基于CT图像特征的肺腺癌预后因素分析

CT image feature-based analysis on the prognostic factors of lung adenocarcinoma

目的:提出一种基于CT图像特征的肺腺癌预后因素分析方法,旨在探究不同种类CT图像特征对肺腺癌预后的影响。方法:首先,对肺部肿瘤进行分割和特征提取;然后,使用Kaplan-Meier方法进行单因素生存分析;使用COX回归模型进行多因素生存分析,得到肺腺癌的独立预后因素。最后,利用支持向量机(SVM)建立分类器对独立预后因素的预后能力进行检验。结果:选用Lung CT-Diagnosis数据库中61例患者进行试验,单因素分析显示径向方差、边缘粗糙度、GLCM熵以及GLCM非相似性与肺腺癌患者生存率显著相关(P<0.05)。COX回归模型多因素分析发现唯有径向方差与肺腺癌患者生存显著相关(P<0.05)。SVM分类器分类结果显示径向方差能够在一定程度上对患者生存时间进行预测。结论:通过对比分析,发现径向方差、边缘粗糙度、GLCM熵、GLCM非相似性与肺腺癌预后有关;径向方差是肺腺癌的独立预后因素。通过提取分析上述图像特征,医生可以对肺腺癌患者进行更加精准的预后进而延长患者生存时间。

Objective To propose a method for the analysis of the prognostic factors of lung adenocarcinoma based on CT image features, and explore the effects of different kinds of CT image features on the prognosis of lung adenocarcinoma. Methods Firstly, the lung tumors were segmented and their features were extracted. Secondly, Kaplan-Meier method was used to perform univariate survival analysis, and a multivariate survival analysis was carried out with COX regression model to obtain independent prognostic factors. Finally, a classifier based on support vector machine was established to test the prognostic ability of independent prognosis factors. Results The data of 61 patients with lung adenocarcinoma were selected form Lung CT-Diagnosis dataset. The univariate analysis showed that several image features, including radial variance, edge roughness, GLCM entropy and GLCM non-similarity, had significant effects on the overall survival of patients with lung adenocarcinoma (P<0.05). The multivariate analysis based on COX regression model revealed that only radial variance was significantly associated with the survival of patients with lung adenocarcinoma (P<0.05).The result of classifier based on support vector machine showed that to some extent, using radial variance could predict the survival time of patients. Conclusion Four features, namely radial variance, edge roughness, GLCM entropy and GLCMnon-similarity, are proved to be associated with the prognosis of patients with lung adenocarcinoma. Moreover, radial variance is regarded as the independent prognostic factor of lung adenocarcinoma. The analysis of the above image features can provide doctors with more accurate prognosis which is helpful for prolonging the survival time of patients with lung adenocarcinoma.