological diagnosis of the IDH status was made at the time of this study, 3 patients in the discovery cohort were excluded due to small tumor volume, and 1 patient in the preoperative cohort was excluded due to poor spectrum quality. A total of 85 subjects were included in this study, including (i) a discovery cohort of 39 glioma patients (35 IDH-mutant glioma patients and 4 IDH-wildtype patients) and 6 normal volunteers (n = 31 for SVS, and n = 41 for CSI), (ii) a prospective preoperative validation cohort of 24 patients (n = 13 for SVS, and n = 22 for CSI), and (iii) a prospective recurrent-lesion validation cohort of 16 patients (n = 13 for SVS, and n = 15 for CSI). Of note, the patient numbers were different for SVS and CSI, either because only SVS or CSI was obtained for some patients or because SVS or CSI was excluded due to not meeting the quality control criteria. The patient clinical and pathologic characteristics are summarized for each cohort in Table 1. The detailed clinical information for patients of the recurrent-lesion validation cohort are given in Supplementary Table S1. Representative spectra of an IDH-mutant and an IDH-wildtype glioma patient are shown in Figs. 1 and 2. Table 1. Patient clinical and pathologic characteristics Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 View Large Table 1. Patient clinical and pathologic characteristics Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 View Large Fig. 1 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the discovery cohort. The white box in panel A shows the SVS location. A prominent 2HG peak was seen at 2.25 ppm in both SVS (2HG/Cr = 0.682, panel B) and the selected CSI spectrum (2HG/Cr = 0.852, panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectrum (D). The 75th percentile 2HG/Cr value of CSI was 0.719. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 1 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the discovery cohort. The white box in panel A shows the SVS location. A prominent 2HG peak was seen at 2.25 ppm in both SVS (2HG/Cr = 0.682, panel B) and the selected CSI spectrum (2HG/Cr = 0.852, panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectrum (D). The 75th percentile 2HG/Cr value of CSI was 0.719. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 2 View largeDownload slide Representative spectroscopies from an IDH-wildtype glioma patient in the preoperative cohort. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in either SVS (panel B) or the selected spectroscopy of CSI (panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectroscopy (D). The 75th percentile 2HG/Cr value of the CSI was 0. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 2 View largeDownload slide Representative spectroscopies from an IDH-wildtype glioma patient in the preoperative cohort. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in either SVS (panel B) or the selected spectroscopy of CSI (panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectroscopy (D). The 75th percentile 2HG/Cr value of the CSI was 0. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Variability of 2HG Measurements For the phantom scans, the mean 2HG/Cr values were 0.968 and 0.952 for the SVS and CSI approaches, respectively, which were in close agreement with the actual ratio of 1. In addition, the standard deviations and the coefficients of variation were 0.036, 0.034 and 0.037, 0.036 for the SVS and CSI methods, respectively, which were readily attributed to machine noise. These data indicated that 2HG could be quantified accurately by both SVS and CSI in this study. Variability of 2HG measurements was assessed by test-retest analysis in 16 patients. Among them, the single-voxel PRESS method was repeated in 10 patients, whereas the CSI approach was repeated in 9 patients. For SVS, the 2HG/Cr ratio difference between scans 1 and 2 ranged between −0.05 and 0.05, with a test-retest correlation of 0.998. For CSI, the 75th percentile 2HG/Cr ratio difference between scans 1 and 2 varied from −0.11 to 0.11, with test-retest correlation of 0.969. These data establish the high reliability of 2HG/Cr quantitation by both the SVS and CSI methods over the tested range (0.00–1.13 for SVS, and 0.00–0.59 for the 75th percentile 2HG/Cr ratios of CSI). The scatter plots are presented in Supplementary Figure S1. Determination of the Optimal Cutoff Values in the Discovery Cohort Using ROC analysis, we investigated the diagnostic accuracy of 2HG/Cr ratios in determining IDH-mutation status in the discovery cohort. The areas under the curves were 0.83 (95% CI: 0.68–0.99) and 0.83 (0.67–0.98) for SVS and the 75th percentile values of CSI, respectively (Supplementary Figure S2). For SVS, 2HG/Cr threshold of 0.11 yielded the highest specificity at sensitivity above 50.00%. In terms of 75th percentile values of CSI, 2HG/Cr threshold of 0.23 generated the highest specificity at sensitivity above 50.00%. The sensitivity, specificity, and diagnostic accuracy corresponding to the potential cutoff values for both methods are listed in Table 2. Table 2 Potential 2HG/Cr cutoff values and the corresponding sensitivity, specificity, and accuracy in the discovery cohort‡ Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) ‡The optimal cutoff values are marked in bold. View Large Table 2 Potential 2HG/Cr cutoff values and the corresponding sensitivity, specificity, and accuracy in the discovery cohort‡ Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) ‡The optimal cutoff values are marked in bold. View Large Diagnostic Performance of 2HG MRS in Preoperative and Recurrent-Lesion Validation Cohorts In the preoperative cohort, using the optimal 2HG/Cr thresholds resulted in a sensitivity, specificity, diagnostic accuracy, positive predictive value, and negative predictive value of 85.71%, 100.00%, 94.12%, 100.00%, and 90.91%, respectively, in distinguishing IDH-mutant gliomas and IDH-wildtype controls for SVS, and 100.00%, 69.23%, 80.00%, 69.23%, and 100.00%, respectively, for CSI (Table 3). Table 3 Sensitivity, specificity, and accuracy corresponding to the optimal 2HG/Cr cutoff values in the validation cohorts Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) View Large Table 3 Sensitivity, specificity, and accuracy corresponding to the optimal 2HG/Cr cutoff values in the validation cohorts Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) View Large When the optimal 2HG/Cr thresholds were applied to the recurrent-lesion cohort, the sensitivity, specificity, diagnostic accuracy, positive predictive value, and negative predictive value for discriminating IDH-mutant glioma recurrence were 40.00%, 62.50%, 53.85%, 40.00%, and 62.50%, respectively, for SVS; and 66.67%, 100.00%, 86.67%, 100.00%, and 81.82%, respectively, for CSI (Table 3). Representative results from a posttreatment IDH-mutant glioma patient with surgically confirmed recurrence are shown in Fig. 3. Existence and uneven distribution of 2HG were observed in CSI with the 75th percentile 2HG/Cr value of 0.68. However, no 2HG peak was seen in SVS. Fig. 3 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the recurrent-lesion cohort. The presence of recurrent tumor was confirmed by surgery. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in SVS (panel B). However, CSI demonstrated the existence and the heterogeneous distribution of 2HG (panels C, D, E). The 2HG/Cr ratios ranged from 0.00 (panel E) to 3.50 (panel D), with the 75th percentile value of 0.68. The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red boxes show the selected voxels for the representative CSI spectra (panels D, E). In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 3 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the recurrent-lesion cohort. The presence of recurrent tumor was confirmed by surgery. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in SVS (panel B). However, CSI demonstrated the existence and the heterogeneous distribution of 2HG (panels C, D, E). The 2HG/Cr ratios ranged from 0.00 (panel E) to 3.50 (panel D), with the 75th percentile value of 0.68. The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red boxes show the selected voxels for the representative CSI spectra (panels D, E). In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Power Analysis The required sample sizes given type I error rate of 0.05 and power 0.7 are shown in Supplementary Table S2. We had enough power to detect the difference of diagnostic performance between the SVS method and the CSI method in both settings. Discussion In this study we prospectively evaluated the diagnostic accuracy of 2HG MRS in identifying IDH-mutant gliomas in patients presenting with newly discovered brain mass as well as in patients with treated IDH-mutant tumors presenting with suspected tumor recurrence. Our results demonstrate that 2HG MRS using the SVS method is highly accurate in diagnosing IDH-mutant gliomas among newly diagnosed brain mass, whereas 2HG MRS using the CSI method provides good accuracy in identifying recurrent IDH-mutant glioma. Several prior studies have reported the 2HG absolute concentration thresholds to distinguish between IDH-mutant and IDH-wildtype gliomas with 97 ms TE single-voxel PRESS sequence, and the most commonly used threshold was the 2HG concentration ≥1 mM with or without the Cramér-Rao lower bound of 2HG ≤30%.10,11 In this study, we used 2HG/Cr ratio rather than absolute 2HG concentrations to diagnose IDH mutations. One reason is that the acquisition of the CSI water reference sequence takes at least 4:30 minutes with a single average. This increased scanning time is clinically challenging due to resource utilization constraints in busy neuroradiology practices, and increases the likelihood for motion artifacts during the CSI-CSI water reference session of about 12 minutes. Furthermore, 2HG absolute quantification by water reference also suffers from the variation of water concentration.8,9 The 2HG absolute quantification studies apply a constant water concentration of normal brain as reference in tumors, assuming that the water concentration is constant between normal brain and tumor tissues, and is invariable among all gliomas. However, prior studies have shown that the brain water content can be increased up to 50% in brain tumors over that in healthy subjects,20,21 and water concentration varies in individual glioma based on a range of factors including edema severity, tumor cellularity, and historical type.8 Although a relatively accurate water concentration estimation method has been proposed,22 the edema, cysts, and surgical cavities need to be accurately segmented, while Cr excludes these confoundings.23 Moreover, endogenous Cr signal reference for 2HG quantification has been used in a recently published paper.23 For better comparison, ratios to Cr were chosen for both the SVS and CSI spectral quantification, which obviated the need of acquiring a water reference sequence and limited our protocol to under 6 minutes for both SVS and CSI acquisitions. The ROC analysis in the discovery cohort yielded a 2HG/Cr cutoff value of 0.11 for SVS, and 0.23 for the 75th percentile 2HG/Cr values of CSI. Assuming a total Cr level between 8 and 10 mmol/L,23 the SVS 2HG/Cr threshold of 0.11 would correspond to a 2HG level around 0.88–1 mmol/L, which is comparable but a little lower than the absolute 2HG concentration threshold reported previously. It is also important to note why there are different thresholds for SVS and CSI 2HG measures: The optimal threshold value for the SVS technique was selected based on the ratio of 2HG to Cr concentrations over our training study population. The optimal threshold value for the CSI technique was calculated based on the mean value of the population of voxels in the top 75th percentile 2HG/Cr concentration, an area that is more representative of likely tumor. By using such a top quartile approach, the sensitivity of detecting a small area of true tumor is expected to increase compared with averaging of the entire voxel area. In other words, if only true tumor voxels were selected and analyzed during CSI analysis, the threshold value would be identical to SVS (such as shown in our phantom results), but this is exactly the main difficulty with posttreatment imaging where tumor voxels are not readily differentiable from nontumor tissues. In the preoperative cohort with newly diagnosed brain mass, we demonstrated that the SVS method achieved an accuracy of 94.12% in diagnosing IDH-mutant lesions preoperatively. This result was concordant with those previously published by Choi et al in newly diagnosed brain mass using 97 ms TE SVS.10 We also found that the SVS approach resulted in higher specificity and accuracy compared with the CSI method. The better diagnostic performance of the SVS could be due to the larger voxel size during the SVS acquisition compared with that during the CSI acquisition, resulting in higher SNRs, which are of benefit in distinguishing 2HG peak from the adjacent resonances of glutamate, glutamine, and GABA. In the cohort with known IDH-mutant tumors presenting with suspected tumor recurrence, the CSI method showed higher sensitivity, specificity, and accuracy in diagnosing IDH-mutant glioma recurrence compared with the SVS approach. Due to the larger spatial coverage and multivoxel acquisition, CSI has the advantage of accounting for the spatial heterogeneity of the abnormal tissues, including gliosis, edema, and tumor, which commonly coexist in the posttreatment setting.8,23 Additionally, because of the smaller volumes of the CSI voxels, the metabolic measures by CSI contain reduced partial-volume effects compared with SVS8 (Fig. 3). These potential advantages of the CSI method need validation by direct tissue sampling using imaging guided resection or biopsy. It is noted that the specificity of CSI was higher than that of SVS. Subgroup analysis was performed based on WHO grade and showed the same results in both the WHO grades II and III groups. Moreover, we explored other possible reasons for the SVS false positives, including spectral quality, tumor volume, and outcome measure, and found them all to be consistent. No specific reason was found to explain the relative lower specificity of SVS versus CSI, and this warrants further investigation. Our study is limited by small patient sample sizes in the validation groups, especially the recurrent-lesion group. While the accuracy of 2HG MRS has been demonstrated in our single-institution study, the diagnostic accuracy in both clinical settings requires further validation by larger prospective trials. Our analysis of the recurrent-tumor cohort is also limited by lack of histological evaluation of all patients but is dependent on follow-up clinical and imaging stability. In conclusion, 2HG MRS provides diagnostic utility for IDH-mutant gliomas both preoperatively and at time of suspected tumor recurrence. SVS has a better diagnostic performance for IDH-mutant gliomas in the untreated patient cohort, whereas CSI demonstrates greater performance in identifying IDH-mutant glioma recurrence. Supplementary Material Supplementary material is available at Neuro-Oncology online. Funding This work was supported by Brigham and Women’s Hospital Institute for the Neurosciences seed grant and an ARRS/ASNR Scholar Award. Conflict of interest statement. David A. Reardon serves as a paid advisory board member for Abbvie, Amgen, BMS, Cavion, Celldex, EMD Serono, Genentech/Roche, Inovio, Juno Pharmaceuticals, Merck, Midatech, Momenta Pharmaceuticals, Novartis, Novocure, Oncorus, Oxigene, Regeneron, and Stemline Terapeutics; receives lab research support (paid to Dana-Farber) from Aceta Pharma, Agenus, Celldex Therapeutics, EMD Serono, Incyte, Inovio, Midatech, and Tragara; is a paid speaker for Genentech/Roche and Merck. Patrick Y. Wen reports grants, personal fees, and nonfinancial support from Agios, nonfinancial support from Angiochem, GlaxoSmithKline, Sanofi-Aventis, and VBI Vaccines, personal fees and nonfinancial support from AstraZeneca, Genentech/Roche, and Vascular Biogenics, personal fees from Cavion, Insys, Monteris, Novogen, Regeneron, and Tocogen. Alexander P. Lin is a consultant for Agios Pharmaceuticals, cofounder of BrainSpec, and consultant to Moncton MRI. No other authors have a conflict of interest. Conflict of interest statement. David A. Reardon serves as a paid advisory board member for Abbvie, Amgen, BMS, Cavion, Celldex, EMD Serono, Genentech/Roche, Inovio, Juno Pharmaceuticals, Merck, Midatech, Momenta Pharmaceuticals, Novartis, Novocure, Oncorus, Oxigene, Regeneron, and Stemline Terapeutics; receives lab research support (paid to Dana-Farber) from Aceta Pharma, Agenus, Celldex Therapeutics, EMD Serono, Incyte, Inovio, Midatech, and Tragara; is a paid speaker for Genentech/Roche and Merck. Patrick Y. Wen reports grants, personal fees, and nonfinancial support from Agios, nonfinancial support from Angiochem, GlaxoSmithKline, Sanofi-Aventis, and VBI Vaccines, personal fees and nonfinancial support from AstraZeneca, Genentech/Roche, and Vascular Biogenics, personal fees from Cavion, Insys, Monteris, Novogen, Regeneron, and Tocogen. Alexander P. Lin is a consultant for Agios Pharmaceuticals, cofounder of BrainSpec, and consultant to Moncton MRI. No other authors have a conflict of interest. Acknowledgments The authors would like to thank the China Scholarship Council for financial support for Min Zhou and Yue Zhou. The authors would also like to thank Dr Changho Choi for the use of his basis set and guidance as well as Dr Weiliang Qiu for the sample size calculations. References 1. Eckel-Passow JE , Lachance DH , Molinaro AM , et al. Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors . N Engl J Med . 2015 ; 372 ( 26 ): 2499 – 2508 . Google Scholar CrossRef Search ADS PubMed 2. Brat DJ , Verhaak RG , Aldape KD , et al. ; Cancer Genome Atlas Research Network . Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas . N Engl J Med . 2015 ; 372 ( 26 ): 2481 – 2498 . Google Scholar CrossRef Search ADS PubMed 3. Yan H , Parsons DW , Jin G , et al. IDH1 and IDH2 mutations in gliomas . N Engl J Med . 2009 ; 360 ( 8 ): 765 – 773 . Google Scholar CrossRef Search ADS PubMed 4. Parsons DW , Jones S , Zhang X , et al. An integrated genomic analysis of human glioblastoma multiforme . Science . 2008 ; 321 ( 5897 ): 1807 – 1812 . Google Scholar CrossRef Search ADS PubMed 5. Dang L , White DW , Gross S , et al. Cancer-associated IDH1 mutations produce 2-hydroxyglutarate . Nature . 2009 ; 462 ( 7274 ): 739 – 744 . Google Scholar CrossRef Search ADS PubMed 6. Jin G , Reitman ZJ , Duncan CG , et al. Disruption of wild-type IDH1 suppresses D-2-hydroxyglutarate production in IDH1-mutated gliomas . Cancer Res . 2013 ; 73 ( 2 ): 496 – 501 . Google Scholar CrossRef Search ADS PubMed 7. Pope WB , Prins RM , Albert Thomas M , et al. Non-invasive detection of 2-hydroxyglutarate and other metabolites in IDH1 mutant glioma patients using magnetic resonance spectroscopy . J Neurooncol . 2012 ; 107 ( 1 ): 197 – 205 . Google Scholar CrossRef Search ADS PubMed 8. Choi C , Ganji SK , DeBerardinis RJ , et al. 2-hydroxyglutarate detection by magnetic resonance spectroscopy in IDH-mutated patients with gliomas . Nat Med . 2012 ; 18 ( 4 ): 624 – 629 . Google Scholar CrossRef Search ADS PubMed 9. Choi C , Ganji S , Hulsey K , et al. A comparative study of short- and long-TE ¹H MRS at 3 T for in vivo detection of 2-hydroxyglutarate in brain tumors . NMR Biomed . 2013 ; 26 ( 10 ): 1242 – 1250 . Google Scholar CrossRef Search ADS PubMed 10. Choi C , Raisanen JM , Ganji SK , et al. Prospective longitudinal analysis of 2-hydroxyglutarate magnetic resonance Spectroscopy identifies broad clinical utility for the management of patients with IDH-mutant glioma . J Clin Oncol . 2016 ; 34 ( 33 ): 4030 – 4039 . Google Scholar CrossRef Search ADS PubMed 11. de la Fuente MI , Young RJ , Rubel J , et al. Integration of 2-hydroxyglutarate-proton magnetic resonance spectroscopy into clinical practice for disease monitoring in isocitrate dehydrogenase-mutant glioma . Neuro-Oncol . 2016 ; 18 ( 2 ): 283 – 290 . Google Scholar CrossRef Search ADS PubMed 12. Wen PY , Macdonald DR , Reardon DA , et al. Updated response assessment criteria for high-grade gliomas: Response Assessment in Neuro-Oncology working group . J Clin Oncol . 2010 ; 28 ( 11 ): 1963 – 1972 . Google Scholar CrossRef Search ADS PubMed 13. van den Bent MJ , Wefel JS , Schiff D , et al. Response Assessment in Neuro-Oncology (a report of the RANO group): assessment of outcome in trials of diffuse low-grade gliomas . Lancet Oncol . 2011 ; 12 ( 6 ): 583 – 593 . Google Scholar CrossRef Search ADS PubMed 14. Capper D , Zentgraf H , Balss J , Hartmann C , von Deimling A . Monoclonal antibody specific for IDH1 R132H mutation . Acta Neuropathol . 2009 ; 118 ( 5 ): 599 – 601 . Google Scholar CrossRef Search ADS PubMed 15. Ramkissoon SH , Bi WL , Schumacher SE , et al. Clinical implementation of integrated whole-genome copy number and mutation profiling for glioblastoma . Neuro Oncol . 2015 ; 17 ( 10 ): 1344 – 1355 . Google Scholar CrossRef Search ADS PubMed 16. Thomas RK , Baker AC , Debiasi RM , et al. High-throughput oncogene mutation profiling in human cancer . Nat Genet . 2007 ; 39 ( 3 ): 347 – 351 . Google Scholar CrossRef Search ADS PubMed 17. Cryan JB , Haidar S , Ramkissoon LA , et al. Clinical multiplexed exome sequencing distinguishes adult oligodendroglial neoplasms from astrocytic and mixed lineage gliomas . Oncotarget . 2014 ; 5 ( 18 ): 8083 – 8092 . Google Scholar CrossRef Search ADS PubMed 18. Wagle N , Berger MF , Davis MJ , et al. High-throughput detection of actionable genomic alterations in clinical tumor samples by targeted, massively parallel sequencing . Cancer Discov . 2012 ; 2 ( 1 ): 82 – 93 . Google Scholar CrossRef Search ADS PubMed 19. Beam CA , Baker ME , Paine SS , Sostman HD , Sullivan DC . Answering unanswered questions: proposal for a shared resource in clinical diagnostic radiology research . Radiology . 1992 ; 183 ( 3 ): 619 – 620 . Google Scholar CrossRef Search ADS PubMed 20. Grasso G , Alafaci C , Passalacqua M , et al. Assessment of human brain water content by cerebral bioelectrical impedance analysis: a new technique and its application to cerebral pathological conditions . Neurosurgery . 2002 ; 50 ( 5 ): 1064 – 1072 ; discussion 1072–1074. Google Scholar PubMed 21. Jansen JF , Backes WH , Nicolay K , Kooi ME . 1H MR spectroscopy of the brain: absolute quantification of metabolites . Radiology . 2006 ; 240 ( 2 ): 318 – 332 . Google Scholar CrossRef Search ADS PubMed 22. Gasparovic C , Song T , Devier D , et al. Use of tissue water as a concentration reference for proton spectroscopic imaging . Magn Reson Med . 2006 ; 55 ( 6 ): 1219 – 1226 . Google Scholar CrossRef Search ADS PubMed 23. Andronesi OC , Loebel F , Bogner W , et al. Treatment response assessment in idh-mutant glioma patients by noninvasive 3D functional spectroscopic mapping of 2-hydroxyglutarate . Clin Cancer Res . 2016 ; 22 ( 7 ): 1632 – 1641 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Society for Neuro-Oncology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Neuro-Oncology Oxford University Press

Diagnostic accuracy of 2-hydroxyglutarate magnetic resonance spectroscopy in newly diagnosed brain mass and suspected recurrent gliomas

Neuro-Oncology , Volume 20 (9) – Sep 1, 2018

Loading next page...
 
/lp/ou_press/diagnostic-accuracy-of-2-hydroxyglutarate-magnetic-resonance-5f3nqi41N7
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Society for Neuro-Oncology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1522-8517
eISSN
1523-5866
D.O.I.
10.1093/neuonc/noy022
Publisher site
See Article on Publisher Site

Abstract

Abstract Background Isocitrate dehydrogenase (IDH) mutations result in abnormal accumulation of 2-hydroxyglutarate (2HG) in gliomas that can be detected by MRS. We examined the diagnostic accuracy of 2HG single-voxel spectroscopy (SVS) and chemical shift imaging (CSI) in both newly diagnosed and posttreatment settings. Methods Long echo time (97 ms) SVS and CSI were acquired in 85 subjects, including a discovery cohort of 39 patients who had postoperative residual or recurrent glioma with confirmed IDH-mutation status and 6 normal volunteers, a prospective preoperative validation cohort of 24 patients with newly diagnosed brain mass, and a prospective recurrent-lesion validation cohort of 16 previously treated IDH-mutant glioma patients with suspected tumor recurrence. The optimal thresholds for both methods in diagnosing IDH status were determined by receiver operating characteristic analysis in the discovery cohort and then applied to the 2 validation cohorts to assess the diagnostic performance. Results The optimal 2HG/creatine thresholds of SVS and 75th percentile CSI for IDH mutations were 0.11 and 0.23, respectively. When applied to the validation sets, the sensitivity, specificity, and accuracy in distinguishing IDH-mutant gliomas in the preoperative cohort were 85.71%, 100.00%, and 94.12% for SVS, and 100.00%, 69.23%, and 81.82% for CSI, respectively. In the recurrent-lesion cohort, the sensitivity, specificity, and accuracy for discriminating IDH-positive recurrent gliomas were 40.00%, 62.50%, and 53.85% for SVS, and 66.67%, 100.00%, and 86.67% for CSI, respectively. Conclusions 2HG MRS provides diagnostic utility for IDH-mutant gliomas both preoperatively and at time of suspected tumor recurrence. SVS has a better diagnostic performance for untreated IDH-mutant gliomas, whereas CSI demonstrates greater performance in identifying recurrent tumors. chemical shift imaging, diagnostic performance, 2-hydroxyglutarate, isocitrate dehydrogenase mutations, single voxel spectroscopy Mutations of isocitrate dehydrogenase 1 and 2 (IDH1 and IDH2) are found in about 80% of lower-grade gliomas (World Health Organization [WHO] grades II and III) and about 85% of secondary glioblastomas.1–4 Mutant IDHs gain the neomorphic ability to catalyze the nicotinamide adenine dinucleotide phosphate–dependent reduction of α-ketoglutarate to 2-hydroxyglutarate (2HG), resulting in accumulation of oncometabolite 2HG in IDH-positive tumor cells.5 Compared with IDH-wildtype tumors or normal brain tissue, 2HG level is increased by hundreds of times in IDH-mutant gliomas,5,6 making 2HG measurement a potential diagnostic marker in distinguishing IDH-mutant gliomas from other brain mass. Magnetic resonance spectroscopy (MRS) is a non-invasive technique which can examine the concentration of metabolites in brain tumors. Recent advances in MRS technique have demonstrated that the signal of the oncometabolite 2HG can be measured in vivo.7–9 Choi et al showed that the single-voxel point-resolved spectroscopy (PRESS) sequence with an optimized long echo time (TE) of 97 ms generated a well-defined narrow 2HG peak at 2.25 ppm, leading to improved differentiation between 2HG and the adjacent glutamate, glutamine, and γ-aminobutyric acid (GABA) signals.9 Also, 2HG MRS has been used in longitudinal follow-up in 76 patients serially.10 The sensitivity of the 97 ms TE single-voxel spectroscopy (SVS) for diagnosing IDH-mutant gliomas was reported to range from 8% to 91%,10,11 which was reflective of patients over the full spectrum of preoperation, postoperation, posttreatment, and recurrence. It is likely that the diagnostic performance may be dependent on the clinical status of the patients at the time of the scan. Compared with short spatial coverage of SVS, chemical shift imaging (CSI) can provide metabolic information for spatially heterogeneous tumor, and has been applied to detect 2HG concentrations in gliomas.8 This approach also has an advantage for the detection of 2HG in a posttreatment setting when nontumor regions are often indistinguishable on standard MRI. In this study, we optimize the threshold in a retrospective cohort of patients for each method in differentiating IDH-mutant gliomas from non-IDH-mutant controls, and then validate them in 2 prospective cohorts of (i) preoperative patients presenting with unknown brain mass, and (ii) postoperative patients with known IDH-mutation status presenting with suspected recurrence to determine the diagnostic value of 2HG MRS under these 2 clinically relevant settings. Materials and Methods Patient Population This single-institution study was approved by the Brigham and Women’s Hospital institutional review board and conducted in compliance with the Health Insurance Portability and Accountability Act. We obtained written informed consent for each subject. The patient cohort consisted of 3 groups: a discovery cohort, a prospective preoperative validation cohort, and a prospective recurrent-lesion validation cohort. The details of the 3 cohorts were specified: 1) The discovery cohort consisted of normal volunteers and subjects with postoperative residual or recurrent glioma retrospectively identified from patients who had MRS using the following inclusion criteria: (i) a 97 ms TE SVS and/or a 97 ms TE CSI acquisition was available and of acceptable spectral quality as detailed in the MRS processing section; (ii) residual or recurrent tumor visible on brain MRI larger than 1.5 × 1.5 × 1.5 cm (axial, sagittal, coronal) to ensure sufficient tissue to detect 2HG signal; (iii) the IDH-mutation status was known, based on immunohistochemistry or gene sequencing. 2) The preoperative validation cohort included prospectively enrolled patients with newly diagnosed brain mass, and MRS exams were obtained as part of standard preoperative evaluation. 3) The recurrent-lesion validation cohort included prospectively enrolled patients with previously treated IDH-mutant glioma presenting with suspected tumor recurrence. The presence of recurrent tumor was confirmed by surgery or 6-month radiological follow-up according to Response Assessment in Neuro-Oncology (RANO) criteria.12,13 From April 2014 to November 2017, 2HG MRS was performed in 96 subjects including a discovery cohort consisting of 42 patients and 6 normal volunteers; a preoperative validation cohort of 32 patients; and a recurrent-lesion validation cohort of 16 patients. Determination of IDH-Mutation Status IDH mutations were determined using immunohistochemistry,14,15 mass spectrometry–based mutation genotyping (OncoMap, Sequenom),15,16 or multiplex exome sequencing (OncoPanel, Illumina),17,18 depending on which genotyping technologies were available at the time of diagnosis. All sequencing assays were performed by the Molecular Diagnostics Division of the Brigham and Women’s Hospital Center for Advanced Molecular Diagnostics, a laboratory environment certified by the Clinical Laboratory Improvement Amendments, without knowledge of the results of the MRS. MRI and MRS Protocol All MRI and MRS exams were performed on one clinical 3.0T MRI scanner (Siemens TIM Skyra) with a 32-channel head coil. Prior to spectroscopy, sagittal 3D fluid attenuated inversion recovery images (FLAIR, repetition time [TR]/inversion time [TI]/TE = 9000 ms/2500 ms/81 ms, field of view = 20 × 20 cm, matrix = 224 × 320) were acquired and reconstructed in the axial and coronal planes with 2 mm slice resolution for accurate localization of the voxel. The MRS protocol included a long-TE (97 ms) single-voxel PRESS sequence and a long-TE (97 ms) semi-LASER (localization by adiabatic selective refocusing) CSI sequence. For the 97 ms TE single-voxel PRESS sequence, the acquisition parameters were: volume of interest = 20 × 20 × 20 mm3, TR/TE = 2 s/97 ms, 128 averages, 833 ms dwell time, 1024 points, and total time = 4:26 minutes. The region-of-interest (ROI) was chosen under the direction of a neuroradiologist (R.Y.H.) to include as much of the lesion as possible while avoiding the surrounding tissue. The voxel for normal volunteers was positioned in the centrum semiovale. Localized shimming was performed by adjustment of first- and second-order shim gradients using the automatic 3-dimensional B0 field mapping technique (Siemens) followed by manual adjustment of the above-mentioned shim gradients to achieve a magnitude peak width of water at half-maximum of 14 Hz or less. Manual shimming was utilized in our study to ensure consistent data quality among all patients so that shimming would not be a variable. After frequency adjustment, water-selective suppression was optimized by using the water suppression enhanced through T1 effects (WET) technique. For the 97 ms TE semi-LASER CSI sequence, the acquisition parameters included: field of view = 160 × 160 × 15 mm3, matrix = 16 × 16 for a voxel resolution = 10 × 10 × 15 mm3, TR/TE = 1700 ms/97 ms, 3 averages. Acceleration was enabled using a weighted distribution, resulting in a total time of 6:53 minutes. The ROI was positioned at the same level of the SVS voxel and was adjusted to cover as much of the lesion as possible, as well as the bilateral normal tissue while avoiding the side of the skull or other areas of susceptibility. Four saturation bands were placed along the margin of the ROI with optimum orientation to minimize lipid contamination from subcutaneous fat. Localized shimming was performed in the same way as was done for the single-voxel PRESS sequence to achieve a magnitude peak width of water at half-maximum of 25 Hz or less. Water suppression was accomplished with the WET technique. MRS Data Processing LCModel v6.2 software was used for both the SVS and CSI spectral fitting, using simulated spectra of 20 metabolites including 2HG as a customized basis set.9 For the CSI, the ROI was reconstructed on a VD17B scanner (Siemens), and the voxels contained within the abnormal hyperintense area on FLAIR images were determined by a neuroradiologist (R.Y.H.). Those spectra were then selected for processing using LCModel. Ratios of 2HG to creatine (2HG/Cr) were obtained and included in the further analysis. Signal-to-noise (SNR) ratio and full width at half maximum (FWHM) were used to assess the quality of the data. Spectra with an SNR <5 or FWHM of Cr peak >0.143 ppm were excluded for both the SVS and the spectra of the selected CSI voxels due to poor quality. For CSI, the 75th percentile 2HG/Cr values of the selected voxels were then calculated for each subject. Postprocessing was done without knowledge of patient IDH status. Statistical Analysis We first performed phantom experiments to establish that 2HG could be differentiated from other metabolites and quantified accurately by both the SVS and CSI methods. This was done using a phantom at pH 7.0 with different metabolites composed of 2HG (4 mM), glutamate, glutamine, GABA, myoinositol, glycine, lactate, N-acetylaspartate, Cr (4 mM), and choline. The SVS and CSI sequences were repeated 6 times. The means, standard deviations, and coefficients of variance for both the 2HG/Cr ratios from SVS and the 75th percentile 2HG/Cr values of CSI were calculated. Reproducibility of both the SVS and the CSI measurements was then assessed by test-retest analysis, in which a series of patients were scanned twice with the voxels placed in the same location. The patients either underwent an initial scan, got out of the scanner for 5 minutes, and then were repositioned and rescanned10 or were scanned twice within the same session based on the time restraints of clinical practices at the time of scanning. In the latter case, the 3D FLAIR sequence and the MRS sequences were repeated by another experienced technologist using the procedure described above. Scatter plot and correlation coefficient between test and retest measurements were presented. Using receiver operating characteristic (ROC) analysis, the sensitivity and specificity of 2HG/Cr ratios for diagnosing IDH-mutant gliomas in the discovery cohort were calculated for SVS and 75th percentile values of CSI. The optimal cutoff values for SVS and 75th percentile values of CSI were respectively chosen based on the analysis of the discovery cohort to optimize specificity with sensitivity above 50%. We then applied the cutoff values to both validation groups to examine the diagnostic performance of 2HG/Cr ratios for IDH-mutant gliomas and IDH-mutant glioma recurrence. The 75th percentile 2HG/Cr values of CSI were prepared with Microsoft Excel 2011, and all other analyses were performed using SPSS v21 (IBM). In addition, we performed post-hoc sample size calculations to check if the sample sizes that were used in the current studies were enough to detect the difference of diagnostic performance between the SVS method and the CSI method with type I error rate of 0.05 and power of 0.7. We applied the sample size calculation method proposed by Beam et al.19 θ1 and θ2 denote the diagnostic performance measures (eg, sensitivity, specificity, or diagnostic accuracy) for SVS and CSI, respectively. We would like to test H0: θ1−θ2=0, versus Ha: θ1−θ2=Δ . We assumed each group had equal sample size n. The formula to calculate n is shown below: n=[zα/2V0+zβVa]2Δ2, where V0=Var(θ̂1−θ̂2|H0)=φa+Δ2, Va=Var(θ̂1−θ̂2|Ha)=φa. Statistical software could provide estimates of variances for each method Var(θ̂1) and Var(θ̂1), but not the estimate of the variance of the difference Var(θ̂1−θ̂2). Note that θ̂1 and θ̂2 were dependent, since they were calculated based on the same set of subjects using different methods. To estimate V0=Var(θ̂1−θ̂2|H0), we performed a permutation analysis. Specifically, we generated 10000 permuted datasets. In each permuted dataset, we permuted the values of true positives and true negatives and kept 2HG/Cr value unchanged. We then detected if a subject was a positive or negative based on the cutoff values generated above. Next, we recalculated sensitivity, specificity, and accuracy. These values were obtained under the null hypothesis H0: θ1−θ2=0. We could calculate 10000 different θ̂1−θ̂2 based on the 10000 permuted datasets. Hence, we could estimate the variance of θ̂1−θ̂2 under H0: V0=Var(θ̂1−θ̂2|H0). To estimate Va=Var(θ̂1−θ̂2|Ha), we performed a bootstrapping analysis. Specifically, we generated 10000 bootstrapping datasets. In each bootstrapping dataset, we resampled subjects with replacement so that some subjects would be sampled multiple times and some subjects would not be sampled to a bootstrapping sample. We then detected if a subject was a positive or negative based on the cutoff values. Next, we recalculated sensitivity, specificity, and accuracy. These values were obtained under the alternative hypothesis Ha: θ1−θ2=0. We could calculate 10000 different θ̂1−θ̂2 based on the 10000 bootstrapping datasets. Hence, we could estimate the variance of θ̂1−θ̂2 under Va=Var(θ̂1−θ̂2|Ha). All power analyses were performed using R v3.3.2. Results Patient Cohorts Seven patients in the preoperative cohort were excluded, since no histological diagnosis of the IDH status was made at the time of this study, 3 patients in the discovery cohort were excluded due to small tumor volume, and 1 patient in the preoperative cohort was excluded due to poor spectrum quality. A total of 85 subjects were included in this study, including (i) a discovery cohort of 39 glioma patients (35 IDH-mutant glioma patients and 4 IDH-wildtype patients) and 6 normal volunteers (n = 31 for SVS, and n = 41 for CSI), (ii) a prospective preoperative validation cohort of 24 patients (n = 13 for SVS, and n = 22 for CSI), and (iii) a prospective recurrent-lesion validation cohort of 16 patients (n = 13 for SVS, and n = 15 for CSI). Of note, the patient numbers were different for SVS and CSI, either because only SVS or CSI was obtained for some patients or because SVS or CSI was excluded due to not meeting the quality control criteria. The patient clinical and pathologic characteristics are summarized for each cohort in Table 1. The detailed clinical information for patients of the recurrent-lesion validation cohort are given in Supplementary Table S1. Representative spectra of an IDH-mutant and an IDH-wildtype glioma patient are shown in Figs. 1 and 2. Table 1. Patient clinical and pathologic characteristics Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 View Large Table 1. Patient clinical and pathologic characteristics Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 Characteristics SVS CSI Discovery Group Validation Group 1 Validation Group 2 Discovery Group Validation Group 1 Validation Group 2 No. of patients 31 17 13 41 22 15 Median (range) age at the time of scan, y 40 (22‒66) 39 (20‒83) 31 (23‒67) 40 (24‒66) 39 (20‒83) 31 (23‒67) Sex  Male 15 10 2 22 13 2  Female 16 7 11 19 9 13 Histological diagnosis  Glioma grade   I 0 1 0 0 1 0   II 15 7 7 18 10 8   III 7 1 6 13 3 7   IV 3 5 0 4 6 0  Infarct 0 2 0 0 2 0  Lymphoma 0 1 0 0 1 0  Normal subjects 6 0 0 6 0 0 IDH mutation status  IDH mutation 22 7 13 31 9 15  IDH wildtype 9 10 0 10 13 0 View Large Fig. 1 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the discovery cohort. The white box in panel A shows the SVS location. A prominent 2HG peak was seen at 2.25 ppm in both SVS (2HG/Cr = 0.682, panel B) and the selected CSI spectrum (2HG/Cr = 0.852, panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectrum (D). The 75th percentile 2HG/Cr value of CSI was 0.719. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 1 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the discovery cohort. The white box in panel A shows the SVS location. A prominent 2HG peak was seen at 2.25 ppm in both SVS (2HG/Cr = 0.682, panel B) and the selected CSI spectrum (2HG/Cr = 0.852, panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectrum (D). The 75th percentile 2HG/Cr value of CSI was 0.719. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 2 View largeDownload slide Representative spectroscopies from an IDH-wildtype glioma patient in the preoperative cohort. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in either SVS (panel B) or the selected spectroscopy of CSI (panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectroscopy (D). The 75th percentile 2HG/Cr value of the CSI was 0. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 2 View largeDownload slide Representative spectroscopies from an IDH-wildtype glioma patient in the preoperative cohort. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in either SVS (panel B) or the selected spectroscopy of CSI (panel D). The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red box shows the selected voxel for the representative CSI spectroscopy (D). The 75th percentile 2HG/Cr value of the CSI was 0. In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Variability of 2HG Measurements For the phantom scans, the mean 2HG/Cr values were 0.968 and 0.952 for the SVS and CSI approaches, respectively, which were in close agreement with the actual ratio of 1. In addition, the standard deviations and the coefficients of variation were 0.036, 0.034 and 0.037, 0.036 for the SVS and CSI methods, respectively, which were readily attributed to machine noise. These data indicated that 2HG could be quantified accurately by both SVS and CSI in this study. Variability of 2HG measurements was assessed by test-retest analysis in 16 patients. Among them, the single-voxel PRESS method was repeated in 10 patients, whereas the CSI approach was repeated in 9 patients. For SVS, the 2HG/Cr ratio difference between scans 1 and 2 ranged between −0.05 and 0.05, with a test-retest correlation of 0.998. For CSI, the 75th percentile 2HG/Cr ratio difference between scans 1 and 2 varied from −0.11 to 0.11, with test-retest correlation of 0.969. These data establish the high reliability of 2HG/Cr quantitation by both the SVS and CSI methods over the tested range (0.00–1.13 for SVS, and 0.00–0.59 for the 75th percentile 2HG/Cr ratios of CSI). The scatter plots are presented in Supplementary Figure S1. Determination of the Optimal Cutoff Values in the Discovery Cohort Using ROC analysis, we investigated the diagnostic accuracy of 2HG/Cr ratios in determining IDH-mutation status in the discovery cohort. The areas under the curves were 0.83 (95% CI: 0.68–0.99) and 0.83 (0.67–0.98) for SVS and the 75th percentile values of CSI, respectively (Supplementary Figure S2). For SVS, 2HG/Cr threshold of 0.11 yielded the highest specificity at sensitivity above 50.00%. In terms of 75th percentile values of CSI, 2HG/Cr threshold of 0.23 generated the highest specificity at sensitivity above 50.00%. The sensitivity, specificity, and diagnostic accuracy corresponding to the potential cutoff values for both methods are listed in Table 2. Table 2 Potential 2HG/Cr cutoff values and the corresponding sensitivity, specificity, and accuracy in the discovery cohort‡ Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) ‡The optimal cutoff values are marked in bold. View Large Table 2 Potential 2HG/Cr cutoff values and the corresponding sensitivity, specificity, and accuracy in the discovery cohort‡ Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) Potential 2HG/Cr Cutoff Values Sensitivity, % Specificity, % Diagnostic Accuracy, % SVS  0.02 90.91 (20/22) 55.56 (5/9) 80.65 (25/31)  0.06 81.82 (18/22) 66.67 (6/9) 77.42 (24/31)  0.11 77.27 (17/22) 88.89 (8/9) 80.65 (25/31)  0.21 50.00 (11/22) 88.89 (8/9) 61.29 (19/31) 75th percentile values of CSI  0.01 93.55 (29/31) 60.00 (6/10) 85.37 (35/41)  0.11 77.42 (24/31) 70.00 (7/10) 75.61 (31/41)  0.23 58.06 (18/31) 80.00 (8/10) 63.41 (26/41)  0.28 51.61 (16/31) 80.00 (8/10) 58.54 (24/41) ‡The optimal cutoff values are marked in bold. View Large Diagnostic Performance of 2HG MRS in Preoperative and Recurrent-Lesion Validation Cohorts In the preoperative cohort, using the optimal 2HG/Cr thresholds resulted in a sensitivity, specificity, diagnostic accuracy, positive predictive value, and negative predictive value of 85.71%, 100.00%, 94.12%, 100.00%, and 90.91%, respectively, in distinguishing IDH-mutant gliomas and IDH-wildtype controls for SVS, and 100.00%, 69.23%, 80.00%, 69.23%, and 100.00%, respectively, for CSI (Table 3). Table 3 Sensitivity, specificity, and accuracy corresponding to the optimal 2HG/Cr cutoff values in the validation cohorts Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) View Large Table 3 Sensitivity, specificity, and accuracy corresponding to the optimal 2HG/Cr cutoff values in the validation cohorts Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) Diagnostic Performance Preoperative Validation Cohort Recurrent-Lesion Validation Cohort SVS CSI SVS CSI Sensitivity, % 85.71 (6/7) 100.00 (9/9) 40.00 (2/5) 66.67 (4/6) Specificity, % 100.00 (10/10) 69.23 (9/13) 62.50 (5/8) 100.00 (9/9) Diagnostic accuracy, % 94.12 (16/17) 81.82 (18/22) 53.85 (7/13) 86.67 (13/15) Positive predictive value, % 100.00 (6/6) 69.23 (9/13) 40.00 (2/5) 100.00 (4/4) Negative predictive value, % 90.91 (10/11) 100.00 (9/9) 62.50 (5/8) 81.82 (9/11) View Large When the optimal 2HG/Cr thresholds were applied to the recurrent-lesion cohort, the sensitivity, specificity, diagnostic accuracy, positive predictive value, and negative predictive value for discriminating IDH-mutant glioma recurrence were 40.00%, 62.50%, 53.85%, 40.00%, and 62.50%, respectively, for SVS; and 66.67%, 100.00%, 86.67%, 100.00%, and 81.82%, respectively, for CSI (Table 3). Representative results from a posttreatment IDH-mutant glioma patient with surgically confirmed recurrence are shown in Fig. 3. Existence and uneven distribution of 2HG were observed in CSI with the 75th percentile 2HG/Cr value of 0.68. However, no 2HG peak was seen in SVS. Fig. 3 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the recurrent-lesion cohort. The presence of recurrent tumor was confirmed by surgery. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in SVS (panel B). However, CSI demonstrated the existence and the heterogeneous distribution of 2HG (panels C, D, E). The 2HG/Cr ratios ranged from 0.00 (panel E) to 3.50 (panel D), with the 75th percentile value of 0.68. The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red boxes show the selected voxels for the representative CSI spectra (panels D, E). In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Fig. 3 View largeDownload slide Representative spectra from an IDH-mutant glioma patient in the recurrent-lesion cohort. The presence of recurrent tumor was confirmed by surgery. Panel A shows the SVS location. No 2HG peak was seen at 2.25 ppm in SVS (panel B). However, CSI demonstrated the existence and the heterogeneous distribution of 2HG (panels C, D, E). The 2HG/Cr ratios ranged from 0.00 (panel E) to 3.50 (panel D), with the 75th percentile value of 0.68. The blue box in panel C shows the selected voxels contained within the abnormal hyperintense area on FLAIR images, and the red boxes show the selected voxels for the representative CSI spectra (panels D, E). In panels B and D, the black spectrum shows the raw data, red spectrum is the fitted data, and the blue spectrum shows the 2HG fit. Power Analysis The required sample sizes given type I error rate of 0.05 and power 0.7 are shown in Supplementary Table S2. We had enough power to detect the difference of diagnostic performance between the SVS method and the CSI method in both settings. Discussion In this study we prospectively evaluated the diagnostic accuracy of 2HG MRS in identifying IDH-mutant gliomas in patients presenting with newly discovered brain mass as well as in patients with treated IDH-mutant tumors presenting with suspected tumor recurrence. Our results demonstrate that 2HG MRS using the SVS method is highly accurate in diagnosing IDH-mutant gliomas among newly diagnosed brain mass, whereas 2HG MRS using the CSI method provides good accuracy in identifying recurrent IDH-mutant glioma. Several prior studies have reported the 2HG absolute concentration thresholds to distinguish between IDH-mutant and IDH-wildtype gliomas with 97 ms TE single-voxel PRESS sequence, and the most commonly used threshold was the 2HG concentration ≥1 mM with or without the Cramér-Rao lower bound of 2HG ≤30%.10,11 In this study, we used 2HG/Cr ratio rather than absolute 2HG concentrations to diagnose IDH mutations. One reason is that the acquisition of the CSI water reference sequence takes at least 4:30 minutes with a single average. This increased scanning time is clinically challenging due to resource utilization constraints in busy neuroradiology practices, and increases the likelihood for motion artifacts during the CSI-CSI water reference session of about 12 minutes. Furthermore, 2HG absolute quantification by water reference also suffers from the variation of water concentration.8,9 The 2HG absolute quantification studies apply a constant water concentration of normal brain as reference in tumors, assuming that the water concentration is constant between normal brain and tumor tissues, and is invariable among all gliomas. However, prior studies have shown that the brain water content can be increased up to 50% in brain tumors over that in healthy subjects,20,21 and water concentration varies in individual glioma based on a range of factors including edema severity, tumor cellularity, and historical type.8 Although a relatively accurate water concentration estimation method has been proposed,22 the edema, cysts, and surgical cavities need to be accurately segmented, while Cr excludes these confoundings.23 Moreover, endogenous Cr signal reference for 2HG quantification has been used in a recently published paper.23 For better comparison, ratios to Cr were chosen for both the SVS and CSI spectral quantification, which obviated the need of acquiring a water reference sequence and limited our protocol to under 6 minutes for both SVS and CSI acquisitions. The ROC analysis in the discovery cohort yielded a 2HG/Cr cutoff value of 0.11 for SVS, and 0.23 for the 75th percentile 2HG/Cr values of CSI. Assuming a total Cr level between 8 and 10 mmol/L,23 the SVS 2HG/Cr threshold of 0.11 would correspond to a 2HG level around 0.88–1 mmol/L, which is comparable but a little lower than the absolute 2HG concentration threshold reported previously. It is also important to note why there are different thresholds for SVS and CSI 2HG measures: The optimal threshold value for the SVS technique was selected based on the ratio of 2HG to Cr concentrations over our training study population. The optimal threshold value for the CSI technique was calculated based on the mean value of the population of voxels in the top 75th percentile 2HG/Cr concentration, an area that is more representative of likely tumor. By using such a top quartile approach, the sensitivity of detecting a small area of true tumor is expected to increase compared with averaging of the entire voxel area. In other words, if only true tumor voxels were selected and analyzed during CSI analysis, the threshold value would be identical to SVS (such as shown in our phantom results), but this is exactly the main difficulty with posttreatment imaging where tumor voxels are not readily differentiable from nontumor tissues. In the preoperative cohort with newly diagnosed brain mass, we demonstrated that the SVS method achieved an accuracy of 94.12% in diagnosing IDH-mutant lesions preoperatively. This result was concordant with those previously published by Choi et al in newly diagnosed brain mass using 97 ms TE SVS.10 We also found that the SVS approach resulted in higher specificity and accuracy compared with the CSI method. The better diagnostic performance of the SVS could be due to the larger voxel size during the SVS acquisition compared with that during the CSI acquisition, resulting in higher SNRs, which are of benefit in distinguishing 2HG peak from the adjacent resonances of glutamate, glutamine, and GABA. In the cohort with known IDH-mutant tumors presenting with suspected tumor recurrence, the CSI method showed higher sensitivity, specificity, and accuracy in diagnosing IDH-mutant glioma recurrence compared with the SVS approach. Due to the larger spatial coverage and multivoxel acquisition, CSI has the advantage of accounting for the spatial heterogeneity of the abnormal tissues, including gliosis, edema, and tumor, which commonly coexist in the posttreatment setting.8,23 Additionally, because of the smaller volumes of the CSI voxels, the metabolic measures by CSI contain reduced partial-volume effects compared with SVS8 (Fig. 3). These potential advantages of the CSI method need validation by direct tissue sampling using imaging guided resection or biopsy. It is noted that the specificity of CSI was higher than that of SVS. Subgroup analysis was performed based on WHO grade and showed the same results in both the WHO grades II and III groups. Moreover, we explored other possible reasons for the SVS false positives, including spectral quality, tumor volume, and outcome measure, and found them all to be consistent. No specific reason was found to explain the relative lower specificity of SVS versus CSI, and this warrants further investigation. Our study is limited by small patient sample sizes in the validation groups, especially the recurrent-lesion group. While the accuracy of 2HG MRS has been demonstrated in our single-institution study, the diagnostic accuracy in both clinical settings requires further validation by larger prospective trials. Our analysis of the recurrent-tumor cohort is also limited by lack of histological evaluation of all patients but is dependent on follow-up clinical and imaging stability. In conclusion, 2HG MRS provides diagnostic utility for IDH-mutant gliomas both preoperatively and at time of suspected tumor recurrence. SVS has a better diagnostic performance for IDH-mutant gliomas in the untreated patient cohort, whereas CSI demonstrates greater performance in identifying IDH-mutant glioma recurrence. Supplementary Material Supplementary material is available at Neuro-Oncology online. Funding This work was supported by Brigham and Women’s Hospital Institute for the Neurosciences seed grant and an ARRS/ASNR Scholar Award. Conflict of interest statement. David A. Reardon serves as a paid advisory board member for Abbvie, Amgen, BMS, Cavion, Celldex, EMD Serono, Genentech/Roche, Inovio, Juno Pharmaceuticals, Merck, Midatech, Momenta Pharmaceuticals, Novartis, Novocure, Oncorus, Oxigene, Regeneron, and Stemline Terapeutics; receives lab research support (paid to Dana-Farber) from Aceta Pharma, Agenus, Celldex Therapeutics, EMD Serono, Incyte, Inovio, Midatech, and Tragara; is a paid speaker for Genentech/Roche and Merck. Patrick Y. Wen reports grants, personal fees, and nonfinancial support from Agios, nonfinancial support from Angiochem, GlaxoSmithKline, Sanofi-Aventis, and VBI Vaccines, personal fees and nonfinancial support from AstraZeneca, Genentech/Roche, and Vascular Biogenics, personal fees from Cavion, Insys, Monteris, Novogen, Regeneron, and Tocogen. Alexander P. Lin is a consultant for Agios Pharmaceuticals, cofounder of BrainSpec, and consultant to Moncton MRI. No other authors have a conflict of interest. Conflict of interest statement. David A. Reardon serves as a paid advisory board member for Abbvie, Amgen, BMS, Cavion, Celldex, EMD Serono, Genentech/Roche, Inovio, Juno Pharmaceuticals, Merck, Midatech, Momenta Pharmaceuticals, Novartis, Novocure, Oncorus, Oxigene, Regeneron, and Stemline Terapeutics; receives lab research support (paid to Dana-Farber) from Aceta Pharma, Agenus, Celldex Therapeutics, EMD Serono, Incyte, Inovio, Midatech, and Tragara; is a paid speaker for Genentech/Roche and Merck. Patrick Y. Wen reports grants, personal fees, and nonfinancial support from Agios, nonfinancial support from Angiochem, GlaxoSmithKline, Sanofi-Aventis, and VBI Vaccines, personal fees and nonfinancial support from AstraZeneca, Genentech/Roche, and Vascular Biogenics, personal fees from Cavion, Insys, Monteris, Novogen, Regeneron, and Tocogen. Alexander P. Lin is a consultant for Agios Pharmaceuticals, cofounder of BrainSpec, and consultant to Moncton MRI. No other authors have a conflict of interest. Acknowledgments The authors would like to thank the China Scholarship Council for financial support for Min Zhou and Yue Zhou. The authors would also like to thank Dr Changho Choi for the use of his basis set and guidance as well as Dr Weiliang Qiu for the sample size calculations. References 1. Eckel-Passow JE , Lachance DH , Molinaro AM , et al. Glioma groups based on 1p/19q, IDH, and TERT promoter mutations in tumors . N Engl J Med . 2015 ; 372 ( 26 ): 2499 – 2508 . Google Scholar CrossRef Search ADS PubMed 2. Brat DJ , Verhaak RG , Aldape KD , et al. ; Cancer Genome Atlas Research Network . Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas . N Engl J Med . 2015 ; 372 ( 26 ): 2481 – 2498 . Google Scholar CrossRef Search ADS PubMed 3. Yan H , Parsons DW , Jin G , et al. IDH1 and IDH2 mutations in gliomas . N Engl J Med . 2009 ; 360 ( 8 ): 765 – 773 . Google Scholar CrossRef Search ADS PubMed 4. Parsons DW , Jones S , Zhang X , et al. An integrated genomic analysis of human glioblastoma multiforme . Science . 2008 ; 321 ( 5897 ): 1807 – 1812 . Google Scholar CrossRef Search ADS PubMed 5. Dang L , White DW , Gross S , et al. Cancer-associated IDH1 mutations produce 2-hydroxyglutarate . Nature . 2009 ; 462 ( 7274 ): 739 – 744 . Google Scholar CrossRef Search ADS PubMed 6. Jin G , Reitman ZJ , Duncan CG , et al. Disruption of wild-type IDH1 suppresses D-2-hydroxyglutarate production in IDH1-mutated gliomas . Cancer Res . 2013 ; 73 ( 2 ): 496 – 501 . Google Scholar CrossRef Search ADS PubMed 7. Pope WB , Prins RM , Albert Thomas M , et al. Non-invasive detection of 2-hydroxyglutarate and other metabolites in IDH1 mutant glioma patients using magnetic resonance spectroscopy . J Neurooncol . 2012 ; 107 ( 1 ): 197 – 205 . Google Scholar CrossRef Search ADS PubMed 8. Choi C , Ganji SK , DeBerardinis RJ , et al. 2-hydroxyglutarate detection by magnetic resonance spectroscopy in IDH-mutated patients with gliomas . Nat Med . 2012 ; 18 ( 4 ): 624 – 629 . Google Scholar CrossRef Search ADS PubMed 9. Choi C , Ganji S , Hulsey K , et al. A comparative study of short- and long-TE ¹H MRS at 3 T for in vivo detection of 2-hydroxyglutarate in brain tumors . NMR Biomed . 2013 ; 26 ( 10 ): 1242 – 1250 . Google Scholar CrossRef Search ADS PubMed 10. Choi C , Raisanen JM , Ganji SK , et al. Prospective longitudinal analysis of 2-hydroxyglutarate magnetic resonance Spectroscopy identifies broad clinical utility for the management of patients with IDH-mutant glioma . J Clin Oncol . 2016 ; 34 ( 33 ): 4030 – 4039 . Google Scholar CrossRef Search ADS PubMed 11. de la Fuente MI , Young RJ , Rubel J , et al. Integration of 2-hydroxyglutarate-proton magnetic resonance spectroscopy into clinical practice for disease monitoring in isocitrate dehydrogenase-mutant glioma . Neuro-Oncol . 2016 ; 18 ( 2 ): 283 – 290 . Google Scholar CrossRef Search ADS PubMed 12. Wen PY , Macdonald DR , Reardon DA , et al. Updated response assessment criteria for high-grade gliomas: Response Assessment in Neuro-Oncology working group . J Clin Oncol . 2010 ; 28 ( 11 ): 1963 – 1972 . Google Scholar CrossRef Search ADS PubMed 13. van den Bent MJ , Wefel JS , Schiff D , et al. Response Assessment in Neuro-Oncology (a report of the RANO group): assessment of outcome in trials of diffuse low-grade gliomas . Lancet Oncol . 2011 ; 12 ( 6 ): 583 – 593 . Google Scholar CrossRef Search ADS PubMed 14. Capper D , Zentgraf H , Balss J , Hartmann C , von Deimling A . Monoclonal antibody specific for IDH1 R132H mutation . Acta Neuropathol . 2009 ; 118 ( 5 ): 599 – 601 . Google Scholar CrossRef Search ADS PubMed 15. Ramkissoon SH , Bi WL , Schumacher SE , et al. Clinical implementation of integrated whole-genome copy number and mutation profiling for glioblastoma . Neuro Oncol . 2015 ; 17 ( 10 ): 1344 – 1355 . Google Scholar CrossRef Search ADS PubMed 16. Thomas RK , Baker AC , Debiasi RM , et al. High-throughput oncogene mutation profiling in human cancer . Nat Genet . 2007 ; 39 ( 3 ): 347 – 351 . Google Scholar CrossRef Search ADS PubMed 17. Cryan JB , Haidar S , Ramkissoon LA , et al. Clinical multiplexed exome sequencing distinguishes adult oligodendroglial neoplasms from astrocytic and mixed lineage gliomas . Oncotarget . 2014 ; 5 ( 18 ): 8083 – 8092 . Google Scholar CrossRef Search ADS PubMed 18. Wagle N , Berger MF , Davis MJ , et al. High-throughput detection of actionable genomic alterations in clinical tumor samples by targeted, massively parallel sequencing . Cancer Discov . 2012 ; 2 ( 1 ): 82 – 93 . Google Scholar CrossRef Search ADS PubMed 19. Beam CA , Baker ME , Paine SS , Sostman HD , Sullivan DC . Answering unanswered questions: proposal for a shared resource in clinical diagnostic radiology research . Radiology . 1992 ; 183 ( 3 ): 619 – 620 . Google Scholar CrossRef Search ADS PubMed 20. Grasso G , Alafaci C , Passalacqua M , et al. Assessment of human brain water content by cerebral bioelectrical impedance analysis: a new technique and its application to cerebral pathological conditions . Neurosurgery . 2002 ; 50 ( 5 ): 1064 – 1072 ; discussion 1072–1074. Google Scholar PubMed 21. Jansen JF , Backes WH , Nicolay K , Kooi ME . 1H MR spectroscopy of the brain: absolute quantification of metabolites . Radiology . 2006 ; 240 ( 2 ): 318 – 332 . Google Scholar CrossRef Search ADS PubMed 22. Gasparovic C , Song T , Devier D , et al. Use of tissue water as a concentration reference for proton spectroscopic imaging . Magn Reson Med . 2006 ; 55 ( 6 ): 1219 – 1226 . Google Scholar CrossRef Search ADS PubMed 23. Andronesi OC , Loebel F , Bogner W , et al. Treatment response assessment in idh-mutant glioma patients by noninvasive 3D functional spectroscopic mapping of 2-hydroxyglutarate . Clin Cancer Res . 2016 ; 22 ( 7 ): 1632 – 1641 . Google Scholar CrossRef Search ADS PubMed © The Author(s) 2018. Published by Oxford University Press on behalf of the Society for Neuro-Oncology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

Neuro-OncologyOxford University Press

Published: Sep 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off