INTRODUCTION

The treatment of metastatic NPC patients is generally palliative chemotherapy, still with a very poor treatment response. Because most metastatic patients will succumb rapidly to the disease, the development of a quick and accurate tool to predict distant metastasis (DM) at initial staging is urgently needed. We aimed to develop and validate a classifier integrated MRI-based textural features with pre-treatment plasma Epstein-Barr Virus (EBV) DNA for the prediction DM in NPC at initial staging.

METHODS

This is a retrospective cohort study. Two hundred and thirty-eight consecutive patients with biopsy-proven NPC were enrolled from August 2009 to January 2013, and follow-up was completed on December 2016. All of radiomic features were extracted from contrast-enhanced T1-weighted (CET1-w) MRI and T2-weighted (T2-w) MR images. Infinite feature selection (Inf-FS) was used to select the most important textural feature associated with DM. The Random Forests (RF) was applied for model development on 80% of the dataset (n=190) and validated on the remaining dataset (n=48). The performance of model was assessed using the area under the receiver operating characteristic curve (AUC).

RESULTS

The radiomic signatures, which consisted of 240 important features, was significantly associated with DM (P < 0.001 for the training and validation cohorts). The combined classifier integrating EBV DNA and radiomic signatures showed significant improvement over clinical variables alone for DM prediction (AUC, 0.840 versus 0.733; P < 0.001). The classifier successfully divided those patients into low- and high-risk groups with significant difference in 5-year survival (P < 0.001). Decision curve analysis showed the classifier outperformed the radiomic signature and the clinical variables alone in terms of clinical usefulness.

DISCUSSION

The present study found a combined classifier that integrated radiomic features with pretreatment EBV DNA for the prediction of DM in NPC patients. The combined classifier successfully stratified patients into high-risk and low-risk of DM, and 5-year survival probabilities of patient subgroups were performed. Combining the radiomic signatures and clinical variables into a classifier provides new tools for making optimal clinical decisions, enabling clinicians to early identify DM risk. To build a combined classifier, the most important feature was selected from 47 candidate features (43 textural and 4 clinical features) by Inf-FS method and repeated the process for 240 times in different scan-texture-parameter combinations. The Inf-FS method evaluates the importance of a given feature while considering all the possible subsets of features. Besides, this method assigned a score of “importance” to each feature, and then obtained a ranking of features. Finally, a total of 240 features were used to develop combined classifier according to random forests algorithm, and have the same performance rate through training and validation. The combined classifier stratified NPC patients into high risk and low risk of DM. Meanwhile, the significance of combined classifier was further evaluated by survival analysis, and was shown to correspond with a significant increase in 5-year survival compared with radiomic signatures and clinical variables alone.

CONCLUSION

By combining quantitative textural features and EBV DNA, our newly developed and validated classifier can be served as a powerful predictor of DM in NPC patients. It provides a good opportunity to improve patient counseling and individualize management of patients.

Acknowledgements

We acknowledge financial support from the National Natural Science Foundation of China (81571664); the Science and Technology Planning Project of Guangdong Province (2014A020212244, 2016A020216020); the Scientific Research General Project of Guangzhou Science Technology and Innovation Commission (201605110912158); and the China Postdoctoral Science Foundation (2016M600145). The authors declare no competing financial interests.

References

1. An X, Wang FH, Ding PR, et al. Plasma Epstein-Barr virus DNA level strongly predicts survival in metastatic/recurrent nasopharyngeal carcinoma treated with palliative chemotherapy. Cancer. Aug 15 2011;117(16):3750-3757.