基于人工神经网络数据挖掘技术构建浸润性膀胱癌预后模型研究

摘要

Objectives Using the data mining techniques to analysis the various factors associated with the prognosis of patients with invasive bladder cancer,then establish survival prediction model of five years in pa-tients with invasive bladder cancer,and evaluate its results by comparing with the traditional Logistic regression analysis. Methods Data of total of 134 cases of patients with invasive bladder cancer from January 2006 to De-cember 2009 in our hospital were collected. All the cases can be divided into two groups:a group as the training sample,used to screen variables and establishment of prediction model,a total of 107 cases are involved in data mining process;The other group (a total of 27 cases)was used as a validation sample,a set of evaluation model of effect,is not involved in data mining process. Artificial neural network was used in the process of data mining technique. Results T stage,tumor diameter,whether to have lymph node metastasis,tumor(single and multi-ple),operation method and pathology classification,six indexes are all related to the five years rate of survival for the patients with invasive bladder cancer(P<0. 05). ANN model predict the accuracy rate of 5 years rate of survival is 85 . 18%,sensitivity to 57 . 14% and specific degrees for 95 . 00%,related evaluation index of Logis-tic regression model,accuracy rate 77. 78%,degree of sensitiveness 44. 44%,specificity 94. 44%. All the in-dicators of ANN are better than Logistic regression model. Conclusions Data mining techniques can excavate meaningful indicators from a lot of information associated with the prognosis of patients with invasive bladder cancer,and establish the prediction model based on these indicators to determine the survival state of the pa-tients after five years.%目的：运用人工神经网络数据挖掘技术分析与浸润性膀胱癌患者预后有关的各种因素建立预测浸润性膀胱癌患者5年生存状态的预后模型，并与传统的Logistic回归分析比较评价其效果。方法收集从2006年1月至2009年12月在我院接受诊治的134例浸润性膀胱癌患者的资料。所采用数据挖掘技术为人工神经网络（ANN）。将所有病例分为两组：一组作为训练样本，不参与数据挖掘过程，共计27例；一组用于筛选变量及建立预测模型，参与数据挖掘过程共计107例。应用Logistic回归模型的相关评价指标来比较两种方法对于评价预后模型的准确度。结果 T分期、肿瘤直径、是否有淋巴结转移、肿瘤单发及多发、手术方式、病理分级，6项指标均与浸润性膀胱癌患者的5生存状态相关（P＜0．05）。ANN模型预测患者5年生存状态的准确率为85．18％、敏感度为57．14％和特异度为95．00％，Logistic回归模型的相关评价指标，准确率77．78％、敏感度44．44％、特异度94．44％。神经网络各项指标均优于Logistic回归模型。结论数据挖掘技术可从与浸润性膀胱癌患者预后相关的大量信息中挖掘出有意义的指标，并利用这些指标建立预测模型来判断患者5年后的生存状态。