2224

Development and validation of a MRI-based radiomics prognostic classifier in patients with primary glioblastoma multiforme

Xin Chen¹, Zaiyi Liu², Xinqing Jiang¹, Xinhua Wei¹, and Zhongping Zhang³

¹Department of radiology, Guangzhou First People's Hospital, Guangzhou, People's Republic of China, ²Department of radiology, Guangdong General Hospital, Guangdong Academy of Medical Sciences, Guangzhou, People's Republic of China, ³GE Healthcare China, Guangzhou, People's Republic of China

Synopsis

Glioblastoma multiforme (GBM) is the most frequent and deadly type of primary malignant tumor in the central nervous system with a median survival range of only 9-15 months. We developed and validated a MRI-based radiomics classifier to predict overall survival(OS) in patients with newly diagnosed GBM. The results showed that a 7-radiomics classifier allows prediction of survival and stratification of patients with newly diagnosed GBM into a low- or high-risk group for OS.

Introduction

Glioblastoma multiforme (GBM) is the most frequent and deadly type of primary malignant tumor in the central nervous system. Standard treatment consists of maximal surgical resection followed by adjuvant chemotherapy and radiation therapy, which could improve the median survival time of GBM from 15.3 months to 21.7 months. However the survival time varies widely from 3 months to 3 years after diagnosis¹, which underscored the limitations of current clinicopathologic markers and grading systems in predicting patient prognosis. Therefore, this study aimed to develop and validate a MRI-based radiomics classifier to predict overall survival(OS) in patients with newly diagnosed GBM.

Methods

Clinical and MR imaging data of 127 patients with GBM were obtained from the Cancer Genome Atlas and the Cancer Imaging Archive. Regions of interest(ROIs) corresponding to enhanced portions of tumor were drawn on the post-contrast T1 weighted images of 127 patients (allocated in a 2:1 ratio to a training[n=85] or validation[n=42] set), and a total of 3824 radiomics features including first-order and texture features were derived from these ROIs. Radiomics features of patients in the training set were subjected to the minimum redundancy maximum relevance (mRMR) algorithm and Cox proportional hazard regression model to predict OS. The performance of a cox proportional hazards model with the Akaike’s information criterion(AIC) criteria was assessed with concordance index(C-index).

Results

Seven radiomics features were selected as a classifier in the training set could stratify patients with GBM into a low- or high-risk group for OS(HR, 3.089, 95% CI,1.793-5.322; P＜0.001), and the results were validated successfully in the validation set(HR, 2.065, 95%CI, 0.996-4.281; P＜0.001)（Figure1). The performance of the radiomics classifier in training set(C-index, 0.790, 95%CI, 0.684-0.896) was similar to that in validation set(C-index, 0.737,95% CI:0.566-0.908).

Discussion

Many findings of previous studies have demonstrated that major differences in phenotypes or gene-protein classifiers can be correlated to radiographic findings^2,3. And a recent literature on the patterns of genomic instability in metastatic pancreatic cancer showed that tumors with more genomic heterogeneity were more likely to have a worse prognosis⁴. Accordingly, tumor heterogeneity assessed through imaging could be translated to an expression of genomic heterogeneity and would exhibit poor prognosis. The 7 radiomics features selected for the radiomics classifier derived from the enhanced portion of tumor in this study are in consistent with the recent researches on risk stratification of survival outcomes of patients⁵. While further studies on a causal relationship between gene expression and image features were needed to be extended.

Conclusion

A 7-radiomics classifier allows prediction of survival and stratification of patients with newly diagnosed GBM.

Acknowledgements

Zhongping Zhang is an employee of GE Healthcare, China, but does not have any conflict of interest with this project.

References

1. Hegi ME, Diserens AC, Gorlia T, et al. mgmt gene silencing and benefit from temozolomide in glioblastoma. N Engl J Med. 2005; 352 (10): 997-1003

2. Lambin P, Rios-Velazquez E, Leijenaar R, et al. radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48(4):441-446

3. Hobbs SK, Shi G, Homer R. et al. Magnetic resonance image-guided proteomics of human glioblastoma multiforme. J Magn Reson Imaging. 2003;18(5): 530-536

4. Campbell PJ, Yachida S, Mudie LJ, et al. The patterns and dynamics of genomic instability in metastatic pancreatic cancer. Nature. 2010;467(7319): 1109-1113

5. Kickingereder P, Burth S, Wick A, et al. Radiomic Profiling of Glioblastoma: Identifying an Imaging Predictor of Patient Survival with Improved Performance over Established Clinical and Radiologic Risk Models. Radiology. 2016;280(3): 880-889

Figures

Figure 1a. Risk score by the radiomics classifier(left panel), Kaplan-Meier survival (right panel) in the training set(a), validation set(b). Data are hazard ratio(95% CI). We calculated p values using the log-rank test.

Figure 1b. Risk score by the radiomics classifier(left panel), Kaplan-Meier survival (right panel) in the training set(a), validation set(b). Data are hazard ratio(95% CI). We calculated p values using the log-rank test.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)

2224