支持向量机在预测鼻咽癌患者5年生存状态中的应用

Application of Support Vector Machine in Predicting 5-Year Survival Status of Patients with Nasopharyngeal Carcinoma

目的运用支持向量机(SVM)建立预测鼻咽癌患者5年生存状态的预测模型,并与医学领域中广泛运用的人工神经网络(ANN)模型进行比较,探索鼻咽癌预后研究的新方法。方法收集2005年1月至2007年12月医院诊治的130例鼻咽癌患者的临床资料,分为两组,一组作为训练样本,用于筛选变量及建立预测模型,计104例;一组作为验证样本,用于评价模型效果,计26例。采用单因素分析筛选建模变量,然后利用ANN及SVM建立鼻咽癌患者5年生存状态预测模型并评价其效果。结果单因素分析显示,年龄、T分期、N分期、M分期、92福州分期、卡氏生活质量评分(KPS评分)、颅底骨质破坏、颅神经损伤、咽旁间隙侵犯、确诊到放疗时间、鼻咽疗效、颈部淋巴结疗效共12项指标与鼻咽癌患者的5年生存状态相关(P<0.05)。验证组验证显示,ANN模型预测患者5年生存状态的准确率、敏感度和特异度分别为88.5%,87.5%和90.0%,而SVM模型预测患者5年生存状态的准确率、敏感度和特异度分别为96.2%,93.8%和100%。结论采用SVM预测模型能较好地判断鼻咽癌患者5年后的生存状态,为个体化地预测患者的预后提供了一种新方法,其效能优于ANN预测模型。

Objective To establish the predictive models of 5-year survival status in the patients with nasopharyngeal carcinoma by using the support vector machine（ SVM）, and to compare it with the widely applied artificial neural network（ ANN） model for exploring the new method for the research of nasopharyngeal carcinoma prognosis.Methods The cinical data in 130 patients with nasopharyngeal carcinoma admitted to our hospital from January 2005 to December 2007 were collected and divided into 2 groups,one group（ 104 cases） as the training sample for screening the variables and establishing the prediction model and the another group（ 26 cases） as confirmation sample for evaluating the model effect.The single factor analysis was adopted to screen the variables for establishing the predictive model.Then ANN and SVM were used to establish the predictive models for 5-year survival status in the patients with nasopharyngeal carcinoma.The effect was also evaluated.Results The single factor analysis showed that 12 variables,including age,T stage,N stage,M stage,92 Fuzhou stage,KPS score,destruction of skull base bone,damage of cranial nerves,invasion of parapharyngeal space,time from diagnosis to radiotherapy,treatment effect of nasopharynx and of cervical lymph node,were related with the 5-year survival status（ P 0.05）.By the evaluation of confirmation group,the accuracy,sensitivity and specificity of the ANN model were 88.5%,87.5% and 90.0% respectively,whereas which of the SVM model were 96.2%,93.8% and 100% respectively.Conclusion The model based on SVM could better predict the 5-year survival status in the patients with nasopharyngeal carcinoma,provides a new method to individually predict the prognosis.The efficacy of the SVM model is superior to that of the ANN model.