基于人工神经网络数据挖掘技术的低位直肠癌预后模型研究

发布时间：2018-05-22 20:20

本文选题：数据挖掘 + 人工神经网络　；参考：《青岛大学》2017年硕士论文

【摘要】：目的:通过收集的包含多项预后因素的临床资料,结合以人工神经网络的数据挖掘技术,建立预测低位直肠癌患者术后5年生存状况模型,并和传统线性统计相关分析模型相较,评价其性能效果。方法:收集2009年1月~2011年8月在青大附院住院接受手术治疗的186例罹患低位直肠癌的相关患者的临床数据信息进行分析。并随机将患者分为两个集合:一组共150例,用于数据挖掘以建立疾病预后模型,为训练集;一组共36例,不参加数据挖掘,用于模型的性能评价,为测试集。此过程中进行数据挖掘的方式是使用人工神经网络(ANN)。结果:T分期、肿瘤最大径、是否有淋巴结转移、是否远处转移、手术方式、血清CEA水平、病理类型7项指标均与低位直肠癌患者术后的5年生存相关(P0.005)。人工神经网络模型(ANN)对于患者的5年生存状况预测,准确度为86.11%,敏感度为75.00%,特异度为89.29%,Logistic回归模型预测的结果显示,准确度77.78%,敏感度55.56%,特异度85.19%。总体来看ANN预测性能要好于Logistic回归分析。结论:ANN数据挖掘技术可从与低位直肠癌患者复杂的临床数据信息中寻找有意义的预后相关的因素,并借助这些因素通过建造预测模型,判别患者接受手术5年后的生存情况。
[Abstract]:Objective: to establish a 5 year survival model for low rectal cancer patients by collecting clinical data including multiple prognostic factors and combining with data mining technology of artificial neural network. Compared with the traditional linear statistical correlation analysis model, the performance of the model is evaluated. Methods: from January 2009 to August 2011, 186 patients with low rectal cancer who were hospitalized in Qingda affiliated Hospital were collected and analyzed. The patients were randomly divided into two sets: one group (150 cases) was used for data mining to establish disease prognosis model, the other group (36 cases) did not participate in data mining, which was used for model performance evaluation and test set. The method of data mining in this process is to use artificial neural network (Ann). Results the tumor size, tumor maximum diameter, lymph node metastasis, distant metastasis, operation mode, serum CEA level and pathological type were all correlated with 5-year survival of patients with low rectal cancer (P 0.005). The Ann model predicted the 5-year survival of patients with accuracy of 86.11, sensitivity of 75.00 and specificity of 89.290.The results of Logistic regression model showed that the accuracy was 77.78, the sensitivity was 55.56, and the specificity was 85.19. Overall, ANN prediction performance is better than Logistic regression analysis. Conclusion\
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R735.37

【参考文献】