A Novel Methodology for Applying Multivoxel MR Spectroscopy to Evaluate Convection-Enhanced Drug Delivery in Diffuse Intrinsic Pontine Gliomas

BACKGROUND AND PURPOSE: Diffuse intrinsic pontine gliomas are inoperable high-grade gliomas with a median survival of less than 1 year. Convection-enhanced delivery is a promising local drug-delivery technique that can bypass the BBB in diffuse intrinsic pontine glioma treatment. Evaluating tumor response is critical in the assessment of convection-enhanced delivery of treatment. We proposed to determine the potential of 3D multivoxel 1H-MR spectroscopy to evaluate convection-enhanced delivery treatment effect in these tumors. MATERIALS AND METHODS: We prospectively analyzed 3D multivoxel 1H-MR spectroscopy data for 6 patients with nonprogressive diffuse intrinsic pontine gliomas who received convection-enhanced delivery treatment of a therapeutic antibody (Phase I clinical trial NCT01502917). To compare changes in the metabolite ratios with time, we tracked the metabolite ratios Cho/Cr and Cho/NAA at several ROIs: normal white matter, tumor within the convection-enhanced delivery infusion site, tumor outside of the infused area, and the tumor average. RESULTS: There was a comparative decrease in both Cho/Cr and Cho/NAA metabolite ratios at the tumor convection-enhanced delivery site versus tumor outside the infused area. We used MR spectroscopy voxels with dominant white matter as a reference. The difference between changes in metabolite ratios became more prominent with increasing time after convection-enhanced delivery treatment. CONCLUSIONS: The comparative change in metabolite ratios between the convection-enhanced delivery site and the tumor site outside the infused area suggests that multivoxel 1H-MR spectroscopy, in combination with other imaging modalities, may provide a clinical tool to accurately evaluate local tumor response after convection-enhanced delivery treatment.

D iffuse intrinsic pontine gliomas (DIPGs) are inoperable high-grade gliomas that account for 10%-15% of pediatric primary CNS tumors and 75%-80% of pediatric brain stem tumors. 1,2 The current standard of care for DIPG is radiation therapy; despite much effort, median survival is Ͻ1 year. Effective drug delivery to these brain tumors is an obstacle. 3 The relatively intact blood-brain barrier presents a major hurdle in drug delivery to DIPGs. Convection-enhanced delivery (CED) is a local drug-delivery technique that bypasses the BBB and uses hydraulic pressure to deliver agents directly to the brain tumor. CED allows enhanced and uniform drug distribution, optimized therapeutic indices, and greatly reduced systemic toxicity to patients. [4][5][6] There is currently no standard method for evaluating treatment response at the CED infusion site that is different from evaluating the whole-tumor treatment response. In this study, we used MR spectroscopic imaging (MRSI), a noninvasive imaging technique that provides information on the tissue metabolites. 7 Due to the unique metabolic characteristics of tumors, MRSI has been used to aid in diagnosing various cancers and evaluating intratumoral heterogeneity and treatment response in both ani-mals and humans. 8 MRSI allows the analysis of many different metabolites in tumor and normal tissue regions of the brain. Some of the major metabolites of interest in examining brain tissue are choline compounds, creatine, and N-acetylaspartate. 7,8 Cho represents the sum of choline compounds, which are components of cell membrane and increase with cellular proliferation. Cr is a biomarker of energy metabolism and is generally stable, functioning primarily as a standard comparison of the other metabolites, though it can vary within certain tumors. 9 NAA is used as a biomarker for neuronal and axonal integrity and will decrease with neuronal damage or loss. The metabolite ratios of Cho/Cr and Cho/NAA were the parameters of interest in this study.
In the current descriptive study, we used multivoxel proton 1 H-MR spectroscopy to compare metabolic changes at the CED infusion site with tumor outside the CED-infused area. We used voxels filled with dominant normal white matter as a reference. We hypothesized that multivoxel 1 H-MR spectroscopy would show a difference in the change of metabolites at the CED-infused area compared with the non-CED infused area.

Participants
We prospectively analyzed MR spectroscopy data from 6 patients diagnosed with DIPG who received CED of a therapeutic antibody in a Phase I clinical trial (NCT01502917), which was approved by the Memorial Sloan Kettering Cancer Center review board. The median age of patients was 7 years (range, 3-17 years). Patients were diagnosed with DIPG on the basis of clinical presentation and MR imaging findings. Before CED treatment, they all completed standard external beam radiation therapy (54 -60 Gy) for at least 9 weeks, but for no more than 14 weeks. MR spectroscopic imaging was performed at Ն3 time points for each patient: 1 week before treatment (preCED), 1 day after treatment (postCED1), and 1 month after treatment (postCED2). All patients were required to have imaging at 1 month after CED (post-CED2). Some patients had additional scans at 1-month intervals afterward (postCED3 and 3 months after CED treatment) if they were able to return more frequently due to their close proximity or willingness to travel for additional scans.

MRI/MRSI Methods
The MR imaging/MRSI studies were performed by using a wholebody 3T Discovery MR750 scanner (GE Healthcare, Milwaukee, Wisconsin) with a quadrature head coil as the transceiver. The MR imaging protocol included precontrast sagittal and axial T1weighted, axial T2-weighted, axial T2 FLAIR, axial diffusionweighted, and postcontrast T1-weighted sequences in axial, sagittal, and coronal planes. The precontrast T2 FLAIR images were obtained by using a fast spin-echo sequence (TI/TE/TR ϭ 2250/ 120/9000 ms, FOV ϭ 24 cm). The 3-mm section thickness with no spacing between sections (320 ϫ 256 acquisition matrix size) served as scouts for placement of the volume of interest for the 3D MRSI data acquisition. MRSI data acquisition was performed following the precontrast MR imaging acquisition, but before gadolinium contrast administration. The 3D MRSI VOI encompassed the lesion (hyperintense area on T2 FLAIR images) and surround-ing normal-appearing brain tissue. Outer volume saturation bands were applied to avoid signal contaminations caused by subcutaneous lipid, bone, and varying magnetic susceptibility effects that might compromise the quality of the spectra. A point-resolved spectroscopy sequence with water suppression by using chemical shift selective suppression was used to collect the 3D 1 H-MRSI datasets with TE ϭ 144 ms, TR ϭ 1000 ms, FOV ϭ 8 cm, section thickness ϭ 10 mm, 3D spatial-encoding (8 ϫ 8 ϫ 8 matrix size, resulting in 1-cm 3 nominal MRSI voxel size), 5000-Hz spectral width, and 4096 data points. The acquisition time for 3D MRSI was approximately 9 minutes, including the prescan for magnetic field shimming. The MRSI VOI location, size, and acquisition parameters were kept the same for each patient during follow-up studies by using anatomic landmarks. The axial postcontrast T1WI MR images were collected with the same section number, location, and thickness as the axial T2 FLAIR images.

MR Spectroscopy Data Processing
Multivoxel 1 H-MR spectroscopy data were analyzed with Func-Tool software (Advantage Workstation, Version 11.3-4.2; GE Healthcare). The MRSI sections were interpolated to 3 mm to match the FLAIR images and generate overlay images of the MRSI grid onto the corresponding T2-FLAIR image. ROIs were equivalent to 1 MR spectroscopy voxel (nominal size of 1 cubic centimeter) and were selected by using the overlaid T2-FLAIR images for anatomic reference. Major resonance peaks of NAA (2.0 ppm), Cr (3.0 ppm), and Cho (3.2 ppm) for each MRSI voxel were assigned and numerically integrated to estimate peak areas. Peak area ratios were calculated and color maps were generated for Cho/NAA and Cho/Cr.

ROI Selection
Several ROIs were chosen to analyze MR spectroscopic data: a voxel at the CED infusion, a voxel that was within the tumor but outside the CED infusion, additional voxels that were within the tumor to calculate the average tumor metabolites, and normal tissue voxels as controls. A radiologist with a Certificate of Added Qualification in radiology (S.H., 11 years of experience) and a physicist (S.B.T., 12 years of experience in brain MR spectroscopy) oversaw all ROI voxel selections. The CED infusion voxel was selected by using the postCED1 scans because the infusion site was visible as hyperintense signal on the T2-FLAIR images ( Fig  1A-D). The CED infusion site voxels for preCED, postCED2, and, when applicable, postCED3 were selected by matching the section numbers and location by using anatomic landmarks to the post-CED1 infusion site voxel. The tumor non-CED infusion site voxel was chosen at least 2 sections away from the CED infusion site and was verified by the radiologist (S.H.) to be outside of the CED infused volume of distribution (Fig 1E-H).
Color metabolite maps were used to observe the relative intensity of each metabolite and metabolite ratio in and around the CED infusion site. Anatomic landmarks and section numbers were used to best match equivalent tumor non-CED infusion site voxels for the remaining scans. Additional tumor voxels were selected on the basis of MR imaging and spectra that were characteristic of tumor (high Cho and low NAA peak compared with unaffected tissue and metabolite ratios: Cho/Cr of Ͼ1.5 and Cho/ NAA of Ͼ1). An average of the tumor metabolite ratios was calculated for each scan by taking an average of the Cho/Cr and Cho/NAA ratios from the following voxels of tumor: the CED infusion site, the tumor non-CED infusion site, and the other additional tumor voxels. Normal tissue voxels were chosen in the white matter cerebellum or cerebrum away from the tumor and had spectra characteristic of normal white matter with metabolite ratios of Cho/Cr of ϳ0.7-1 and Cho/NAA of ϳ0.5-1. 10-12

Statistical Analysis
Statistical analysis was performed on metabolite ratios, which were treated as data with unknown distributions. Exploratory data analysis included group means/medians with time and graphing individual measurements with time. To test the trend of changes in metabolite ratios, we performed paired difference tests by using the Wilcoxon signed rank test in R (http://www. r-project.org).

RESULTS
The metabolite ratios Cho/Cr and Cho/NAA for the CED infusion site, tumor non-CED infusion site, average tumor, and normaltissue ROIs were compared during the course of treatment. Color metabolite maps demonstrated the higher intensity of Cho, Cho/ Cr, and Cho/NAA and lower intensity of NAA in the area of the pons with tumor (Fig 2).
The metabolite ratios for normal tissue remained stable and within the normal range for each patient throughout the course of treatment (Fig 3). In 4 patients, Cho/Cr and Cho/ NAA decreased at the CED infusion site with time, while the ratios for the tumor non-CED infusion site generally decreased by a lesser degree or increased with time (Fig 4A-D). In 2 patients, Cho/Cr and Cho/NAA increased at the infusion site with time, but a more substantial increase occurred in Cho/Cr and Cho/NAA at the tumor non-CED infusion site and for the average tumor (Fig 4E, -F). In 2 patients, Cho/Cr and Cho/ NAA in the CED infusion site ROI at postCED1 were high ( Fig  4B, -E).
Statistical analysis was performed to assess the change in Cho/Cr and Cho/NAA at the CED infusion site compared with the tumor non-CED infusion site in all patients (Fig 5). There was a relative difference in the percentage change in both Cho/Cr and Cho/NAA for the CED infusion site compared with the tumor non-CED infusion site. The mean and P value decreased for both Cho/Cr and Cho/NAA metabolite ratios with time when comparing preCED with postCED1, preCED with postCED2, and preCED with postCED3. The mean (SD) for the change in Cho/Cr between infusion and non-CED infusion sites was similar between postCED1 and postCED2 compared with preCED: 1.

DISCUSSION
Despite much therapeutic effort, patients with DIPG continue to have a poor prognosis with a median survival of Ͻ1 year. Radia-tion therapy is the standard of care because most chemotherapeutic agents do not prolong survival. CED is a promising technique that allows direct localized delivery of therapeutic agents to the tumor. In this study, patients with nonprogressive DIPG received CED treatment with a therapeutic antibody in a Phase I clinical trial (NCT01502917). Evaluating local treatment response at the CED infusion site is critical because it may suggest that there is some treatment effect when whole-tumor response may be undetectable. A CED therapeutic agent may be falsely deemed ineffective if whole-tumor response is assessed without evaluating the response at the local infusion site.
Evaluating DIPG treatment response with conventional MR imaging alone is inadequate due to nonspecific changes on imaging that can arise from recurrent tumor, inflammatory responses, and necrotic changes from treatment. 12,13 Furthermore, conventional MR imaging results alone have not been shown to correlate with survival in patients with DIPG. 14 Diffusion tensor imaging and diffusion-weighted imaging have been studied in patients with DIPG, but the diffuse infiltration of DIPG has made using these techniques difficult. The pattern and amount of tract involvement varies among patients, and the significance of diffusion properties in DIPG varies anatomically. 15,16 Perfusion-weighted MR imaging has been used to assess the angiogenesis of the tumor, but relative tumor blood volume has a weaker correlation to disease progression than increased choline levels with 1 H-MR spectroscopy. 17  1 H-MR spectroscopy is effective in providing clinically important information in brain tumors and has increased diagnostic sensitivity, specificity, and accuracy when used with MR imaging. 8 1 H-MR spectroscopy uses the radiofrequency signal from protons when a magnetic field is applied to determine metabolite concentrations by displaying a spectrum of peaks corresponding to different metabolites. The area under the curve of the peak is then used to calculate the concentration and ratios of the metabolites. 7 Due to the abnormal metabolism characteristic of tumor cells, MR spectroscopy can be used to identify tumorous tissue with good accuracy. Cho, Cr, and NAA are some of the most common metabolites analyzed for brain tumors. Increases in Cho correspond to increased membrane turnover or cellular density and thus make Cho a biomarker for tumor tissue. 7 Cr is usually stable because it is a biomarker of energy metabolism, but it can vary within certain tumors, and a decrease in Cr can signify tissue necrosis. 7,9 Decreasing levels of NAA signify a decrease in neuronal integrity. 7 Cho/Cr and Cho/NAA ratios have often been used to identify the metabolic changes in tumor tissue. 11,12 Increases in both Cho/Cr and Cho/NAA indicate tumor because Cho increases, Cr generally stays stable or decreases, and NAA decreases in tumor tissue.
Single-voxel 1 H-MR spectroscopy has been used to assess prognosis and overall treatment response to radiation therapy and chemotherapeutics in DIPG tumors. 8,18,19 However, multivoxel 1 H-MR spectroscopy allows the interrogation of local changes in metabolism within different regions of the tumor. 13 A previous study used multivoxel 1 H-MR spectroscopy to look at DIPG during radiation therapy and compared changes in tumor versus normal brain tissue. This study showed a correlation with low Cho/Cr and Cho/NAA during response to radiation therapy and high Cho/Cr and Cho/NAA with relapse. 12 Other studies have also shown increases in Cho/Cr and Cho/NAA ratios in DIPG tumors corresponding to prognosis and relapse after treatment. 20,21 In our study, we wanted to evaluate the utility of using multivoxel 1 H-MR spectroscopy for assessing CED treatment. We prospectively analyzed the multivoxel 1 H-MR spectroscopy data from 6 patients with DIPG receiving CED treatment. We compared the Cho/Cr and Cho/NAA ratios across time at the CED infusion site with the tumor non-CED infusion site outside the CED infused volume of distribution.
Based on the MR spectroscopy data for these patients, there is a comparative reduction in Cho/Cr and Cho/NAA after CED treatment at the CED infusion site compared with the tumor non-CED infusion site. This reduction becomes more pronounced with increasing time after CED treatment. These MR spectroscopic changes, when combined with findings from other modalities, such as DWI, DTI, PWI, and PET and compared with patient outcomes, may reflect CED treatment response. In 2 patients, metabolite ratios for the CED infusion site ROI at postCED1 were high compared with their baseline metabolite ratios at preCED. Patient 2 had Cho/Cr ϭ 4.22 for the CED infusion site and 3.88 for the tumor noninfusion site, respectively, at post-CED1 (Fig 4B). Patient 5 had Cho/NAA ϭ 2.78 and Cho/Cr ϭ 2.17 for the tumor noninfusion site at postCED1 (Fig 4E). Because postCED1 is 1 day after CED treatment, there may be an influence of the infused CED volume on the MR spectroscopic readings. The infusate may produce susceptibility arti- facts that could result in poor shimming and hence broader line widths. This result could lead to inaccurate quantification of metabolites at postCED1. Despite this limitation, we were able to see a difference in metabolite ratios at the infusion site versus the noninfused tumor by using our methodology with multivoxel 1 H-MR spectroscopy. Our methodology for ROI selection to assess CED treatment response can also be used in parallel to analyze results from other imaging modalities such as DWI and PWI, which may provide additional biomarkers to help evaluate the response in CED treatment.
The study is limited in that a reduced metabolite ratio trend without statistical significance was observed on the basis of the small number of patients analyzed with only a few scans. Additionally, the patients started with different tumor volumes and with different metabolite ratios from their baseline preCED scan. These variables may influence the response to CED treatment. It may be of interest for future studies to derive absolute measures of metabolites that could provide an accurate way to study the local changes with treatment and to standardize the patient's tumor size and baseline metabolic ratios.

CONCLUSIONS
Our methodology for analysis of multivoxel 1 H-MR spectroscopy highlighted a trend of decreasing P values for the metabolite ratios Cho/Cr and Cho/NAA after CED treatment at the CED infusion site compared with the tumor non-CED infusion site. Although statistical tests were not significant with the small number of patients, the results are encouraging, and if confirmed, MRSI could be used to obtain early markers of tumor response in individual patients. Our methodology may also be applicable to other imaging modalities, with some optimization for determining ROI size and placement. When combined with other imaging techniques, such as PET, DTI, DWI, and PWI, our methodology may provide a clinical tool to more accurately evaluate local tumor response after CED treatment.