Saumya S. Gurbani1,2, Eric Mellon3, Brent D. Weinberg2, Eduard Schreibmann1, Andrew A. Maudlsey4, Sulaiman Sheriff4, Peter B. Barker5, Lawrence Kleinberg6, Lee A. D. Cooper7, Hui-Kuo Shu1, and Hyunsuk Shim1,2
1Department of Radiation Oncology, Winship Cancer Institute of Emory University, Atlanta, GA, United States, 2Department of Radiology and Imaging Sciences, Emory University School of Medicine, Atlanta, GA, United States, 3Department of Radiation Oncology, University of Miami Miller School of Medicine, Miami, FL, United States, 4Department of Radiology, University of Miami Miller School of Medicine, Miami, FL, United States, 5Department of Radiology and Radiological Science, The Johns Hopkins University, Baltimore, MD, United States, 6Department of Radiation Oncology, The Johns Hopkins University, Baltimore, MD, United States, 7Department of Biomedical Informatics, Emory University School of Medicine, Atlanta, GA, United States
Synopsis
Glioblastoma (GBM)
is a grade IV primary brain tumor with poor outcomes despite surgical
resection, chemotherapy, and radiation. Often, disease will recur in regions in
the penumbra of the treatment volume, hypothesized to occur because anatomic
MRI does not fully capture neoplastic infiltration. Spectroscopic magnetic
resonance imaging (sMRI) enables in vivo whole-brain
analysis of metabolic activity, and has been shown to sensitively and
specifically identify regions of non-enhancing, infiltrating tumor. We present
an ongoing prospective clinical study to target metabolically active tumor identified
by sMRI for a radiation boost, with the aim of improving outcome in patients
with GBM.
Introduction
Glioblastoma (GBM)
is a malignant primary adult brain tumor with an annual incidence of 7,000
cases in the United States (1). Current treatment recommendations
include maximal safe resection followed by concurrent radiation therapy (RT)
and temozolomide chemotherapy (2). High dose RT is targeted using T1-weighted
contrast-enhanced (CE-T1w) MRI, in which areas of enhancement represent areas
of tumor with leaky neovasculature. Lower dose RT is targeted using T2-weighted
fluid-attenuation inversion recovery (FLAIR) MRI, in which areas of
hyperintensity include a combination of infiltrative tumor, inflammation, and
vasogenic edema (3). Despite aggressive treatment,
median survival remains 15 months (4, 5). Often, disease recurs
within the penumbra of the treatment volume, hypothesized to occur because
CE-T1w and FLAIR do not capture the entirety of tumor, including areas of
proliferation beyond the tumor core.
Spectroscopic
magnetic resonance imaging (sMRI) is an evolution of MR spectroscopy that
enables 3D whole-brain volumes of metabolic activity to be obtained in vivo without any exogenous contrast
agents (6, 7) (Fig. 1). We have previously shown that the ratio of choline to
N-acetylaspartate (Cho/NAA) sensitively and specifically identifies tumor
proliferation; regions of metabolic activity expressing a Cho/NAA ratio more
than two times that of normal appearing white matter are be prone to recurrence
if left untreated (8). Despite its
potential to improve disease control, clinical adoption of sMRI has been
limited by difficulty incorporating it into clinical display software without
manual workarounds (9) and artifacts due
to magnetic field inhomogeneities (10-12). We present a
cloud software platform designed specifically to incorporate sMRI into the RT
planning workflow. and demonstrate its feasibility in guiding RT based on
elevated Cho/NAA in a multisite clinical study.Methods
We built the Brain
Imaging Collaboration Suite (BrICS), a web-based software designed to integrate
sMRI with clinical 3D MRI volumes, enabling physicians to evaluate metabolic
activity, review underlying spectra, and delineate target volumes for RT
planning (Fig. 2a). Data from
spectroscopy software, such as the Metabolite Imaging and Data Analysis Suite
(MIDAS), are automatically registered via affine transformation with clinical DICOM
volumes (CE-T1w MRI, FLAIR MRI). BrICS encapsulates algorithmic modules, such
as automated contouring of residual contrast enhancement, which reduces user
bias and improve reproducibility across patients (Fig. 2b).
While MIDAS
eliminates spectra that have broadened peak linewidths, some artifactual
spectra remain. We developed a convolutional neural network, trained on MR
experts’ quality classifications of spectral artifacts (poor/unacceptable
spectra), to perform additional filtering (Fig.
2c). It takes as input unfitted spectra and outputs a classification of “good”
or “poor” quality for each voxel; poor quality spectra are then filtered out.
To assess
feasibility of using BrICS in the multisite settings, patients were enrolled
from three institutions – Emory, Johns Hopkins, and U. Miami. Each underwent a
pre-RT sMRI and regions of the brain with a two-fold increase in Cho/NAA were automatically
contoured. Data were reviewed collaboratively in BrICS by two spectroscopists,
who assessed raw and fitted spectra, and a neuroradiologist; voxels with poor quality
spectra were eliminated from the target volume. Two radiation oncologists then
made final edits and validated volume for RT treatment. All user edits were
tracked in a digital “paper trail” for safety and retrospective review of the
pipeline. Finally, BrICS generated DICOM RT sets that are imported into
clinical LINAC stations. For this study, elevated Cho/NAA and residual
contrast-enhancing lesions were targeted for boosted radiation to 75 Gy; the remaining
tissue received standard-of-care therapy to 60/51 Gy (Fig. 3). Results
Two cases from this
study are shown. Patient one is a 21-year-old female with a frontal GBM at post-resection.
A small amount of residual contrast-enhancement (1.62 cm3) remains after
surgery (Fig. 4a), though the
Cho/NAA abnormality is much larger (48.21 cm3) and includes a
contralateral projection through the genu of the corpus callosum (Fig. 4b). Based on these two contours,
a 75 Gy boost (PTV3) was planned (Fig.
4c). Patient two is a 57-year-old male with a parieto-occipital GBM. Residual
parietal enhancement (2.59 cm3) and an anterior projection of
Cho/NAA (23.84 cm3) abnormality were used as the target volume for
RT boost (Fig. 5). Importantly,
analyses are performed on a central server, ensuring unbiased reproducibility
of contours.Conclusion
We have developed a cloud platform
specifically designed to enable collaborative sMRI-guided radiation therapy in
the multisite setting, paving the way for future consortium trials to evaluate
the efficacy of sMRI as a prognostic tool. With continued evolution of this
software, the data processing pipeline, and data from future studies, sMRI
moves closer to clinical use, and ultimately to improving patient outcomes.Acknowledgements
This
work is funded by the following grants from the National Institutes of Health: U01 CA172027 (Shim, Barker, Shu); R01 CA214557 (Shim, Shu); and F30 CA206291 (Gurbani).References
- Ostrom QT, Gittleman H, Liao P, Rouse C, Chen Y, Dowling J, et al. CBTRUS Statistical Report: Primary Brain and Central Nervous System Tumors Diagnosed in the United States in 2007–2011. Neuro Oncol. 2014;16(suppl_4):iv1-iv63.
- Wen PY, Macdonald DR, Reardon DA. Updated response assessment criteria for high-grade gliomas: response assessment in neuro-oncology working group. J Clin Oncol. 2010.
- Tsuchiya K, Mizutani Y. Preliminary evaluation of fluid-attenuated inversion-recovery MR in the diagnosis of intracranial tumors. Am J Neuroradiol. 1996.
- Stupp R, Hegi ME, Gilbert MR, Chakravarti A. Chemoradiotherapy in malignant glioma: standard of care and future directions. J Clin Oncol. 2007;25(26):4127-36.
- Stupp R, Hegi ME, Mason WP, van den Bent MJ, Taphoorn MJB, Janzer RC, et al. Effects of radiotherapy with concomitant and adjuvant temozolomide versus radiotherapy alone on survival in glioblastoma in a randomised phase III study: 5-year analysis of the EORTC-NCIC trial. Lancet Oncol. 2009;10(5):459-66.
- Law M. MR spectroscopy of brain tumors. Topics in Magnetic Resonance Imaging. 2004;15(5):291-313.
- Maudsley AA, Domenig C, Sheriff S. Reproducibility of serial whole‐brain MR Spectroscopic Imaging. NMR in Biomedicine. 2010.
- Cordova JS, Shu H-KG, Liang Z, Gurbani SS, Cooper LAD, Holder CA, et al. Whole-brain spectroscopic MRI biomarkers identify infiltrating margins in glioblastoma patients. Neuro Oncol. 2016;18(8):1180-9.
- Cordova JS, Gurbani SS, Olson JJ, Liang Z, Cooper LAD, Shu H-KG, et al. A systematic pipeline for the objective comparison of whole-brain spectroscopic MRI with histology in biopsy specimens from grade III glioma. Tomography. 2016;2(2):106-16.
- Kreis R. Issues of spectral quality in clinical 1H magnetic resonance spectroscopy and a gallery of artifacts. NMR in Biomedicine. 2004;17(6):361-81.
- Pedrosa de Barros N, McKinley R, Knecht U, Wiest R, Slotboom J. Automatic quality control in clinical (1)H MRSI of brain cancer. NMR Biomed. 2016;29(5):563-75.
- Pedrosa de Barros N, McKinley R, Wiest R, Slotboom J. Improving labeling efficiency in automatic quality control of MRSI data. Magn Reson Med. 2017.