Prediction of Glioblastoma Prediction of Glioblastoma Multiforme Patient Time to Multiforme Patient Time to Recurrence Using MR Image Recurrence Using MR Image

Features and Gene ExpressionFeatures and Gene Expression

Nicolasjilwan, M.1Nicolasjilwan, M.1··Clifford, R.2Clifford, R.2··Flanders, A. Flanders, A. E.3·Scarpace, L.4E.3·Scarpace, L.4··Raghavan, P.1Raghavan, P.1··Hammoud, Hammoud,

D.5D.5··Huang, E.6Huang, E.6··Jaffe, C.7Jaffe, C.7··Freymann, J.2Freymann, J.2··Kirby, Kirby, J.2J.2··Buetow, K.5Buetow, K.5··Huang, S.8Huang, S.8··Holder, Holder, C.8C.8··Gutman,D.8Gutman,D.8··Wintermark, M.1Wintermark, M.1

1University of Virginia, Charlottesville, VA, 2SAIC-Frederick, Inc., Frederick, 1University of Virginia, Charlottesville, VA, 2SAIC-Frederick, Inc., Frederick, MD, 3Thomas Jefferson University Hospital, Philadelphia, PA,4Henry MD, 3Thomas Jefferson University Hospital, Philadelphia, PA,4Henry

Ford University Hospital,Detroit,MI,5National Institute of Health, Ford University Hospital,Detroit,MI,5National Institute of Health, Bethesda, MD, 6National Cancer Institute, Bethesda, MD, 7Boston Bethesda, MD, 6National Cancer Institute, Bethesda, MD, 7Boston

University School of Medicine, Boston, MA, 8Emory University Hospital, University School of Medicine, Boston, MA, 8Emory University Hospital, Atlanta, GA.Atlanta, GA.

DISCLOSUREDISCLOSURE

Nothing to discloseNothing to disclose

PURPOSEPURPOSE

Utilize conventional MRI imaging Utilize conventional MRI imaging features to predict time to recurrence of features to predict time to recurrence of patients with glioblastoma multiforme patients with glioblastoma multiforme (GBM) after initial diagnosis.(GBM) after initial diagnosis.

Linear regression models incorporating Linear regression models incorporating MR imaging features and tumor gene MR imaging features and tumor gene expression to predict patient time to expression to predict patient time to recurrence.recurrence.

Important role in selecting treatment Important role in selecting treatment options. options.

Materials & Methods

The study is part of The Cancer Genome The study is part of The Cancer Genome Atlas (TCGA) MR imaging (MRI) Atlas (TCGA) MR imaging (MRI) characterization project of the National characterization project of the National Cancer Institute. Cancer Institute.

MR images for 70 GBM patients made MR images for 70 GBM patients made available through the National Biomedical available through the National Biomedical Imaging Archive were reviewed Imaging Archive were reviewed independently by six neuroradiologists. independently by six neuroradiologists.

The VASARI feature scoring system for The VASARI feature scoring system for human gliomas, developed at Thomas human gliomas, developed at Thomas Jefferson University Hospital, was employed.Jefferson University Hospital, was employed.

30 features clustered by categories.30 features clustered by categories.– Lesion LocationLesion Location– Morphology of Lesion SubstanceMorphology of Lesion Substance– Morphology of Lesion MarginMorphology of Lesion Margin– Alterations in Vicinity of LesionAlterations in Vicinity of Lesion– Extent of resectionExtent of resection

620 genes associated with angiogenesis used in this investigation; Verhaak et al. (Cancer cell 17:98)(Cancer cell 17:98)

Associations between imaging features and Associations between imaging features and time to recurrence were assessed using time to recurrence were assessed using linear regression models. Time to linear regression models. Time to recurrence was the outcome; imaging recurrence was the outcome; imaging features were the predictors.features were the predictors.

Well marginated Non-enhancingWell marginated Non-enhancing

F4 Enhancement Quality: 1=None 2=Mild/Minimal 3=Marked/Avid

F13 Definition of the non-enhancing margin 1= n/a 2= Smooth 3= Irregular Courtesy Dr Adam Flanders

Predominantly Non-Predominantly Non-enhancingenhancing

F5 Proportion Enhancing: 1= n/a 2=None (0%) 3= <5% 4= 6-33% 5= 34-67% 6= 68-95% 7= >95% 8=All (100%)

Courtesy Dr Adam Flanders

RESULTSRESULTS

Time to recurrence values obtained Time to recurrence values obtained for for 15 15 patients. patients.

Median time to recurrence: Median time to recurrence: 263 263 daysdays. .

Univariate Analysis of Association Univariate Analysis of Association between VASARI Features and Time to between VASARI Features and Time to

RecurrenceRecurrence

Individually, 2 MRI features show association Individually, 2 MRI features show association to time to recurrence with an unadjusted p-to time to recurrence with an unadjusted p-value < 0.05.value < 0.05.

Negative correlation with survival:Negative correlation with survival: Ependymal extension (F19), (P = 0.0445)Ependymal extension (F19), (P = 0.0445)

Positive correlation with survival: Positive correlation with survival: Location of the tumor in the left (usually Location of the tumor in the left (usually

dominant) hemisphere (F2), dominant) hemisphere (F2), (P = 0.0084).(P = 0.0084).

Multivariate Analysis incorporating VASARI Multivariate Analysis incorporating VASARI Imaging Features and The Expression of Imaging Features and The Expression of

Angiogenesis-related Genes.Angiogenesis-related Genes.

Optimal model constructed by addition and Optimal model constructed by addition and substraction of variables: substraction of variables: Vasari feature 2 Vasari feature 2 (location of the tumor in the left or right (location of the tumor in the left or right hemispherehemisphere) and expression of ) and expression of STAT1STAT1, , ARHGAP24ARHGAP24 and and SSTR2SSTR2 genes. genes.

Predicted time to recurrence based on the Predicted time to recurrence based on the model shows a Pearson correlation of model shows a Pearson correlation of 0.9720.972 (P = (P = 1.42e−09) with observed times to recurrence. 1.42e−09) with observed times to recurrence.

In contrast, tumor localization to the left In contrast, tumor localization to the left hemisphere alone shows a correlation of hemisphere alone shows a correlation of 0.6520.652 (P = 8.43e−03) with time to recurrence.(P = 8.43e−03) with time to recurrence.

CONCLUSIONCONCLUSION

A subset of VASARI imaging features correlate well A subset of VASARI imaging features correlate well with patient time to recurrence. with patient time to recurrence.

Linear regression models incorporating multiple Linear regression models incorporating multiple imaging features or a single VASARI feature imaging features or a single VASARI feature (localization of the tumor to the left or right (localization of the tumor to the left or right hemisphere) and tumor gene expression can be hemisphere) and tumor gene expression can be used to predict patient time to recurrence. used to predict patient time to recurrence.

We are refining these models and are investigating We are refining these models and are investigating whether including patient clinical characteristics into whether including patient clinical characteristics into linear models can improve their predictive power.linear models can improve their predictive power.

AKNOWLEDGMENTAKNOWLEDGMENT

University of Virginia Health SystemUniversity of Virginia Health SystemSAIC-FrederickSAIC-Frederick

Thomas Jefferson University Hospital Thomas Jefferson University Hospital

Henry Ford University Hospital Henry Ford University Hospital

National Institute of HealthNational Institute of Health

National Cancer InstituteNational Cancer Institute

Emory University HospitalEmory University Hospital

Boston University School of MedicineBoston University School of Medicine