基于数据挖掘技术大肠肿瘤高危人群发病预警研究

A study on early warning of colorectal cancer in high-risk population based on data mining technology

目的评价基于肿瘤标志物检测与APCS评分系统筛查的人工神经网络对大肠肿瘤高危人群发病风险预警的价值。方法经病理确诊大肠癌101例、大肠息肉组110例和对照组96例,进行肿瘤标志物癌胚抗原(CEA)、糖类抗原19-9(CA19-9)联合检测并接受APCS评分系统筛查,同时应用数据挖掘技术(人工神经网络技术)建立肿瘤标志物-APCS评分-人工神经网络模型(简称数据挖掘模型)进行发病风险预测。结果数据挖掘模型诊断大肠癌敏感度为81.07%、特异度为87.21%、准确度为84.69%、ROC曲线下面积为0.853;APCS评分系统筛查敏感度为86.32%、特异度为74.86%、准确度为70.05%、ROC曲线下面积为0.602;肿瘤标志物联合检测敏感度为92.38%、特异度为60.47%、准确度为67.41%、ROC曲线下面积为0.501。结论基于肿瘤标志物检测与APCS评分系统筛查的人工神经网络模型提高了对大肠癌的预测准确性。

Objective To evaluate the value of artificial neural network based on tumor marker detection and APCS score system screening for early warning of the risk of colorectal cancer in high-risk population.Methods A total of 101 cases of colorectal cancer and 110 cases of large intestine polyps were confirmed by pathology,and 96 healthy cases were selected as the control group.All cases were tested for CEA and CA19-9,and were screened by APCS scoring system.At the same time,artificial neural network was used to establish tumor marker-APCS score and predict the risk of disease.Results The sensitivity of data mining model for colorectal cancer diagnosis was 81.07%,the specificity was 87.21%,the accuracy was 84.69%,and the area under ROC curve was 0.853.The sensitivity of APCS scoring system was 86.32%,the specificity was 74.86%,the accuracy was 70.05%,and the area under ROC curve was 0.602.The sensitivity of combined detection of tumor markers was 92.38%,the specificity was 60.47%,the accuracy was 67.41%,and the area under ROC curve was 0.501.Conclusion The artificial neural network model based on tumor marker detection and APCS score system screening improves the accuracy of colorectal cancer prediction.