Computed tomography radiomics to predict EBER positivity in Epstein–Barr virus-associated gastric adenocarcinomas: a retrospective study

Abstract

Background

The relevance of Epstein–Barr virus (EBV) in gastric carcinoma has been represented by the existence of EBV-encoded small RNA (EBER) in the tumor cells and has prognostic significance in gastric cancer, while gastric adenocarcinoma represents the most frequently occurring gastric malignancy.

Purpose

To observe the capacity of radiomic features extracted from contrast-enhanced computed tomography (CE-CT) images to differentiate EBER-positive gastric adenocarcinoma from EBER-negative ones.

Material and Methods

A total of 54 patients with gastric adenocarcinoma (EBER-positive: 27, EBER-negative: 27) were retrospectively examined. Radiomic imaging features were extracted from all regions of interest (ROI) delineated by two experienced radiologists on late arterial phase CT images. We distinguished related radiomic features through the two-tailed t test and applied them to construct a decision tree model to evaluate whether EBER in situ hybridization positive had appeared.

Results

Nine radiomics features were significantly related to EBER in situ hybridization status (P < 0.05), four of which were used to build the decision tree through backward elimination: Correlation_ AllDirection_offset7, Correlation_ angle135_offset7, RunLengthNonuniformity_ AllDirection_offset1_SD, and HighGreyLevelRunEmphasis_ AllDiretion_offset1_SD. The decision tree model consisted of seven decision nodes and six terminal nodes, three of which demonstrated positive EBER in situ hybridization. The specificity, sensitivity, and accuracy of the model were 84%, 80%, and 81.7%, respectively. The area under the curve of the decision tree model was 0.87.

Conclusion

Radiomics based on CE-CT could be applied to predict EBER in situ hybridization status preoperatively in patients with gastric adenocarcinoma.

Keywords

Computed tomography radiomics gastric adenocarcinoma

Introduction

Gastric carcinoma is one of the most common gastrointestinal malignancies around the globe (1). In 2015, gastric cancer ranked ninth for cancer incidence and eighth for cancer deaths in the United Kingdom (2). Since 1997, Epstein–Barr virus (EBV) has been regarded as causally related to gastric carcinomas in a small portion of patients (3 –5). Studies have found that EBV-associated gastric carcinoma is related to improved survival and can be efficiently treated with chemotherapy (6). EBV-encoded RNA (EBER) is the expression product of EBV, which exists in high nuclear numbers in nuclei. It is applied as an accurate supplementary meant to diagnose EBV positive malignancies histopathologically (16). Since gastric adenocarcinoma represents the most frequently occurring gastric carcinoma, the diagnosis of EBV-positive gastric adenocarcinoma is of great significance, both for treatment decision and prognosis of gastric cancers. However, the identification of EBER in situ hybridization is usually possible through postoperative testing of biopsy tissues or surgical specimens, while the opportunity to confirm EBER status preoperatively is limited. Thus, the modality to confirm EBER in situ hybridization status preoperatively is of great significance for efficient and precise treatment of gastric adenocarcinoma.

Multidetector computed tomography (MDCT) is shown to be a successful tool for preoperative staging work as well as detection of distant nodal metastasis for gastric carcinoma. It has fine spatial resolution and can quantify biological and physical characteristics of the microenvironment that attenuates X-ray (7 –10).

Radiomics, which can be applied to different imaging modalities including ultrasound, positron emission tomography, radiography, CT, and magnetic resonance imaging (MRI), is a non-invasive technique that involves extraction of a large quantity of high-throughput imaging features. These features quantify tumoral heterogeneity by measuring spatial variation in gray-level intensities in imaging information. Radiomics on CT has demonstrated promising result on the prediction of pathological staging, therapy response, and occult peritoneal carcinomatosis of gastric carcinoma (11,14,15). However, there is still no study on the application of radiomics from CT as a modality to predict EBER in situ hybridization in gastric adenocarcinoma. Therefore, the aim of the present study was to examine the potential of CT radiomics to identify EBER status of gastric adenocarcinoma.

Material and Methods

Study population

The Clinical Research Ethics Committee approved the study and the requirement for written informed consent was waived. A total of 83 patients diagnosed pathologically with EBV-positive gastric adenocarcinoma from January 2015 to December 2018 and 106 patients diagnosed pathologically with EBV-negative gastric adenocarcinoma from January 2015 to December 2015 are included in the study. The inclusion criteria were as follows: (i) pathological diagnosis of gastric adenocarcinomas by gastroscopic biopsy or surgery; (ii) pathologically confirmed with positive EBER; (iii) available contrast-enhanced CT (CE-CT) examination before therapy; (iv) patient did not undergo any preoperative chemotherapy or surgery before the examination. The exclusion criteria were as follows: (i) lesions without absolute borders for region of interest (ROI); and (ii) poor CT image quality due to artifact. Finally, a total of 54 patients (27 patients with EBV-positive adenocarcinoma and 27 patients with EBV-negative adenocarcinoma; age range = 24–75 years; mean age = 53 years) were included in the study. The flow chart of this study is shown in Fig. 1. The clinicopathological characteristics of the study cohort are shown in Table 1.

Fig. 1.

Flow of patients through the study.

Table 1.

Demographic and clinical characteristics of the study population (n = 54).

	EBV (+) (n = 27)	EBV (–) (n = 27)	P value
Sex (M/F)	23/4	19/8	<0.05
Age (years)*	51 (24–75)	55 (37–74)	<0.05
T staging (T1/T2/T3/T4)	4/5/17/1	4/3/20/0	<0.05
N staging (N0/N1/N2/N3)	8/5/13/1	2/11/14/0	<0.05
Metastatic status (M0/M1)	26/1	27/0	0.327
Major location (cardia and fundus/greater curvature of corpus/lesser curvature of corpus/angular notch/pylorus)	6/6/3/11/1	4/1/4/1/17	<0.05
Differentiation degree (poor/poor–moderate/moderate/well)	15/9/3/0	15/6/6/0	<0.05
Lauren classification (diffused/mixed/intestinal)	12/8/7	12/7/8	<0.05
Vascular invasion (Yes/No)	18/9	13/14	<0.05
Neural invasion (Yes/No)	20/7	15/12	<0.05

*Values are given as median (range).

EBV, Epstein–Barr virus.

CT acquisition

An informed consent was signed by each patient for CT examination, and patients were required to fast from solid food for at least 2 h before the examination and were instructed to drink 500 mL of water to reach gastric distension. Two 64-section and one 128-section spiral MDCT scanners were used. The patients were instructed to hold their breath during the scan. All patients were in the supine position while the scan covered the entire or upper abdomen. After the unenhanced scan, a high-pressure syringe (Medrad Stellant CT; Bayer HealthCare LLC, Whippany, NJ, USA) was used to inject 100–120 mL of iodinated contrast medium (Optiray 320 mg iodine/mL; Jiangsu Hengrui Medicine, Lianyungang, Jiangsu, PR China) sampler system at a velocity of 2.5 mL/s; late arterial phase with a post-injection delay of 35–40 s. The parameters of the CT included in this study were: tube current = 250–300 mA; tube voltage = 120 kVp; slice interval = 5 mm; slice thickness = 5 mm; matrix = 512 × 512; field of view = 36–42 cm; rotation time = 0.5–0.8 s; pitch = 0.844–0.993. CT acquisition was performed in the form of a spiral dataset; meanwhile, imaging review was performed using 5-mm continuous axial reconstruction.

Pathological evaluation

In accordance with the World Health Organization (WHO) categorization of tumors of the Digestive System (2010 edition), tumor differentiation through endoscopic biopsy was evaluated as one of the preoperative indicators in our retrospective study. The EBER in situ hybridization status was confirmed through needle biopsy. Among the 54 patients with gastric adenocarcinomas, 52 patients received radical gastrectomy (whole or partial) and the remaining two received palliative resection. Every patient had one identified lesion. A pathologist with 30 years of experience in gastrointestinal pathology retrospectively analyzed the postoperative histopathological specimens. Lauren classification, the level of differentiation, and neural as well as vascular invasion of every tumor were assessed and noted down according to the WHO categorization. The results of the gastric carcinomas were as follows: well differentiated, n = 0; moderately differentiated, n = 8; poorly to moderately differentiated, n = 16; and poorly differentiated, n = 30 poorly differentiated. There were 24 diffuse, 15 intestinal, and 15 mixed types in accordance with the Lauren categorization; there were 35 and 19 gastric adenocarcinoma cases with and without neural invasion, respectively; and there were 31 and 23 gastric adenocarcinoma cases with and without vascular invasion, respectively. The average tumor size in the training and validation cohorts was 38 cm³ (range = 7–73 cm³) and 22 cm³ (range = 9–55 cm³). Furthermore, the amount of gastric cancer with lymphoid stroma (GCLS) was also noted down as another primary pathological characteristic of EBV gastric carcinoma (27). Generally, there were 19 lesions with a small amount of GCLS, 17 lesions with a medium amount of GCLS, and 18 lesions with a high amount of GCLS. The pathology images of GCLS are demonstrated in Fig. 2, and the pathological characteristics of EBER are shown in Fig. 3.

Fig. 2.

Example of gastric cancer with lymphoid stroma (white arrow) and cancer cells (black arrow) under 200× microscope.

Fig. 3.

Comparison of EBER-negative gastric adenocarcinoma cell (black arrow) and EBER-positive gastric adenocarcinoma cell (white arrow) under 400× microscope.

Feature selection

Two radiologists segmented the ROI from late arterial CT images independently. They were blinded to the pathological and clinical information of the patients including age, gender, name, TNM stage (tumor, lymph node, and metastasis), degree of tumor differentiation, histological type, neural and vascular invasion of the lesion, and Lauren classification, as this is the effective phase for identifying the mucosal enhancement for early gastric cancers (Fig. 4) (26).

Fig. 4.

Example of segmentation of gastric adenocarcinoma in late arterial phase images. (a) Original image. (b, c) Segmentation by two experienced radiologists. (d) Ground truth tumor image showing the area of overlap between the two readers.

Late arterial phase CE-CT images for all patients were derived from the picture archiving and communication system (PACS), after which the images were uploaded into in-house software (Artificial Intelligence Kit; GE, Boston, MA, USA). The two radiologists manually segmented a polygonal ROI along the edge of the tumor on each slice the lesion involved. The gastric cavity and artifacts were carefully avoided when drawing the ROIs.

Feature extraction was performed on all ROIs in late arterial phase CT images of the 54 patients. The final tumor area was determined by the overlapping regions of two ROIs that were segmented separately by the two radiologists. Areas with uncertainty were not included in the delineation.

Through manual delineation by two radiologists, a total of 385 radiomic features were extracted from the tumor region. The feature extraction was processed using in-house software (Artificial Intelligence Kit; GE). Quantitative radiomic features were separated into three subcategories: ﬁrst-order statistical features; textural features; and shape- and size-based features. First-order statistical features demonstrate distribution of voxel intensities within the CT images through basic and commonly used indicators (e.g. entropy, kurtosis, energy). Textural features were calculated using gray-level co-occurrence (GLCM) and gray-level run-length texture matrices (GLRLM). Gray-level co-occurrence (GLCM) is an effective texture feature type that utilizes second-order statistics to distinguish two adjacent pixel values at specific sites. The GLCM features applied in our research were energy, entropy, correlation, inertia, inverse different moment, cluster shade, cluster prominence, and Haralick correlation (22 –24). GLRLM is another texture feature type that can extract the spatial plane features of each pixel. In our study, 10 GLRLM features were selected, i.e. long run emphasis, short run emphasis, run length non-uniformity, gray-level non-uniformity, high gray-level run emphasis, low gray-level run emphasis, short run low gray-level emphasis, long run low gray level emphasis, short run high gray level emphasis, long run high gray level emphasis (25,26). Features based on shape and size provide an indication of the tumor’s compactness, volume, and area, as well as the tumor’s spherical, circularity, or slenderness (17).

The ﬁrst-order statistical, textural, and shape- and size-based features account for 42, 324, and 19 features, respectively.

Statistical analysis

The Kolmogorov–Smirnov normality test was applied to evaluate the distribution characteristics of the CT radiomics parameters of gastric adenocarcinomas. Depending on the normality test results, univariate analysis was performed by the Mann–Whitney U test for continuous characteristics and Pearson’s chi-square test or Fisher’s exact test for categorical characteristics. Statistical power and 95% confidence intervals (CI) of the difference of statistically significant CT texture parameters in univariate analysis were calculated. The selected features were applied to construct a discriminant model through the decision tree method to predict the presence of EBER in situ hybridization. The model was constructed through the classification and regression tree method, which is based on empirical statistical techniques for recursive partition analysis and generate binary decision trees. With decision tree analysis, diagnostic threshold values were iteratively decided to maximize the accuracy of diagnosis while minimizing false-positive EBER in situ hybridization. Unless the Gini index does not improve by > 0.001, the tree does not expand its node. In order to limit overfitting, the minimum number of cases for the node is set to nine parent nodes and three child nodes. Therefore, if the output branch of the decision node contains three or fewer patients in any diagnostic category, no further separations of the tree were required. We assessed the efficacy of the final decision tree through receiver operating characteristic (ROC) curve and specificity, sensitivity, and accuracy values. Statistical analysis was processed using SPSS (IBM Corp., Armonk, NY, USA). P < 0.05 was regarded as statistically significant.

Results

The volumes of the ROIs were calculated as VolumeCC in the shape- and size-based feature. The mean volume of ROI in the training cohort and validation cohorts was 35 cm³ (range = 5–68 cm³) and 26 cm³ (range = 4–60 cm³), respectively.

Nine radiomics features were found significantly related with EBER in situ hybridization status (P < 0.05). Four features were selected from the nine features through backward selection to build the decision tree according to the classification regression tree. The selected features are listed in Table 2, comparing the distribution of the four features according to EBER in situ hybridization status. Statistical significance was identified in four radiomic features between EBER-positive patients and EBER-negative patients (P < 0.05). The Correlation_AllDirection_offset7 (0.00109 ± 0.00098; P = 0.04859), Correlation_ angle135_offset7 (0.00085 ± 0.00092; P = 0.04418), RunLengthNonuniformity_AllDirection_offset1_SD (375.79 ± 457.97; P = 0.02111), HighGreyLevel RunEmphasis_AllDiretion_offset1_SD (8.11 ± 7.39; P = 0.04661) had a significant ability to differentiate positive EBER in situ hybridization from negative EBER in situ hybridization (0.00064 ± 0.00064, 0.00043 ± 0.00052, 147.77 ± 172.72, and 4.93 ± 3.03). The box plot in Fig. 5 shows the differences in the four imaging features.

Table 2.

Comparison of tumor radiomic features parameters between EBV-positive and EBV-negative patients.

	EBV (+) (n = 27)	EBV (–) (n = 27)	P value
InverseDifferenceMoment_ AllDirection_offset1_SD	0.0155 ± 0.00831	0.01039 ± 0.00563	0.01119
Correlation_angle90_offset4	0.00232 ± 0.00206	0.00138 ± 0.00113	0.04220
Correlation_angle135_offset4	0.00169 ± 0.00143	0.00097 ± 0.00098	0.03429
Correlation_AllDirection_offset7	0.00109 ± 0.00098	0.00064 ± 0.00064	0.04859
Correlation_angle90_offset7	0.00142 ± 0.00147	0.00073 ± 0.00063	0.03131
Correlation_angle135_offset7	0.00085 ± 0.00092	0.00043 ± 0.00052	0.04418
RunLengthNonuniformity_ AllDirection_offset1_SD	375.78688 ± 457.97291	147.76973 ± 172.72332	0.02111
HighGreyLevelRunEmphasis_ AllDirection_offset1_SD	8.10504 ± 7.38854	4.93308 ± 3.03053	0.04661
Maximum3DDiameter	97.51083 ± 36.34577	80.01221 ± 22.09363	0.03726

Values are given as mean ± SD.

EBV, Epstein–Barr virus.

Fig. 5.

A box plot comparing the distribution of EBV-positive gastric carcinoma with that of EBV-negative gastric carcinoma. (a) HighGreyLevelRunEmphasis_ AllDiretion_offset1_SD (375.79 ± 457.97; P = 0.02111). (b) Correlation_angle135_offset7 (0.00085 ± 0.00092; P = 0.04418). (c) RunLengthNonuniformity_ AllDirection_offset1_SD (375.79 ± 457.97; P = 0.02111). (d) Correlation_ AllDirection_offset7 (0.00109 ± 0.00098; P = 0.04859). The central line in the box plot indicates the median value of the data. The lower and upper boundary lines of the central box represent the 25% and 75% quartiles. The box indicates the 95% confidence interval.

The final decision tree consisted of seven nodes (Fig. 6). The first node of the decision tree used the Correlation_angle135_offset7 with a threshold value of 0.000585785. Second, the Correlation_angle135_ offset7 and HighGreyLevelRunEmphasis_ AllDirection_offset1_SD were evaluated. Third, the RunLengthNonuniformity_AllDirection_offset1_SD and Correlation_ AllDirection_offset7 were assessed. Finally, the RunLengthNonuniformity_AllDirection_ offset1_SD and Correlation_ angle135_offset7 were assessed. Therefore, the final decision tree consisted of seven decision nodes and six terminal nodes, three of which distinguished positive EBER in situ hybridization and the remaining three identified negative EBER in situ hybridization. The area under the curve (AUC) was 0.87 (95% CI = 0.766–0.975; P < 0.001), showing the model had a relatively high efficacy in differentiating EBER-positive gastric adenocarcinomas from EBER-negative ones. The proposed model demonstrated a fine specificity, sensitivity, and accuracy, which were 85%, 89%, and 87%, respectively. The ROC curve is shown in Fig. 7.

Fig. 6.

Decision tree for identification of EBER in situ hybridization status in patients with gastric adenocarcinoma using CT imaging features of late arterial phase.

Fig. 7.

Receiver operating characteristic curve showing the performance of the decision tree model.

Discussion

In the present study, the decision tree analysis was used for the prediction of EBER in situ hybridization status through late arterial phase of CE-CT imaging features to determine the therapeutic strategy and predict the prognosis of gastric adenocarcinoma. Formerly, invasive biopsy or surgeries are required to confirm the status of EBER in situ hybridization in clinical procedures. CE-CT imaging-based radiomics could be utilized as a non-invasive diagnostic modality to overcome these limitations.

Several studies attempted to study CT radiomics characteristics as a biomarker for gastric cancer diagnosis as well as prognosis. Giganti et al. (14) found significant associations between pretreatment CT-based texture analysis and treatment response in gastric carcinoma, with a specificity, sensitivity, and accuracy of 71%, 75%, 74% for entropy, and a specificity sensitivity, and accuracy of 73%, 64%, and 67% for volume. Kim et al. (15) show that preoperative CT texture features can be used as a predictive tool for occult peritoneal carcinomatosis in patients with advanced gastric cancer. The specificity, sensitivity, and AUC of that study were 80.5%, 75.6%, and 0.77, respectively. Liu et al. (12) found CT texture analysis useful for identifying pathological and histological features of gastric cancers; the specificity, sensitivity, and accuracy were 83.3%, 78.7%, and 81.3%, respectively, with an AUC of 0.774. Similarly, Liu et al. (11) reported CT texture analysis of gastric carcinoma demonstrated significant relationships with TNM staging. However, to our knowledge, this is the first study on the relationship between CE-CT radiomics features and EBER in situ hybridization of gastric adenocarcinoma. CT imaging holds the advantages of fewer artifacts and higher spatial resolution, which is suitable for clinical staging and texture feature extraction compared with MRI on the diagnosis of gastric carcinoma. Previous studies have shown different clinical treatment and prognosis of EBER-positive gastric cancer from EBER-negative patients, while still there is no study to determine the relationship between CT imaging characteristics and EBER in situ hybridization status. Therefore, in this preliminary study, the decision tree model built through CT radiomics features can be applied as a non-invasive tool to support clinical decisions. Moreover, this study selected a total of three commercial models of CT machines commonly used in clinical practice. Therefore, the results of the study are more robust, with greater reference for future transformation in clinical practice.

For the prediction of EBER in situ hybridization status, we used a decision tree model, which is able to efficiently handle large, complex datasets without imposing complex parameter structures. The decision tree model is both easy to understand and perform. By applying decision tree analysis based on CT radiomics, we were able to identify the EBER in situ hybridization status simply through related imaging features. The specificity, sensitivity, and accuracy of the decision tree to differentiate EBER-positive and EBER-negative gastric adenocarcinomas are 85%, 89%, and 87%, respectively. Compared with previous studies, the model applied in this study performed better as a biomarker for gastric cancer.

The present study has some limitations. First, prospective studies are needed for patients whose data are not used to build models in the future, and external validation of the proposed models is equally required. Second, the population included in this research is relatively small, so the reliability of the constructed model requires being testified in prospective studies with more cases included. Lastly, radiomic features were merely extracted from late arterial phase CE-CT images in this study. Radiomics analysis can be improved with the involvement of other CE-CT images (such as portal phase images and delayed phase images).

In conclusion, the present study shows that late arterial phase CE-CT imaging has the potential to be a non-invasive and preoperative modality to predict the EBER in situ hybridization status in patients with gastric adenocarcinoma. The proposed model can also assist with deciding on appropriate therapeutic strategies for patients with gastric adenocarcinoma.

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) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Cheng Zhang

References

Jemal

Bray

Center

, et al. Global cancer statistics. CA Cancer J Clin 2001; 61:69–90.

Fitzmaurice

Allen

Barber

, et al. Global, regional, and national cancer incidence, mortality, years of life lost, years lived with disability, and disability-adjusted life-years for 32 cancer groups, 1990 to 2015: a systematic analysis for the global burden of disease study. JAMA Oncol 2017; 3:524–548.

Arikawa

Tokunaga

Satoh

, et al. Morphological characteristics of Epstein-Barr virus-related early gastric carcinoma: a case-control study. Pathol Int 1997; 47:360–367.

Murphy

Pfeiffer

Camargo

, et al. Meta-analysis shows that prevalence of Epstein-Barr virus-positive gastric cancer differs based on sex and anatomic location. Gastroenterology 2009; 137:824–833.

Lee

Kim

Han

, et al. Clinicopathological and molecular characteristics of Epstein-Barr virus-associated gastric carcinoma: a meta-analysis. J Gastroenterol Hepatol 2009; 24:354–365.

Camargo

Kim

Chiaravalli

, et al. Improved survival of gastric cancer with tumor Epstein–Barr virus positivity: an international pooled analysis. Gut 2014; 63:236–243.

Tsurumaru

Miyasaka

Nishimuta

, et al. Differentiation of early gastric cancer with ulceration and resectable advanced gastric cancer using multiphasic dynamic multi-detector CT. Eur Radiol 2016; 26:1330–1337.

Saito

Kurokawa

Takiguchi

, et al. Accuracy of multi-detector row CT in diagnosing lymph node metastasis in patients with gastric cancer. Eur Radiol 2015; 25:368–374.

Ganeshan

Miles

KA.

Quantifying tumor heterogeneity with CT. Cancer Imaging 2013; 13:140–149.

10.

Davnall

Yip

Ljunggvist

, et al. Assessment of tumor heterogeneity: an emerging imaging tool for clinical practice? Insights Imaging 2012; 3:573–589.

11.

Liu

Shi

, et al. Preoperative CT texture analysis of gastric cancer: correlations with postoperative TNM staging. Clin Radiol 2018; 73:756.e1–756.e9.

12.

Liu

, et al. Application of CT texture analysis in predicting histopathological characteristics of gastric cancers. Eur Radiol 2017; 27:4951–4959.

13.

Giganti

Antunes

Salerno

, et al. Gastric cancer: texture analysis from multi-detector computed tomography as a potential preoperative prognostic biomarker. Eur Radiol 2017; 27:1831–1839.

14.

Giganti

Marra

Ambrosi

, et al. Pre-treatment MDCT-based texture analysis for therapy response prediction in gastric cancer: Comparison with tumor regression grade at final histology. Eur J Radiol 2017; 90:129–137.

15.

Kim

Yun

, et al. Could texture features from preoperative CT image be used for predicting occult peritoneal carcinomatosis in patients with advanced gastric cancer? PLoS ONE 2018; 13:e0194755.

16.

Iwakiri

Takada

Role of EBERs in the pathogenesis of EBV infection.

Adv Cancer Res 2010; 107:119–136.

17.

Patil

Mahadevaiah

Dekker

. An approach toward automatic classification of tumor histopathology of non–small cell lung cancer based on radiomic features. Tomography 2016; 2:374–377.

18.

Chaddad

Desrosiers

Bouridane

, et al. Multi texture analysis of colorectal cancer continuum using multispectral imagery. PLoS One 2016; 11:e0149893.

19.

Haralick

Shanmugam

Textural features for image classification. IEEE Trans Syst Man Cybern 1973; 3:610–621.

20.

Sachdeva

Kumar

Gupta

, et al. Segmentation, feature extraction, and multiclass brain tumor classification. J Digit Imaging 2013; 26:1141–1150.

21.

Öztürk

Akdemir

Application of feature extraction and classification methods for histopathological image using GLCM, LBP, LBGLCM, GLRLM and SFTA. Procedia Comput Sci 2018; 132:40–46.

22.

Galloway

MM.

Texture analysis using gray level run lengths. Computer Graphics and Image Processing 1975; 4:172–179.

23.

Karahaliou

Skiadopoulos

Boniatis

, et al. Texture analysis of tissue surrounding microcalcifications on mammograms for breast cancer diagnosis. Br J Radiol 2007; 80:648–656.

24.

Krishnan

Sudhakar

Automatic classification of liver diseases from ultrasound images using GLRLM texture features. International Workshop Soft Computing Applications 2013; 195:611–624.

25.

Fonarow

Adams

Abraham

, et al. ADHERE Scientific Advisory Committee, Study Group, and Investigators. Risk stratification for in-hospital mortality in acutely decompensated heart failure: classification and regression tree analysis. JAMA 2005; 293:572–580.

26.

Chen

Hsu

, et al. Gastric cancer: preoperative local staging with 3D multidetector row CT-correlation with surgical and histopatological results. Radiology 2007; 242:472–482.

27.

Pyo

Kim

Kang

DW.

Clinicopathological significance of EBV-infected gastric carcinomas: a meta-analysis. Medicina 2020; 56:345.