Introduction

The overall survival (OS) of head and neck squamous cell carcinoma (HNSCC) patients is extremely low because the rate of regional and distant metastases at diagnosis is high. Besides, This ratio has improved slightly last several years, despite advances in multidisciplinary management.(1-2) The term radiomics has aroused widespread concern in recent years, and it is the process of conversion of imaging data into high-dimensional, mineable data by using a large number of automatically extracted data-characterization algorithms.(3-4) This method can quantitatively measure the heterogeneity of the tumor, so patients can be assessed well even for the same stage. Nowadays the most widely used imaging modality in radiomics research has been magnetic resonance imaging (MRI), which can quantify tissue density. Evidence has been accumulating to suggest that radiomics features derived from magnetic resonance imaging (MRI) could bring valuable prognostic information in glioblastoma,(5-8) and prostate cancer.(9) While study conducted to investigate the potential value of this technique in prognostic prediction of the prognosis of HNSCC patients with T2-weighted (T2W) MRI.

Methods

Patient From January 2013 to December 2015, medical records and images from patients with histopathologically confirmed HNSCC who underwent preoperative MRI and received treatment at our institution were included. Patients were excluded for any of the following reasons: (1) tumor size smaller than 5 mm, (2) imaging artifacts (motion or susceptibility artifacts) that impaired the correct segmentation of cancer, (3) any prior history of head and neck cancer, or (4) treatment (surgery, chemotherapy, or radiation) for the cancer before the MRI scan. MRI and Segmentation MRI examinations were performed on a 1.5-T scanner (Signa Excite; GE Medical Systems; Milwaukee, Wisconsin, USA). The imaging protocol included axial T1-weighted imaging (repetition time [TR]/echo time [TE], 400-600/10 ms), axial T2W imaging (TR/TE, 3200/100 ms). Primary tumors of all patients were separately segmented slice by slice by two experts with more than 7 years of experience. Forty cases were segmented respectively by two radiologists in the double-blinded way. The rest of CT scans were segmented by one radiologist and reviewed by another radiologist. The inter-class correlation coefficient (ICC) was used to determine the inter-observer agreement of these features, and an ICC greater than 0.75 was considered as a mark of excellent reliability(10). Radiomics Feature extraction and Statistical Analysis Within each volume of interest ,485 features consisted of four type: first-order, shape and size, textural and wavelet features were extracted within the image region defined by the segmentation using in-house algorithms.We divided all patients into a training cohort and a validation cohort at a 1:1 ratio using random number generated by computer. Feature normalization was used to correction factors. The LASSO penalized Cox proportional hazards regression was adopted to select critical features with weights. Finally, the radiomics signature was built by critical features with weights according to that scores can be computed for each patient.The potential association of the radiomics signature with OS of patients was assessed in the training cohort and then validated in the testing cohort with a univariate Cox analysis. The patients were classified into high-risk or low-risk groups according to the median value of the Rad-score in the training cohort.

Results

A total of 170 patients were included. Eighty-five patients each were allocated to the training and testing cohorts. Of the 485 features were extracted from T2W images for each patient, 53 features were associated with OS in the univariate analysis. Seven features including X0_fos_entropy_p, X0_fos_uniformity_p, X3_fos_skewness, X4_fos_maximum, X4_fos_maximum, X7_GLCM_correlation, X1_GLRLM_mean were selected by LASSO Cox regression model. The radiomics signature was associated with OS in the training cohort (P < 0.0001; hazard ratio [HR], 3.83; 95% confidence interval [CI], 2.28-6.44) and in the testing cohort (P = 0.0013; HR, 1.84; 95% CI, 1.35-2.51; Fig 2). On the basis of the median Rad-score of 6.0 in the training cohort, the patients in each cohort were classified into a high-risk group (Rad-score > 6.0) or a low-risk group (Rad-score ≤ 6.0). The log-rank test showed that patients in the low-risk group had better OS than patients in the high-risk group in both training cohort (P < 0.001) and testing cohort (P = 0.0013).

Discussion

Radiomics features with the highest prognostic power were selected and a novel 7-feature based radiomics signature was built, which was validated in the testing cohort and confirmed as an independent predictor for OS in HNSCC patients.

Conclusion

A pretreatment T2W MRI–based radiomics signature was developed and validated as a convenient approach to predict OS in patients with HNSCC.

Acknowledgements

We acknowledge the National Natural Science Foundation of China (81227901, 81771924, 81501616, 61231004, 81671851, and 81527805), National Key R&D Program of China (2017YFA0205200, 2017YFC1308700, 2017YFC1308701, 2017YFC1309100), the Science and Technology Service Network Initiative of the Chinese Academy of Sciences (KFJ-SW-STS-160), the Instrument Developing Project of the Chinese Academy of Sciences (YZ201502), the Beijing Municipal Science and Technology Commission (Z161100002616022), and the Youth Innovation Promotion Association CAS.

References

1. Siegel RL, Miller KD, Jemal A. Cancer Statistics, 2017. CA Cancer J Clin 67:7-30, 2007 2. Torre LA, Bray F, Siegel RL, et al: Global cancer statistics, 2012.CA Cancer J Clin 65:87-108, 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 [Erratum: Nat Commun 5:4644, 2014] 4. Gillies RJ, Kinahan PE, Hricak H: Radiomics: Images are more than pictures, they are data. Radiology 278:563-577, 2016 5. Cui Y, Ren S, Tha KK, et al: Volume of high-risk intratumoral subregions at multi-parametric MR imaging predicts overall survival and complements molecular analysis of glioblastoma. Eur Radiol. 2017 Feb 6. doi: 10.1007/s00330-017-4751-x. [Epub ahead of print]. 6. Kickingereder P, Götz M, Muschelli J, et al: Large-scale radiomic profiling of recurrent glioblastoma identifies an imaging predictor for stratifying anti-angiogenic treatment response. Clin Cancer Res. 2016;22:5765-5771. 7. Prasanna P, Patel J, Partovi S, et al: Radiomic features from the peritumoral brain parenchyma on treatment-naïve multi-parametric MR imaging predict long versus short-term survival in glioblastoma multiforme: Preliminary findings. Eur Radiol. 2016 Oct 24. doi: 10.1007/s00330-016-4637-3. [Epub ahead of print]. 8. Zinn PO, Singh SK, Kotrotsou A, et al: Clinically applicable and biologically validated MRI radiomic test method predicts glioblastoma genomic landscape and survival. Neurosurgery. 2016;63 Suppl 1:156-7. 9. Gnep K, Fargeas A, Gutiérrez-Carvajal RE, et al: Haralick textural features on T2 -weighted MRI are associated with biochemical recurrence following radiotherapy for peripheral zone prostate cancer. J Magn Reson Imaging. 2017;45:103-17. 10. Barry WT, Kernagis DN, Dressman HK, Griffis RJ, Hunter JD, Olson JA, et al. Intratumor heterogeneity and precision of microarray-based predictors of breast cancer biology and clinical outcome. J Clin Oncol. 2010; 28:2198–206.