使用胸片检测新型冠状肺炎的深度集成网络

Deep Integrated Network for Detection of COVID-19 Using Chest Radiographs

随着新型冠状病毒的爆发,该病毒引起的疫情已成为全球医疗体系最大的威胁之一。由于目前新型冠状肺炎的强传染性,导致感染人群较多,同时肺炎也成为了影响新型冠状病毒检测的主要因素之一,因而快速诊断检测已经成为主要的挑战。胸部X光检测是一种安全、灵活、速度快和有利的检测方式。该文提出了一种融合深度可分离网络、卷积自编码器和VGG16的深度集成网络,构建了一个强有力的分类模型,用于对新型冠状肺炎的检测与分类。使用Kaggle存储库的COVID-19放射学标准数据集中的胸部X光图像进行验证。实验结果表明,该模型对于COVID-19、肺炎和正常的分类准确率为96.15%,灵敏度为98.92%,F1评分为94.92%。最后,将该模型与现有的模型进行了实验性能对比,同时结合了基于梯度的鉴别定位来区分不同类型肺炎的X光图像的异常区域。实验结果表明,提出的模型优于现有的模型,具有较好的鲁棒性,可以作为检测新型冠状肺炎的辅助工具。

With the outbreak of New Coronavirus,the epidemic caused by the virus has become one of the biggest threats to the global medical system.Due to the strong contagion of the new type of coronary pneumonia,a large number of people are infected,and pneumonia has become one of the main factors that affect the detection of New Coronavirus.Therefore,rapid diagnosis and detection has become a major challenge.Chest X-ray detection is a safe,flexible,fast and favorable detection method.We propose a deep integration network integrating deep separable network,convolutional self-encoder and VGG16,and construct a powerful classification model for the detection and classification of new coronary pneumonia.We use chest X-ray images from the COVID-19 radiology standard dataset of the Kaggle repository for validation.The experiment shows that the classification accuracy of COVID-19,pneumonia and normal is 96.15%,the sensitivity is 98.92%,and the F1 score is 94.92%.Finally,the experimental performance of the model is compared with the existing models,and the gradient based differential localization is combined to distinguish the abnormal areas of X-ray images of different types of pneumonia.The experimental results show that the proposed model is superior to the existing models with good robustness,which can be used as an auxiliary tool for the detection of new coronary pneumonia.