探讨Bayes判别分析在食管癌淋巴结转移CT诊断中的价值

摘要

Objective:To construct mechanism model for the diagnosis of LNs (Lymph nodes metastasis)of esophageal cancer by Bayes discriminatory analysis.Methods:208 thoracic esophageal cancer cases were divided into two groups,inclu-ding training set (166 cases)and test set (42 cases).Training set was used to construct mechanism model,test set was used to analyze its value of diagnosis.Results:Non-metastatic LNs:Y0 = -7.499X1 +7.957X2 +167.761X3 -0.087X4 +0.459X5-5.528X6-0.711X7-66.080;metastatic LNs:Y1= -6.697X1+8.231X2+181.686X3-0.106X4+2.219X5-3.331X6-0.124X7-83.183.X1= short-axis diameter,X2= long-axis diameter,X3= aspect ratio,X4= cross-sectional area, X5= fuzzy rim,X6= cental low density area,X7= clustered distribution.The accuracy was 87.7%,the P (the probability of false prediction)was 12.3%,with similar result by cross validation.For the test set,the accuracy of Bayesian analysis was 84.7%,the P was 15.3%.Conclusion:It is feasible to use Bayes discriminatory analysis for diagnosis of LNs metastasis of esophageal cancer.%目的：利用Bayes判别分析初步建立诊断食管癌淋巴结转移的各种CT征象的联合诊断模型。方法：共搜集胸段食管癌208例，将其分为实验组即训练样本（166例，80％）及检验组即验证样本（42例，20％），实验组用以建立诊断方程，检验组用以验证方程。结果：经计算获得的方程为：Y0＝－7．499X1＋7．957X2＋167．761X3－0．087X4＋0．459X5－5．528X6－0．711X7－66．080，Y1＝－6．697X1＋8．231X2＋181．686X3－0．106X4＋2．219X5－3．331X6－0．124X7－83．183，其中Y0为非转移组，Y1为转移组，X1为淋巴结最大短径，X2为淋巴结最大长径，X3为横纵比，X4为最大截面积，X5为是否边缘模糊，X6为是否中央低密度，X7为是否成簇分布。利用自身检验法所得诊断模型的符合率为87．7％，误判率为12．3％，交叉检验法与自身检验法所得结果相近。当利用验证样本数据代入方程，所得模型诊断符合率为84．7％，误判率为15．3％。结论：通过Bayes判别分析法所建立的不同CT征象对食管癌淋巴结转移的联合诊断模型具有一定的诊断价值，但诊断模型还有待进一步完善。