基于机器学习的结直肠癌血清标志物筛选及早期诊断模型评估

ML-based Colorectal Cancer Serum Tumor Marker Screening and Evaluation of Early Stage Diagnostic Models

目的提高结直肠癌患者的早期诊断率,帮助结直肠癌患者及早发现病情获得最佳治疗效果(治疗早期直肠癌能达到超过90%的五年存活率)。方法在机器学习理论和实践的基础上,提出了采用向前法作逐步逻辑回归(Logistic Regression,LR)分析筛选出最具有诊断参考性的血清标志物,并利用支持向量机(Support Vector Machine,SVM)与后向传播(Back Propagation,BP)神经网络等模型建立结直肠癌早期诊断模型的方法。结果实验结果显示CEA、CA1724、CA242、CA153和HSP60这5种肿瘤标志物对结直肠癌均有一定的诊断价值,该五种肿瘤标志物LR模型联合检测效果明显高于五种肿瘤标志物任一指标。结论联合检测有助于提高结直肠癌检测的灵敏度,而且基于LR建立的结直肠癌检测模型相较于基于SVM建立的模型具有更高的诊断价值。

Objective To improve the early diagnostic rate of colorectal cancer in order to have eolorectal cancer patients diagnosed as early as possible for the best treatment result. Methods Based on the Machine Learning theory and practices, the forward method was adopted inlogistic regression（LR） in order to analyze and screen the serum marker of the best referential val- ue. Support Vector Machine （SVM）and Back Propagation （BP）were used to establish an early diagnostic model for colorectal cancer. Results The experiment results showed that such serum markers as CEA,CA1724 ,CA242,CA153 and HSP60 were of di- agnostic valuefor colorectal cancer. The LR eolorectal cancer diagnostic model that combined the above mentioned five types of serum markers outperformed that of each individual serum marker. Conclusion Combined tests can enhance the detection sensi- tivity to eolorectal cancer. The LR coloreetal cancer diagnostic model has a higher diagnostic value than the model based on SVM.