基于深度学习的慢性阻塞性肺病与哮喘-慢性阻塞性肺疾病重叠分类

Deep learning-based classification of chronic obstructive pulmonary disease and asthma-chronic obstructive pulmonary disease overlap syndrome

目的:基于医疗大数据的深度学习分析算法,提出了一种具有辅助诊断价值的慢性阻塞性肺病与哮喘-慢性阻塞性肺疾病重叠的鉴别诊断方法。方法:选择COPDGene数据集,利用Fisher评分的方法进行特征选择,使用准确率和ROC曲线对深度信念网络模型和支持向量机模型构建鉴别诊断模型,进行分析和比较。结果:使用深度信念网络模型得出的与COPD和ACO分类有关的敏感特征与已知临床诊断策略具有较高的吻合度,支持向量机模型和深度信念网络模型的分类准确率分别为85.28%和93.56%,灵敏度分别为89.73%和95.21%,特异度分别为74.10%和89.29%。结论:利用深度信念网络构建的COPD与ACO自动分类模型具有较高的鉴别诊断能力,可以有效协助临床医生对患者进行诊断。

Objective To study the accessory differential diagnostic method of chronic obstructive pulmonary disease( COPD) and asthma-COPD overlap syndrome based on deep learning algorithm of medical big data. Methods The characteristics of COPDGene dataset were selected with Fisher score algorithm. A differential diagnosis model was established for COPD and asthma-COPD overlap syndrome according to the accuracy and ROC cure of deep belief network( DBN) model and supporting vector model( SVM)and was comparatively analyzed. Results The sensitive characteristics of DBN model were highly consistent with those of the existing clinical diagnostic strategies in differential diagnosis of COPD and asthma-COPD overlap syndrome. The accuracy,sensitivity and specificity of SVM and DBN model were 85.28% and 93. 56%,89. 73% and 95. 21%,74. 10%and 89. 29% respectively in differential diagnosis of COPD and asthma-COPD overlap syndrome.Conclusion The automatic classification model of COPD and asthma-COPD overlap syndrome established based on DBN model is rather effective in differential diagnosis of COPD and asthmaCOPD overlap syndrome and can effectively assist the clinicians in diagnosis of COPD and asthma-COPD overlap syndrome.