Evaluation of a deep learning-based automated detection algorithm for identifying COVID-19 pneumonia on chest X-ray images

抄録

This study evaluated the diagnostic accuracy of a deep learning-based automated detection （DLAD） algorithm for coronavirus disease 2019 （COVID-19） pneumonia using chest X-ray images and the consistency of its performance across different variants of the severe acute respiratory syndrome coronavirus 2 （SARS-CoV-2）. A retrospective observational study was conducted at Showa Medical University Hospital. This study included 699 adult patients （aged ≥20 years） diagnosed with COVID-19 who underwent chest radiography and computed tomography （CT） scan between February 2020 and March 2022. Chest X-ray images were analyzed using the DLAD software developed by Fujifilm, which generated abnormality scores of 0％-100％. CT scan was used as the reference standard for diagnosing pneumonia. Diagnostic accuracy was evaluated via a receiver operating characteristic curve analysis. Subgroup analyses based on sex, age, body mass index, and SARS-CoV-2 variant were conducted. Of the 699 patients, （28.6％） had abnormal CT scan findings. The DLAD algorithm had a strong predictive performance, with an area under the curve of 0.85 （95％ confidence interval ［CI］: 0.82-0.88）, a sensitivity of 75.4％, and a specificity of 79.5％ at an optimal cutoff score of 28.5％. No significant differences were found in terms of the area under the receiver operating characteristic curve between the subgroups stratified according to sex, age, body mass index, or SARS-CoV-2 variant （p＞0.05）. The DLAD algorithm had a high diagnostic accuracy for COVID-19 pneumonia on chest radiography and a consistent performance across SARS-CoV-2 variants. These findings support the use of DLAD as a reliable tool for detecting COVID-19 pneumonia in clinical settings.