基于CT定量比较两种算法对慢性阻塞性肺疾病危重程度的分级
Comparison of two algorithms for severity classification of chronic obstructive pulmonary disease based on quantitative CT

摘要：目的 基于CT定量分析比较C5.0决策树模型和TAN贝叶斯网络模型对慢性阻塞性肺疾病(chronic obstructive pulmonary disease, COPD)危重程度分类预测的正确率。方法 回顾性收集2015年3月至2017年9月“数字肺”数据库中心COPD患者的CT扫描图像与肺功能测试结果，按《2018年慢性肺疾病诊断、治疗及预防全球策略》诊断标准，将患者分为4级。利用“数字肺”自动分析平台得到所有患者肺实质及支气管的相关指标。以肺功能分级为参照，建立C5.0决策树模型和TAN贝叶斯网络模型，比较2个模型对COPD分级的正确率。结果 C5.0的决策树模型训练样本和测试样本的正确率分别为90.76%和63.63%，TAN贝叶斯网络模型训练样本和测试样本的正确率分别为83.19%和52.73%。 结论 基于CT定量分析，应用C5.0决策树模型能更好地预测COPD疾病的危重程度。
ABSTRACT: Objective To compare the ability of C5.0 decision tree model and Tree-Augmented Navie (TAN) Bayesian Network model in the severity classification of chronic obstructive pulmonary disease (COPD) based on CT quantitative parameters. Methods CT scan images and the results of pulmonary function test (PFT) of COPD patients were retrospectively collected from the "Digital Lung" database from March 2015 to September 2017. All the patients were divided into four grades according to The 2018 Global Strategy for the Diagnosis, Management and Prevention of Chronic Obstructive Pulmonary Disease. Related parameters of lung parenchyma and bronchi of all the patients were obtained using the "Digital Lung" automatic analysis platform. With PFT as the reference, the C5.0 decision tree model and TAN Bayesian network model were established, and the accuracy of the two models was compared. Results The accuracy of training and testing samples of C5.0 decision tree reached 90.76% and 63.63%, and of TAN Bayesian network reached 83.19% and 52.73%, respectively. Conclusion The C5.0 decision tree is a better predictive model in the severity classification of COPD based on quantitative CT