A Deep-learning Pipeline for the Diagnosis and Discrimination of Viral, Non-viral and COVID-19 Pneumonia from Chest X-ray Images
Guangyu Wang; Xiaohong Liu; Jun Shen; Chengdi Wang; Zhihuan Li; Linsen Ye; Xingwang Wu; Ting Chen; Kai Wang; Xuan Zhang; Zhongguo Zhou; Jian Yang; Ye Sang; Ruiyun Deng; Wenhua Liang; Tao Yu; Ming Gao; Jin Wang; Zehong Yang; H. Cai; Guangming Lu; Lingyan Zhang; Lei Yang; W. Xu; Winston Wang; Andrea Olvera; Ian Ziyar; Charlotte Zhang; Oulan Li; Weihua Liao; Jun Liu; Wen Chen; Wei Chen; Jichan Shi; Lianghong Zheng; Longjiang Zhang; Zhihan Yan; Xiaoguang Zou; Gigin Lin; Guiqun Cao; Laurance L Lau; Manmei Long; Yong Liang; Michael Roberts; Evis Sala; Carola‐Bibiane Schönlieb; Manson Fok; Johnson Y. N. Lau; Tao Xu; Jianxing He;
Kang Zhang; Weimin Liu;
Tianxin Lin
Summary
Guangyu Wang and his team unveiled a fully automated deep-learning system that revolutionizes the diagnosis of viral pneumonia, especially COVID-19. This AI-driven system stands out for its ability to diagnose and differentiate COVID-19 from other types of pneumonia using chest X-ray images, a critical advancement given the global impact of the pandemic.
The novelty of this research is multi-faceted. Firstly, the system showcases superior accuracy in detecting COVID-19 and determining its severity compared to traditional methods. This precision is crucial in helping doctors make more informed decisions about patient care, particularly in managing the severity of the infection. Secondly, the system’s capability to outperform experienced radiologists regarding specificity and sensitivity marks a significant milestone in AI’s role in healthcare.
Training the AI system involved a comprehensive dataset of CXR images from multiple hospitals, enhancing its ability to differentiate between various types of viral pneumonia. This broad training base is instrumental in its high accuracy and adaptability to different cases. Furthermore, the system’s robustness and explainability are notable; they provide clear insights for healthcare professionals and enable junior radiologists to perform diagnoses at a senior level.
This innovation aims to equip the medical community with a powerful tool against respiratory viral pandemics, particularly when access to expert medical opinion and laboratory testing might be limited, such as in remote areas or during peak crisis times.
While acknowledging certain limitations, such as the need for further validation in early-stage COVID-19 diagnosis and differentiation from non-focal acute respiratory distress syndrome, the researchers emphasize the system’s effectiveness in identifying COVID-19 pneumonia and distinguishing it from other forms of pneumonia. They note that its accuracy is higher in severe cases than in non-severe cases, suggesting an area for future enhancement.
G. Wang et al., “A deep-learning pipeline for the diagnosis and discrimination of viral, non-viral and COVID-19 pneumonia from chest X-ray images,” Nat Biomed Eng, vol. 5, no. 6, pp. 509–521, Apr. 2021, doi: .
