The Feasibility Study Of Radiomics About CT Image In The Diagnosis And Pathological Classification Of Gastric Cancer

Objectives To explore the easibility of radiomics about CT plain scan images in the diagnosis and pathological classification of gastric cancer.Methods 59 patients with gastric cancer confirmed by Tumor surgery and gastrointestinal surgery of Tangshan Workers' Hospital were reviewed from November2017 to March 2018.There were 14 cases of signet ring cell carcinoma,39 cases of adenocarcinoma,6 cases of adenocarcinoma in signet ring cell carcinoma and signet ring cell carcinoma in adenocarcinoma.CT plain scans were performed in all the 59 patients.The images of 59 patients with CT plain scan were studied by using the pathological results as the gold standard.First,all patients were filled with low-pressure gas before CT examination.The CT images were introduced into Huiyi Huiying Radcloud platform for manual mapping of Region of interest(including diseased tissues and some normal gastric tissues),extract 1029 features.Variance threshold,select K best and Least absolute shrinkage and selection operator algorithm methods were uesed to gradually select the optimal features.80% of the computer-generated data(47 patients in the diagnosis of gastric cancer.31 cases of adenocarcinoma and 11 cases of adenocarcinoma ring cell carcinoma,42 patients in total in pathological classification of gastric cancer)were randomly assigned to the training data group for training,20%(12 patients in the diagnosis of gastric cance.8 cases of adenocarcinoma and 3 cases of signet ring cell cancer,11 patients in total in pathological classification of gastric cancer)were assigned to the verification group to verify the training group.receiver operating characteristic curve was used to illustrate the prediction performance of radiation characteristics.Finally,the more relevant features extracted from them were analyzed statistically.Results 1 In the diagnosis of gastric cancer,25 features with differential ability were selected in the training set.AUC was above 0.9,and the sensitivity and specificity were also high.The diagnostic value of Inter-quartile Range,Energy,Run Entropy and Maximum were greater.The results were statistically significant.2 In the study of pathological types of gastric cancerr,in the training group,8 features were selected to distinguish gastric carcinoma from signet ring cell carcinoma.The AUC was above 0.9,the specificity was high and the sensitivity was relatively high.The diagnostic value of Difference Entropy and Skewness were greater.The results were statistically significant.Conclusions 1 Radiomics about CT Plain scan images can be used to differentiate gastric cancer tissue from normal tissue,The diagnostic effectiveness of Inter-quartile Range,Energy,Run Entropy and Maximum were greater which can provide a reference for the diagnosis.2 Radiomics about CT Plain scan images has reference value in the study of pathological classification of gastric cancer.The diagnostic effectiveness of Difference Entropy and Skewness were greater which can reduce the invasive examination and provide reference for the treatment of clinical gastric cancer.Figure 11;Table 7;Reference 128.