Application of CT Texture Analysis in Predicting Preoperative Lauren Classification of Gastric Cancer

To explore the application of CT texture analysis in predicting preoperative Lauren classification of gastric cancers. Patients of gastric cancer were retrospectively collected. All patients underwent enhanced CT abdomen scan including arterial, portal venous phases before surgery, and confirmed by pathology. A total of 108 patients, including 50 cases of intestinal -type gastric cancer, and 58 cases of diffuse-type gastric cancer; 72 cases were assigned into the training set, the remaining 36 cases into the testing set, according to a ratio of 2:1.The region of interest (ROI)were delineated manually by ITK Snap software, A.K. software was used for texture extraction. Then the LASSO regression was used to select image features and to reduce the dimensionality. A predictive mode of diagnosing the different types of Lauren classification was established based on logistic multiple regression correlation analysis, and the diagnostic efficiency of the characteristic parameters was analyzed by using the AUC. A total of 396 quantitative image features were extracted in the training set in arterial or portal venous phase. Five imaging features were selected after LASSO regression and dimensionality reduction in arterial phase, including GLCM entropy_all direction_offset4, GLCM entropy_angle135_offset7, quantile0.75, mean deviation and maximum intensity with AUC of 0.86 in the training set and 0.69 after applying it in the testing set. Six imaging features were selected after LASSO regression and dimensionality reduction in portal venous phase, including mean value, GLCM entropy_all direction_offset7, percentile95, correlation_all direction_offset1, long run low grey level emphasis_all direction_offset4 and short run low grey level emphasis_all direction_offset1 with AUC of 0.92 in the training set and 0.78 after applying it in the testing set. CT texture analysis held great potential in predicting Lauren classification of gastric cancers, and can be applied to preoperative evaluation of patients.