Objective To investigate the feasibility of the prediction of colorectal cancer based on the tex-ture features of enhanced CT images.Methods The clinical data of forty-eight patients with pathological diag-nosis of colorectal cancer in Yanzhou District People's Hospital of Jining from February 2021 to December 2022 were retrospectively selected as group A,and the clinical data of thirty-five healthy people were selected as group B.CT image texture features and clinical data were compared between the two groups.Binary Logistics regression was used to screen independent risk factors for colorectal cancer prediction and establish models.The receiver operating characteristic(ROC)curve was used to analyze the diagnostic efficiency of independent risk factors and binary regression prediction models.Results There was no significant difference in age,sex,carbohydrate antigen 199,carbohydrate antigen 242 and carcinoembryonic antigen between the two groups(all P＞0.05).Five texture features were obtained by using MaZda4.7 built-in minimum mean correlation coefficient and minimum classification error method.Binary logistic regression equation showed that Run Length Non-uniformity,Compactness 1 and Inertia were independent factors for predicting Ki-67 expression in colorectal cancer.The model was built as follows:Logit(P)=1.433+Run Length Non-uniformity×4.114+Compactness 1×3.015+Inertia×2.356.The area under ROC curve was 0.944,the threshold was 31.804,and the sensitivity and specificity were 91.7%and 88.9%.Conclusion The prediction of colorectal cancer based on the texture fea-tures of CT enhanced images can provide a non-invasive method for evaluating the degree of proliferation and prognosis of colorectal cancer.
关键词
CT增强图像纹理特征/结直肠癌/预测模型
Key words
CT enhanced image texture features/Colorectal cancer/Prediction model
维普
万方数据