基于CT增强图像纹理特征预测结直肠癌Ki-67表达的可行性研究

Feasibility of Predicting Ki-67 Expression in Colorectal Cancer Based on CT Enhanced Image Texture Features

目的 探讨基于CT增强图像纹理特征预测结直肠癌的可行性。方法 回顾性选取2021年2月—2022年12月济宁市兖州区人民医院病理诊断为结直肠癌的48例患者的临床资料为A组,35名健康体检人群的临床资料为B组,比较两组CT图像纹理特征与临床数据,采用二元Logistic回归筛选预测结直肠癌的独立风险因素并建立模型。对独立风险因素与二元回归预测模型进行受试者操作特征(receiver operating characteristic,ROC)曲线分析诊断效能。结果 两组患者年龄、性别、糖类抗原199、糖类抗原242、癌胚抗原比较,差异无统计学意义(P均>0.05)。使用MaZda4.7内置最小平均相关系数与最小分类误差法降为后获得5个纹理特征,二元Logistic回归方程显示游程不均匀性、紧密度、惯性为预测结直肠癌Ki-67表达的独立风险因素,建立模型为:Logit(P)=1.433+Run Length游程不均匀性×4.114+紧密度×3.015+惯性×2.356,ROC曲线下面积为0.944,取阈值31.804的灵敏度及特异度为91.7%、88.9%。结论 基于CT增强图像纹理特征预测结直肠癌,可为评估结直肠癌增殖程度和治疗预后提供无创性检测手段。

Objective To investigate the feasibility of the prediction of colorectal cancer based on the texture features of enhanced CT images.Methods The clinical data of forty-eight patients with pathological diagnosis 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 Nonuniformity,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 features of CT enhanced images can provide a non-invasive method for evaluating the degree of proliferation and prognosis of colorectal cancer.