基于注意力机制和迁移学习的COVID-19深度学习诊断方法

目的:结合2019新型冠状病毒（COVID-19）肺炎患者肺CT影像学特征，提出一种多级空间注意力机制（ML-SAM）下的肺CT图像自动诊断模型，探讨该模型在COVID-19辅助诊断上的价值。方法:收集目前公开的COVID-19患者肺CT数据样本，在深度迁移学习框架下引入空间注意力多级聚焦策略，将数据样本、注意力机制与深度迁移学习卷积神经网络相结合，构建可在肺CT图像上自动诊断COVID-19肺炎的融合模型。结果:本文建立的融合模型对肺CT图像具有较好的分类性能，模型对COVID-19的正确识别率可达95%，同时实现了弱监督条件下肺CT图像关键特征的有效聚焦和提取。结论:本文建立的融合模型可被放射科医生或医疗保健专业人员作为COVID-19爆发期间快速、有效筛查COVID-19病例的智能辅助工具。 

COVID-19 Deep Learning Diagnosis Method Based on Attention Mechanism and Transfer Learning

Objective: This paper proposed a lung CT image automatic diagnosis model under multi level spatial attention mechanism (ML-SAM) associated with new coronavirus (COVID-19) infection in combination with the correcting CT imaging features. Methods: The published lung CT dataset samples of COVID-19 patients were collected and utilized to construct a fusion model by incorporating the attention mechanism and transfer learning strategy into the deep network. Results: The fusion model established in this paper realizes the rapid and effective auxiliary diagnosis of COVID-19. In the test dataset, the correct recognition rate of the model for COVID-19 can reach 95%. Conclusion: The deep transfer learning model established in this paper can be used by radiologists or health care professionals as an artificial intelligence tool to quickly and accurately screen COVID-19 cases during the outbreak of COVID-19.