Tumor heterogeneity in oral and oropharyngeal squamous cell carcinoma assessed by texture analysis of CT and conventional MRI: a potential marker of overall survival

Abstract

Background

The combined prognostic value of texture analysis on computed tomography (CT) and conventional magnetic resonance imaging (MRI) in oral and oropharyngeal squamous cell carcinoma (OOSCC) has rarely been reported.

Purpose

To evaluate the ability of texture-based heterogeneity indices extracted from CT and conventional MRI to predict the overall survival (OS) of OOSCC patients.

Material and Methods

Eighty-eight patients with OOSCC underwent contrast-enhanced CT, and T2-weighted (T2W) and contrast-enhanced T1-weighted (T1W) imaging. For each imaging protocol, four histogram (HISTO) and six gray-level co-occurrence matrix (GLCM) parameters were extracted. Intraclass correlation coefficients (ICCs) were used to evaluate variation in inter-observer delineation. The association of clinicopathologic factors and texture parameters with OS was analyzed using log-rank testing and Cox regression. The Harrell Concordance Index (C-index) was used to evaluate the prognostic performance of the predictors and Cox model.

Results

Higher inter-observer delineation agreement of texture analysis on contrast-enhanced T1W imaging was observed relative to that on CT and T2W imaging. T stage (P = 0.026, hazard ratio [HR] = 2.35), GLCM_Entropy on CT (P = 0.003, HR = 3.15), and HISTO_Energy (P = 0.036, HR = 0.27) and HISTO_Skewness (P = 0.026, HR = 0.41) on contrast-enhanced T1W imaging were independent predictors for the OS of OOSCC patients. The multivariate Cox model performed strongly in predicting OS, with a significantly higher C-index than the independent predictors separately (C-index: 0.751 vs. 0.600–0.627, P < 0.005 for all).

Conclusion

Texture analysis using CT and conventional MRI could provide a valuable method for survival stratification of OOSCC patients.

Keywords

Head and neck cancer heterogeneity computed tomography magnetic resonance imaging prognosis texture analysis

Introduction

Oral and oropharyngeal squamous cell carcinoma (OOSCC) is one of the most common malignant tumors worldwide (1,2). Nearly two-thirds of patients present with locally advanced disease at the diagnosis and only 30–50% of these patients live for three years after the initiation of standard therapy (3,4). The intra-tumor heterogeneity of OOSCC has been proposed to be one of the major underlying factors of tumor progression and resistance to radiochemotherapy (5,6). It is therefore necessary to explore the relationship between intra-tumor heterogeneity and survival in OOSCC.

Percutaneous biopsy and postoperative pathology are commonly used for intra-tumor characterization. However, these procedures are invasive and fail to provide comprehensive information for the entire tumor. It has been accepted that intra-tumor heterogeneity occurs at the molecular level but is reflected macroscopically by the texture patterns within medical images (7).

“Image texture” refers to the complex pattern within an image. Texture analysis allows mathematic evaluation of the subtle spatial arrangement of an image’s pixels. Over the past few years, texture analysis has increasingly been investigated as a method to predict treatment outcome and survival in patients with colorectal cancer, pulmonary adenocarcinoma, esophageal cancer, or glioblastoma (7 –10). Studies have shown that the texture features of computed tomography (CT) images are predictive of the overall survival (OS) and treatment outcome of patients with OOSCC (2,11). Compared with CT, magnetic resonance imaging (MRI) provides enhanced contrast of soft tissue. However, the prognostic value of texture analysis on conventional MRI in OOSCC has been reported rarely (12,13).

We wished to evaluate the ability of the indices of texture-based heterogeneity extracted from CT and conventional MRI to predict the OS of patients with OOSCC.

Material and Methods

Patients

The protocol for this retrospective study was approved by the ethics committee of our hospital and the requirement for obtaining informed consent from patients was waived.

A comprehensive search of our institutional database for medical records from July 2011 to April 2015 was undertaken to identify patients with OOSCC. Patients were included based on the following criteria: (i) OOSCC confirmed by postoperative pathology (tumor–node–metastasis (TNM) stage: I–IV); (ii) short axis of mass ≥5 mm; and (iii) contrast-enhanced CT and conventional MRI undertaken before treatment. The exclusion criteria were: (i) prior history of head and neck cancer or receipt of treatment before imaging; and (ii) poor image quality that made tumor segmentation difficult and texture analysis inaccurate (e.g. motion or metal artifacts).

The clinicopathologic characteristics of patients were recorded: demographic information (age, sex); risk factors (tumor volume, tobacco use, location of primary mass); pathologic TNM stage; histology grade; and treatment. The TNM stage was determined using the 7th American Joint Committee on Cancer TNM Staging System Manual (14). In the present study, OS was defined as the time from preoperative examination until the date of death from any cause, or the date of the final follow-up.

Image acquisition

Patients underwent CT using 64 detector-row machines (Brilliance 64; Philips Medical Systems, Cleveland, OH, USA). The parameters of contrast-enhanced CT were: 120 or 140 kV; 240 or 398 mA; slice thickness = 0.6 or 1 mm; matrix = 512 × 512; field of view (FOV) = 240 × 240 mm; 60s delay after intravenous injection of contrast (60–100 mL; ioversol and iohexol).

Patients underwent MRI using a head and neck array coil with a 1.5-T scanner (Signa Excite; GE Medical Systems, Milwaukee, WI, USA). The parameters of MRI acquisition were: axial fat-suppressed T2-weighted (T2W) imaging sequence (repetition time [TR]/echo time [TE] = 4520/87 ms; matrix = 512 × 512; FOV = 240 × 240 mm; thickness = 4 mm; gap = 3 mm); and axial fat-suppressed contrast-enhanced T1W imaging sequence (TR/TE = 700/9 ms; matrix = 512 × 512; FOV = 240 × 240 mm; thickness = 3 mm; gap = 4 mm; delay of 8–10 min after intravenous injection of contrast (0.1 mmol/kg body weight; Magnevist, Schering, Berlin, Germany).

Image analysis

For each primary cancer, a volume of interest (VOI) was drawn on all slices of a visible tumor using ITK-SNAP software (www.itksnap.org). The necrosis components were included in the VOI. For each case, the tumor was outlined by two independent radiologists (with three and seven years of experience in CT- and MRI-based interpretation of head and neck diseases, respectively) blinded to the final survival results. The mean value of the two measurement results was adopted for survival analysis.

All texture features were calculated using LIFEx software (www.lifexsoft.org) (15). Four histogram (HISTO) features (kurtosis, skewness, energy, and entropy) and six gray-level co-occurrence matrix (GLCM) features (contrast, correlation, dissimilarity, energy, entropy, and homogeneity) were extracted from each imaging protocol. Before computation of texture features, voxel intensities were discretized to 128 gray levels. In addition, voxel values were rescaled automatically between mean ± 3SD, where the mean and SD are the mean value and standard deviation of the voxels within the VOI. The texture features were interpreted in detail and are shown in the Supplementary File.

Statistical analysis

R statistical and computing software (www.r-project.org) was used to undertake statistical analysis.

The inter-observer delineation agreements for texture analysis on CT and MRI were assessed using intraclass correlation coefficients (ICCs) based on the level of agreement: 0–0.40 = poor; 0.40–0.59 = fair; 0.60–0.74 = good; and 0.75–1.00 = excellent (16). ICC ≥0.75 was selected as an accepted reliability. Only the texture features considered to be “stable” were used for survival analysis.

The Kaplan–Meier method and log-rank testing were used for univariate analysis. The analyzed variables were clinicopathologic factors and texture parameters. Continuous variables, such as age, tumor volume, and texture parameters, were dichotomized before log-rank testing based on the optimal threshold values obtained using the “surv_cutpoint” function of the “survminer” R package. The function could determine the cut-off point by using the maximally selected rank statistics from the “maxstat” R package, which is an outcome-oriented method providing a value that corresponds to the most significant relation with survival. For log-rank testing, P < 0.05 was considered significant. Due to the small sample size, signiﬁcant textural parameters yielding <20 patients per group for comparison were not reported or censored.

Multivariate Cox regression analysis (stepwise) was used to identify the independent predictors of OS. The Harrell Concordance Index (C-index) was used to evaluate the prognostic performance of the predictors and Cox model.

Results

The study population consisted of 88 patients (61 men, 27 women; mean age = 58 ± 10.6 years). Median follow-up was 1055 days. During follow-up, 28 patients (32%) died and the probability of 36-month OS was 65%. The clinicopathologic characteristics of patients are summarized in Table 1.

Table 1.

Patient characteristics.

Characteristic	Study population (n = 88)
Age (years, mean±SD)	58 ± 10.6
Sex
Female	27 (31)
Male	61 (69)
Tobacco use
Yes	38 (43)
No	50 (57)
Location of primary mass
Oral cavity	64 (73)
Oropharynx	24 (27)
T stage
1–2	60 (68)
3–4	28 (32)
N stage
0	39 (44)
1–2	49 (56)
Overall stage
I–II	28 (32)
III–IV	60 (68)
Histopathology grade
Well differentiated	30 (34)
Moderately differentiated	36 (41)
Poorly differentiated	22 (25)
Treatment*
1	25 (28)
2	39 (44)
3	24 (27)

Values are presented as n (%) unless otherwise indicated.

*1, 2, and 3 denotes surgery, surgery + radiotherapy, and surgery + radiotherapy + chemotherapy, respectively.

Higher delineation agreement of texture analysis based on contrast-enhanced T1W imaging (ICCs = 0.72–0.94) between observers was shown relative to that based on CT (0.31–0.85) and T2W imaging (0.44–0.93). Five features from CT, five features from T2W imaging, and nine features from contrast-enhanced T1W imaging considered to be stable were used for survival analysis. The ICCs for these three imaging sequences are shown in Table 2.

Table 2.

Inter-observer ICCs for texture analysis for different imaging protocols.

Texture parameters	ICC (95% CI)
Texture parameters	CT	T2W imaging	Contrast-enhanced T1W imaging
HISTO_Kurtosis	0.33 (0.23–0.56)	0.44 (0.21–0.58)	0.72 (0.60–0.80)
HISTO_Skewness	0.31 (0.11–0.49)	0.55 (0.39–0.68)	0.85 (0.78–0.90)*
HISTO_Energy	0.80 (0.71–0.86)*	0.70 (0.57–0.78)	0.76 (0.67–0.84)*
HISTO_Entropy	0.79 (0.70–0.86)*	0.67 (0.54–0.77)	0.82 (0.74–0.88)*
GLCM_Contrast	0.69 (0.56–0.79)	0.75 (0.64–0.83)*	0.85 (0.78–0.90)*
GLCM_Correlation	0.64 (0.51–0.75)	0.78 (0.68–0.85)*	0.82 (0.74–0.88)*
GLCM_Dissimilarity	0.72 (0.60–0.81)	0.79 (0.70–0.86)*	0.86 (0.80–0.91)*
GLCM_Energy	0.85 (0.77–0.89)*	0.62 (0.48–0.74)	0.93 (0.89–0.95)*
GLCM_Entropy	0.81 (0.72–0.87)*	0.93 (0.88–0.95)*	0.94 (0.90–0.96)*
GLCM_Homogeneity	0.76 (0.66–0.84)*	0.86 (0.79–0.90)*	0.86 (0.79–0.91)*

*Texture parameters considered to be stable (ICC ≥ 0.75).

ICC, intraclass correlation coefficient; CI, conﬁdence interval; HISTO, histogram; GLCM, gray-level co-occurrence matrix.

With regard to clinicopathologic factors, tumor volume, T stage, N stage, and TNM stage were associated significantly with the OS of patients with OOSCC (P < 0.05 for all); significance was not found in age, sex, tobacco use, location of primary mass, histology grade, or treatment (P > 0.05 for all). With respect to texture features, higher GLCM_Entropy on CT and HISTO_Entropy, GLCM_Entropy and GLCM_Homogeneity on contrast-enhanced T1W imaging were associated significantly with lower OS (P < 0.05 for all). Lower GLCM_Energy on CT and HISTO_Energy, HISTO_Skewness and GLCM_Contrast on contrast-enhanced T1W imaging were associated significantly with lower OS (P < 0.05 for all). Significant differences in OS were not found according to the texture features from T2W imaging.

Multivariate Cox analysis showed that T stage (P = 0.026, hazard ratio [HR] = 2.35), GLCM_Entropy on CT (P = 0.003, HR = 3.15), HISTO_Energy (P = 0.036, HR = 0.27), and HISTO_Skewness on contrast-enhanced T1W imaging (P = 0.026, HR = 0.41) were independent predictors of OS. The C-index did not demonstrate significant differences between the independent predictors (C-index = 0.600–0.627, P > 0.05 for all) and was slightly higher in GLCM_Entropy on CT. The multivariate Cox model (C-index = 0.751) showed significantly better prognostic performance than those shown by the independent factors separately (P < 0.005 for all). The signiﬁcant variables predicting OS in univariate and multivariate analysis are given in Table 3. The prognostic performance of the predictors and Cox model are shown in Table 4. The survival-curve results of the independent predictors are summarized in Fig. 1. Representative cases for the comparison of texture parameters between two patients with different OS are shown in Fig. 2.

Fig. 1.

Kaplan–Meier curves of T stage (a), GLCM_Entropy on CT (b), HISTO_Energy (c) and HISTO_Skewness (d) on contrast-enhanced T1W imaging. These variables were independent prognostic factors for the overall survival of patients with OOSCC.

Fig. 2.

Representative cases for a comparison of texture parameters between two patients with different overall survival. (a, b) Patient with short survival (24 months). (c, d) Patient with relatively long survival (>36 months).

Table 3.

Signiﬁcant variables predicting overall survival.

Variables	Categories	Univariate analysis			Multivariate analysis
Variables	Categories	HR	95% CI	P value	HR	95% CI	P value
Clinicopathological factors
Tumor volume (mL)	>0.66 vs. <0.66	4.06	1.69–9.68	0.038
T stage	T3–4 vs. T1–2	2.56	1.12–5.84	0.010	2.35	1.11–4.98	0.026
N stage	N1–2 vs.N0	2.21	1.05–4.64	0.042
TNM stage	III–IVvs. I–II	3.41	1.58–7.36	0.015
Texture on CT
GLCM_Energy	>0.726 × 10⁻³ vs. <0.726 × 10⁻³	0.36	0.17–0.80	0.006
GLCM_Entropy	>3.261 vs. <3.261	2.73	1.21–6.15	0.006	3.15	1.47–6.76	0.003
Texture on contrast-enhanced T1W imaging
HISTO_Energy	>0.015 vs. <0.015	0.27	0.12–0.59	0.020	0.27	0.08–0.92	0.036
HISTO_Entropy	>1.897 vs. <1.897	3.27	1.50–7.07	0.020
HISTO_Skewness	>0.091 vs. <0.091	0.39	0.18–0.85	0.013	0.41	0.20–0.89	0.026
GLCM_Contrast	>351.8 vs. <351.8	0.46	0.21–1.02	0.034
GLCM_Entropy	>3.325 vs. <3.325	2.08	0.96–4.49	0.048
GLCM_Homogeneity	>0.159 vs. <0.159	2.81	1.33–5.94	0.019

HISTO, histogram; GLCM, gray-level co-occurrence matrix; HR, hazard ratio; CI, conﬁdence interval.

Table 4.

Prognostic performance of the predictors and model.

Model	C-index	95% CI	P value*
T stage	0.611	0.523–0.699	0.002
GLCM_Entropy (CT)	0.627	0.537–0.717	0.003
HISTO_Energy (contrast-enhanced T1W imaging)	0.600	0.512–0.688	0.001
HISTO_Skewness (contrast-enhanced T1W imaging)	0.610	0.516–0.704	0.001
Cox model	0.751	0.639–0.861

*The C-index was compared between the multivariate Cox model and independent factors separately.

C-index, Harrell’s Concordance Index; CI, conﬁdence interval; HISTO, histogram; GLCM, gray-level co-occurrence matrix.

Discussion

For patients with OOSCC, intra-tumor heterogeneity is a major attribute contributing to a poor and variable prognosis (5,17). Considering the variation observed in intra-tumor heterogeneity using different imaging protocols, we characterized this heterogeneity by extracting the texture features from CT and conventional MRI. We found texture analysis on contrast-enhanced T1W imaging to have better inter-observer agreement than that on CT and T2W imaging. Univariate and multivariate survival analysis showed that T stage, GLCM_Entropy on CT, and HISTO_Energy and HISTO_Skewness on contrast-enhanced T1W imaging were independent predictors of OS. The multivariate Cox model incorporating the independent predictors showed good performance for predicting OS.

In recent years, measurement of spatial heterogeneity by texture analysis has increasingly been used to predict treatment outcome or survival in oncology patients (11,18–23). In the head and neck, Zhang et al. (2) found entropy and skewness on CT images to be associated with the OS of patients with locally advanced OOSCC, who had been treated by induction chemotherapy with TPF (cisplatin, 5-fluorouracil, and docetaxel). In addition, Kuno et al. (11) recently demonstrated that texture features can be used to predict local failure in patients with OOSCC treated with chemoradiotherapy. As an alternative imaging method used routinely to diagnose and monitor OOSCC, MRI provides both better soft-tissue contrast and more comprehensive information with multiple sequences. Therefore, we chose to evaluate the prognostic value of intra-tumor heterogeneity-associated texture features from CT and conventional MR images. This strategy has rarely been reported.

In this context, a histogram shows the accumulation of pixels of the same intensity within the whole image and the correlated features represent the statistical properties without taking account of the relative spatial position of pixels (24). The GLCM describes the correlation between two pixels with certain distances and directions, which is a typical method for the description of second-order textures (25). Therefore, the features of a histogram and GLCM can characterize the global and local heterogeneity of OOSCC on different scales (7). Only the features reflecting intra-tumor heterogeneity were used because we wished to investigate the impact of the different heterogeneous characterizations on the prognosis, although useful image information may have been missed as a result.

Texture features from T2W imaging and contrast-enhanced (CT and T1W) images reflect the heterogeneity of tumor water content and blood supply, respectively (12,19). However, texture features from T2W imaging were not found to be significantly associated with OS. Taken together, the heterogeneity of the blood supply may have more impact on the prognosis. In addition, high GLCM_Entropy and HISTO_Entropy, and low GLCM_Energy representing significant pixel gray-level heterogeneity were identified to predict poor OS. There was no direct comparison between the mathematic importance of these features and underlying tumor histopathology. Nevertheless, the pixel gray-level heterogeneity on CT and contrast-enhanced T1W imaging could represent increased heterogeneity or hypoxic voids within the tumor. In multivariate Cox analysis, GLCM_Entropy on CT, HISTO_Energy, and HISTO_Skewness on contrast-enhanced T1W imaging were identified as independent predictors. These findings suggested that the local heterogeneity on CT and global heterogeneity on contrast-enhanced T1W imaging had key roles in OS prediction. Furthermore, the slightly higher C-index in GLCM_Entropy on CT also indicated that CT can provide complementary information about tumor heterogeneity to contrast-enhanced T1W imaging, although it has relatively inferior contrast of soft tissue. When T stage and texture parameters were incorporated, the Cox model showed good performance in predicting OS, with a C-index of 0.751. This finding suggested that texture analysis combining CT with contrast-enhanced T1W imaging could provide a readily available biomarker for survival stratification of patients with OOSCC.

Besides the prognostic value in OOSCC, we explored the delineation variability of texture analysis on CT and conventional MRI between observers. VOI delineation is an important step in the workflow of texture analysis and is a manual process. Recently, Pavic et al. (26) reported that the inter-observer variability of radiomics analysis was influenced strongly by the tumor type and that only 59% of the total parameters on CT had acceptable stability in head and neck squamous cell carcinoma. In the present study, texture analysis on contrast-enhanced T1W images had better delineation agreement between observers than that on CT. This observation is likely attributed to better soft-tissue contrast of the former to avoid variation in sample selection. In addition, entropy and skewness of the histogram on CT and T2W imaging had poor inter-observer agreement, a finding that correlates with a recent study of texture analysis using CT in hepatocellular carcinoma (27). These results may suggest that measurement of global heterogeneity on CT and T2W imaging was affected readily by sample bias.

The present study had four major limitations. First, our study was based on a relatively small sample and a subgroup analysis of tumor site was not allowed. Second, there is evidence that OOSCCs have biologic differences regarding the molecular alterations associated with the prognosis, such as infection with the human papilloma virus (28), which we did not take into account. Third, the possibility of type-I errors >0.05 could not be eliminated without adoption of multiple-testing correction. However, given the small sample size and exploratory purpose of our study, multiple-testing correction would not have been appropriate (29). Finally, although gray-level normalization was done before texture-parameter extraction, whether the predictive factor or model can be applied to different scanning systems and software platforms (especially the texture parameters for MRI) is not known.

In conclusion, texture analysis on contrast-enhanced T1W imaging had better delineation agreement than that on CT and T2W imaging between observers. Intra-tumor heterogeneity characterized by texture features on CT and conventional MRI was related to the OS of patients with OOSCC. As a non-invasive imaging biomarker, a combined predictive model using CT and contrast-enhanced T1W imaging texture analysis provided a valuable and practical method for survival stratification.

Supplemental Material

Supplemental material for Tumor heterogeneity in oral and oropharyngeal squamous cell carcinoma assessed by texture analysis of CT and conventional MRI: a potential marker of overall survival

Supplemental Material for Tumor heterogeneity in oral and oropharyngeal squamous cell carcinoma assessed by texture analysis of CT and conventional MRI: a potential marker of overall survival by Jiliang Ren, Ying Yuan, Yiqian Shi and Xiaofeng Tao in Acta Radiologica

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This work was supported by funds from the National Scientific Foundation of China (91859202, 81771901).

References

Jemal

Siegel

Ward

, et al. Cancer statistics, 2008. CA Cancer J Clin 2008; 58:71–96.

Zhang

Graham

Elci

, et al. Locally advanced squamous cell carcinoma of the head and neck: CT texture and histogram analysis allow independent prediction of overall survival in patients treated with induction chemotherapy. Radiology 2013; 269:801–809.

Qin

Luo

Zhu

, et al. Combination of taxanes, cisplatin and fluorouracil as induction chemotherapy for locally advanced head and neck cancer: a meta-analysis. PLoS One 2012; 7:e51526.

Seiwert

Cohen

EE.

State-of-the-art management of locally advanced head and neck cancer.

Br J Cancer 2005; 92:1341–1348.

Zhang

Mitchell

, et al. Tumor evolution and intratumor heterogeneity of an oropharyngeal squamous cell carcinoma revealed by whole-genome sequencing. Neoplasia 2013; 15:1371–1378.

Wang

Cheng

Lee

, et al. Heterogeneity of (18)F-FDG PET combined with expression of EGFR may improve the prognostic stratification of advanced oropharyngeal carcinoma. Int J Cancer 2016; 138:731–738.

Liu

Yin

, et al. Relationship between glioblastoma heterogeneity and survival time: an MR imaging texture analysis. Am J Neuroradiol 2017; 38:1695–1701.

Yip

Landau

Kozarski

, et al. Primary esophageal cancer: heterogeneity as potential prognostic biomarker in patients treated with definitive chemotherapy and radiation therapy. Radiology 2014; 270:141–148.

Miles

Ganeshan

Griffiths

, et al. Colorectal cancer: texture analysis of portal phase hepatic CT images as a potential marker of survival. Radiology 2009; 250:444–452.

10.

Kim

Park

CM.

The prognostic value of CT radiomic features for patients with pulmonary adenocarcinoma treated with EGFR tyrosine kinase inhibitors. PLOS ONE 2017; 12:e0187500.

11.

Kuno

Qureshi

Chapman

, et al. CT texture analysis potentially predicts local failure in head and neck squamous cell carcinoma treated with chemoradiotherapy. Am J Neuroradiol 2017; 38:2334–2340.

12.

Dang

Lysack

, et al. MRI texture analysis predicts p53 status in head and neck squamous cell carcinoma. Am J Neuroradiol 2015; 36:166–170.

13.

Ravanelli

Grammatica

Correlation between human papillomavirus status and quantitative MR imaging parameters including diffusion-weighted imaging and texture features in oropharyngeal carcinoma. Am J Neuroradiol 2018; 39:1878–1883.

14.

Paleri

Mehanna

Wight

RG.

TNM classification of malignant tumours 7th edition: what’s new for head and neck?

Clin Otolaryngol 2010; 35:270–272.

15.

Nioche

Orlhac

Boughdad

, et al. LIFEx: a freeware for radiomic feature calculation in multimodality imaging to accelerate advances in the characterization of tumor heterogeneity. Cancer Res 2018; 78:4786–4789.

16.

Shrout

Fleiss

JL.

Intraclass correlations: uses in assessing rater reliability.

Psychol Bull 1979; 86:420–428.

17.

Melchardt

Magnes

Hufnagl

, et al. Clonal evolution and heterogeneity in metastatic head and neck cancer-An analysis of the Austrian Study Group of Medical Tumour Therapy study group. Eur J Cancer 2018; 93:69–78.

18.

Parmar

Grossmann

Rietveld

, et al. Radiomic machine-learning classifiers for prognostic biomarkers of head and neck cancer. Front Oncol 2015; 5:272.

19.

Jansen

Gupta

, et al. Texture analysis on parametric maps derived from dynamic contrast-enhanced magnetic resonance imaging in head and neck cancer. World J Radiol 2016; 8:90–97.

20.

Vallieres

Kay-Rivest

Perrin

, et al. Radiomics strategies for risk assessment of tumour failure in head-and-neck cancer. Sci Rep 2017; 7:10117.

21.

Hayano

Tian

Kambadakone

, et al. Texture analysis of non-contrast-enhanced computed tomography for assessing angiogenesis and survival of soft tissue sarcoma. J Comput Assist Tomogr 2015; 39:607–612.

22.

Zhou

Huang

, et al. CT-based radiomics signature: a potential biomarker for preoperative prediction of early recurrence in hepatocellular carcinoma. Abdom Radiol (NY) 2017; 42:1695–1704.

23.

Bogowicz

Riesterer

Stark

, et al. Comparison of PET and CT radiomics for prediction of local tumor control in head and neck squamous cell carcinoma. Acta Oncol 2017; 56:1531–1536.

24.

Just

Improving tumour heterogeneity MRI assessment with histograms.

Br J Cancer 2014; 111:2205–2213.

25.

Haraclick

RM.

Texture features for image classification. IEEE Trans Smc 1973; 3:610–621.

26.

Pavic

Bogowicz

Wurms

, et al. Influence of inter-observer delineation variability on radiomics stability in different tumor sites. Acta Oncol 2018; 57:1070–1074.

27.

Mule

Thiefin

Costentin

, et al. Advanced hepatocellular carcinoma: pretreatment contrast-enhanced CT texture parameters as predictive biomarkers of survival in patients treated with sorafenib. Radiology 2018; 288:445–455.

28.

Nagpal

Das

BR.

Oral cancer: reviewing the present understanding of its molecular mechanism and exploring the future directions for its effective management.

Oral Oncol 2003; 39:213–221.

29.

Rothman

KJ.

No adjustments are needed for multiple comparisons.

Epidemiology 1990; 1:43–46.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.32 MB