The COVID-19 pandemic not only made an impact on the discipline of radiology as a whole but also on how we use specific tools in its detection. This was especially seen in the role of chest radiography when it was utilized as a diagnostic tool at the beginning of the pandemic “when microbiological resources were scarce,” evolving into its use focusing on the detection and monitoring of COVID-19 lung involvement. However, the sensitivity of a chest radiograph in the diagnosis of COVID-19 is moderate. Therefore, in this study, the authors compared the performance of a deep learning algorithm (Ensemble4Covid) on the first clinical encounter with a group of radiologists with varying years of experience. The authors were able to show that the performance of Ensemble4Covid from the onset of the disease was considerably higher compared to the radiologists.
Key points
- The sensitivity of COVID-19 detection using conventional CXR depends on the evolution of the disease.
- Compared to a group of radiologists, AI achieves a higher sensitivity at the onset of the disease.
- This ability for early detection of COVID-19 could be applied in settings lacking resources, providing an immediate radiological report for each patient studied with CXR. Assistance with this tool could also be supplied by teleradiology in places where there is a lack of microbiological testing.
Article: A comparison of Covid-19 early detection between convolutional neural networks and radiologists
Authors: Alberto Albiol, Francisco Albiol, Roberto Paredes, Juana María Plasencia-Martínez, Ana Blanco Barrio, José M. García Santos, Salvador Tortajada, Victoria M. González Montaño, Clara E. Rodríguez Godoy, Saray Fernández Gómez, Elena Oliver-Garcia, María de la Iglesia Vayá, Francisca L. Márquez Pérez & Juan I. Rayo Madrid
