Large-scale radiomic profiling of glioblastoma identifies an imaging signature for predicting and stratifying antiangiogenic treatment response.
Philipp Kickingereder1, Michael Götz2, John Muschelli3, Antje Wick4, Ulf Neuberger5, Russell T Shinohara6, Alexander Radbruch7, Heinz-Peter Schlemmer7, Wolfgang Wick4, Martin Bendszus5, Klaus H Maier-Hein2, and David Bonekamp7

1Department of Neuroradiology, University Hospital Heidelberg, Heidelberg, Germany, 2Division Medical and Biological Informatics, DKFZ - German Cancer Research Center, Heidelberg, Germany, 3Department of Biostatistics, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD, United States, 4Department of Neurology, University of Heidelberg Medical Center, Heidelberg, Germany, 5Department of Neuroradiology, University of Heidelberg Medical Center, Heidelberg, Germany, 6Department of Biostatistics and Epidemiology, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, United States, 7Department of Radiology, DKFZ - German Cancer Research Center, Heidelberg, Germany

Synopsis

To analyze the potential of radiomics, an emerging field of research that aims to utilize the full potential of medical Imaging (1,2), for predicting and stratifying treatment response to antiangiogenic therapy in patients with recurrent glioblastoma.

Introduction

Antiangiogenic treatment with bevacizumab is the single most widely used therapeutic agent for patients with recurrent glioblastoma (GB). A major challenge is that there are currently no validated biomarkers that can delineate molecular activity and predict treatment Response (1,2). Here we analyze the potential of radiomics, an emerging field of research that aims to utilize the full potential of medical Imaging (3,4).

Methods

A total of 4842 quantitative MRI features (including first-order, volume and shape features and texture features) were automatically extracted from the complete three-dimensional tumor bulk of 129 patients with recurrent GB prior antiangiogenic therapy following transformation of arbitrary MR signal into quantitative intensities (hybrid-white-stripe statistical normalization(5)), then screened for association with treatment outcome using bootstrap-resampling methods, and finally analyzed using unsupervised hierarchical clustering for radiomic signature discovery (Fig. 1). The utility of the obtained radiomic clusters on stratifying response to antiangiogenic therapy was modeled via Cox proportional hazards models with bootstrap resampling for internal validation.

Results

Leveraging a high throughput approach, we identified a radiomic signature consisting of 82 MRI features that segmented patients to three distinct consensus clusters (Fig. 2) which predict and stratify treatment response to antiangiogenic therapy by means of both progression-free and overall survival in patients with recurrent GB (Fig. 3). We furthermore provide evidence that the cluster of tumors with the most favorable treatment response arises from an anatomically distinct niche location.

Conclusions

Our radiomic signature emerges as a putative imaging biomarker for predicting response to antiangiogenic therapy in patients with recurrent GB, advances the knowledge in the non-invasive analysis and characterization of brain tumours, and stresses the role of radiomics as a novel tool for improving decision-support in cancer treatment at low cost.

Acknowledgements

No acknowledgement found.

References

1. Lu-Emerson C, Duda DG, Emblem KE, et al: Lessons from anti-vascular endothelial growth factor and anti-vascular endothelial growth factor receptor trials in patients with glioblastoma. J Clin Oncol 33:1197-213, 2015

2. Mayer TM: Can We Predict Bevacizumab Responders in Patients With Glioblastoma? J Clin Oncol 33:2721-2, 2015

3. Aerts HJ, Velazquez ER, Leijenaar RT, et al: Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 5:4006, 2014

4. Lambin P, Rios-Velazquez E, Leijenaar R, et al: Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 48:441-6, 2012

5. Shinohara RT, Sweeney EM, Goldsmith J, et al: Statistical normalization techniques for magnetic resonance imaging. Neuroimage Clin 6:9-19, 2014

Figures

Image post-processing workflow.

Heatmap of radiomic features within the three consensus clusters. Each row corresponds to one patient, each column to one z-score normalized radiomic feature. Cluster 2 primarily consists of features with low z-scores (“low z-score signature”), Cluster 3 of features with high z-score values (“high z-score signature”) and the remaining Cluster 1 of features with intermediate z-scores (“indifferent signature”)

Kaplan-Meier plot for overall and progression-free survival (OS, PFS) stratified by the three radiomic consensus clusters.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0289