迁移学习应用于新型冠状病毒肺炎诊断综述

Review of Transfer Learning Applied to Diagnosis of COVID-19

新型冠状病毒肺炎(COVID-19)疫情爆发以来,由于该病毒具有极强的传染性,所导致的感染人数与死亡人数持续增加。筛查疑似患者和早期诊断COVID-19是防止疫情恶化的重要措施之一。通过核酸检测和人工检查等方法在感染早期诊断出COVID-19是防止其在社会中爆发的最佳途径。然而核酸检测效率低下,仅仅依靠放射科专家诊断X射线图像和CT扫描图像存在耗时长且易出现诊断误差等问题。研究人员相继提出了基于迁移学习的计算机辅助诊断算法,可以最大程度地减少传统诊断方法所产生的问题,但目前关于迁移学习在新冠肺炎成像中的应用综述较少,因此总结和分析了当前国内外基于迁移学习技术诊断COVID-19的研究成果。针对模型类型进行分类讨论,分别从数据集来源、数据预处理方法、基于迁移学习的诊断模型、模型可视化、评价指标以及模型性能6个角度进行分析和比较。并指出了当前所面临的挑战和未来的发展方向,为今后进一步的研究工作奠定了基础。

Since the outbreak of the coronavirus disease 2019(COVID-19) epidemic, the number of infections and deaths caused by the epidemic has continued to increase due to its highly contagious nature. The best way to prevent outbreaks of COVID-19 in society is to detect it early in infection through methods such as reverse transcriptionpolymerase chain reaction(RT-PCR) test and manual examination. However, such laboratory procedures are inefficient, and radiologists’ diagnosis of X-ray and CT images is time-consuming and prone to diagnostic errors.Researchers have proposed computer-aided diagnostic algorithms based on transfer learning, which can minimize the problems arising from traditional diagnostic methods. But there are few reviews on the application of transfer learning in the imaging of COVID-19. Therefore, this paper summarizes and analyzes the current research results on the diagnosis of COVID-19 based on transfer learning techniques at home and abroad. The model types are categorized and discussed, and six perspectives of dataset sources, data preprocessing methods, transfer learning based diagnostic models, model visualization, evaluation metrics, and model performance are analyzed and compared, respectively. The current challenges and future development directions are also pointed out, laying foundation for further research in the future.