基于密度分布特征的深度神经网络模型诊断COVID-19的价值

目的:评价基于密度分布特征（CDD）的深度神经网络（DNN）模型对新型冠状病毒肺炎（COVID-19）的诊断价值。方法:收集42例COVID-19病例和43例非COVID-19肺炎病例。将所有患者的211份胸部CT图像分为训练集（n=128）和验证集（n=83）。参考北美放射学会发布的COVID-19相关性肺炎的CT结构化报告，构建基于CT影像特征的DNN模型（DNN-CTIF）。根据胸部CT图像上肺炎CDD建立DNN-CDD模型。采用ROC曲线分析和决策曲线分析对两种模型进行评价。结果:DNN-CTIF模型的AUC在训练集为0.927，在验证集为0.829。DNN-CDD模型的AUC在训练集为0.965，在验证集为0.929。DNN-CDD模型在验证集的AUC高于DNN-CTIF模型（P=0.047）。决策曲线分析表明在0.04~1.00概率阈值范围内，DNN-CDD模型相比DNN-CTIF模型使患者的净获益更高。结论:DNN-CTIF和DNN-CDD模型对COVID-19均具有较好的诊断性能，其中DNN-CDD模型优于DNN-CTIF模型。

Diagnostic value of deep neural network model based on characteristics of density distribution in COVID-19

Abstract: Objective To evaluate the efficacy of a deep neural network (DNN) model based on characteristics of density distribution (CDD) in diagnosing corona virus disease 2019 (COVID-19). Methods A total of 42 cases of COVID-19 and 43 cases of non-COVID-19 pneumonia were enrolled in the study. The 211 chest CT images of these patients were divided into a training set (n=128) and a validation set (n=83). Referring to the CT structured report of COVID-19-related pneumonia issued by Radiological Society of North America, the CT imaging features (CTIF) based DNN model (DNN-CTIF) was constructed. Meanwhile, the DNN-CDD model was constructed based on the pneumonia CDD in the chest CT images. ROC curve analysis and decision curve analysis were used to evaluate the diagnostic performances of the two models. Results The AUC of DNN-CTIF model and DNN-CDD model was 0.927 and 0.965 in training set. The AUC of DNN-CDD model in validation set was significantly higher than that of DNN-CTIF model (0.829 vs 0.929, P=0.047). Moreover, the decision curve analysis showed that DNN-CDD model provided more net benefit than DNN-CTIF model in the range of 0.04-1.00 probability threshold.Conclusion Both DNN-CTIF and DNN-CDD models have good diagnostic performance for COVID-19, and DNN-CDD model is superior to DNN-CTIF model.