A Comparative Study Of Imaging Histology And Histopathology Of Gastric Cancer

Objective: This study is to investigate the value of six machine learning models based on arterial phase and portal vein in patients with gastric adenocarcinoma in the comparison of postoperative pathological immunohistochemistry and preoperative gastric cancer texture features.Materials and Methods: This study was a retrospective study of 254 patients with gastric adenocarcinoma who underwent CT examination from January 2017 to September 2018 and confirmed by postoperative pathology.Patient data was exported from the PACS system(Picture Archiving and Communication Systems)using the AW4.6 workstation,including arterial,portal-axis images.Two radiologists with more than five years of work experience determined the boundary range of gastric cancer tumors based on histopathological results.The ITK-snap software kept 2-3 mm from the edge of the tumor.Combined with the location of the lesions recorded after surgery,the lesions were manually drawn layer by layer.The region forms a three-dimensional ROI and extracts 396 texture parameters based on Artificial Intelligence Kit(AK)analysis software.254 patients with gastric adenocarcinoma were randomly divided into training set and validation set according to 7:3.The number of training sets and validation sets were composed according to different pathological groups.Redundant features were removed by feature selection,and 18 and 17 eigenvalues of the invasive or non-invasive arteries and portal veins were extracted respectively;Lymphatic vessels were invaded or not,and there were 18 and 15 characteristic values of arterial and portal veins;no identifiable eigenvalues were extracted from nerve invasion or invasive;4 and 5 characteristic values of Ki67 positive and negative arterial and portal phases in patients with poorly differentiated gastric cancer;5 and 4 characteristic values of Ki67 positive and negative arterial and portal phases in patients with low-medium differentiated gastric cancer;Two and four characteristic values of Ki67 positive and negative arterial and portal phases in patients with moderately differentiated gastric cancer;HER2 positive,negative artery and portal phase 1,9 characteristic values.The extracted features are based on Support Vector Machine(SVM,1.SVM: C_SVC & LINEAR 2.SVM: C_SVC & RBF 3.SVM: C_SVC & sigmoid),Logistic Regression,K Nearest Neighbor(K-NN)Naive Bayesian(NB)is trained in six machine learning models to obtain an optimal machine learning model.At the same time,accuracy,sensitivity,specificity,ROC curve,and area under the ROC curve(AUC)were obtained.Results: The optimal model of the vascular group in the six machine learnings is Logistic Regression.The AUC value is 0.707,the portal phase is 0.724,the portal phase is higher than the arterial phase,and the classification effect is good;The optimal model of lymphoid group was C_SVC&LINEAR.The arterial AUC value was 0.737,the portal phase was 0.700,and the classification effect was good;The nerve group failed to extract meaningful identification features;The differentially differentiated gastric cancer Ki67 group had poorer discriminant effects,and the AUC values were 0.563 and 0.665;The optimal model of the arterial phase in the Ki67 group of moderately and moderately differentiated gastric cancer is C_SVC&sigmoid,the AUC value is 0.702,the optimal model of portal phase is SVC&RBF,the AUC value is 0.731,the differentiation effect is good,and the portal phase is superior to the arterial phase;The AUC value of the moderately differentiated gastric cancer Ki67 group was 0.664,and the classification effect was general.The portal vein effect was good,the AUC value was 0.786,and the optimal model was C_SVC&LINEAR;There is only one feature in the HER2 group of arterial phase for identification,so there is no good machine model.The optimal model of portal phase is Logistic Regression,with an AUC value of 0.720 and good classification.Conclusion: Establishing machine learning model based on texture features of contrast-enhanced CT images has better classification effect in differentiating pathological types of gastric adenocarcinoma tissues and HER2 expression,which is expected to provide important reference value for preoperative optimization of treatment plan and evaluation of prognosis of gastric cancer patients.