Texture analysis based on CT for predicting the differentiation of esophageal squamous cancer: An observational study.

ABSTRACT

To explore the feasibility and application value of texture analysis based on computed tomography (CT) for predicting the differentiation of esophageal squamous cell carcinoma (ESCC). Patients diagnosed with ESCC who underwent chest contrast-enhanced CT before treatment were selected. Based on the pathological results, the patients were stratified into poorly differentiated and moderately well-differentiated groups. FireVoxel software was used to analyze the region of interest based on venous phase CT images. Texture parameters including the mean, median, standard deviation (SD), inhomogeneity, skewness, kurtosis, and entropy were obtained automatically. Differences in the texture parameters and their relationship with the degree of differentiation between the 2 groups were analyzed. The value of CT texture parameters in identifying poor differentiation and moderate-well differentiation of esophageal cancer was analyzed using the ROC curve. A total of 48 patients with ESCC were included, including 24 patients in the poorly differentiated group and 24 patients in the moderate-well-differentiated group. There were negative correlations between SD, inhomogeneity, entropy, and the degree of differentiation of esophageal cancer (P < .05). The correlation of inhomogeneity was the highest (r = -0.505, P < .001). SD, inhomogeneity, and entropy could effectively distinguish between the poorly and moderately well-differentiated groups, with statistically significant differences between the 2 groups (P < .05). The best critical values for SD, inhomogeneity, and entropy were 17.538, 0.017, and 3.917, respectively. The areas under the ROC curve were 0.793, 0.792, and 0.729, respectively, with the SD and inhomogeneity being the best. The application of texture analysis on venous phase CT images holds promise as a method for forecasting the degree of differentiation in esophageal cancers, which could significantly contribute to the preoperative noninvasive evaluation of tumor differentiation.