COVID-19 on the Chest Radiograph: A Multi-Reader Evaluation of an AI System

K. Murphy, H. Smits, A. Knoops, M. Korst, T. Samson, E. Scholten, S. Schalekamp, C. Schaefer-Prokop, R. Philipsen, A. Meijers, J. Melendez, B. van Ginneken and M. Rutten

Radiology 2020;296:E166-E172.

DOI PMID Cited by ~12

Background Chest radiography (CXR) may play an important role in triage for COVID-19, particularly in low-resource settings. Purpose To evaluate the performance of an artificial intelligence (AI) system for detection of COVID-19 pneumonia on chest radiographs. Methods An AI system (CAD4COVID-Xray) was trained on 24,678 CXR images including 1,540 used only for validation while training. The test set consisted of a set of continuously acquired CXR images (n=454) obtained in patients suspected for COVID-19 pneumonia between March 4th and April 6th 2020 in a single center (223 RT-PCR positive subjects, 231 RT-PCR negative subjects). The radiographs were independently analyzed by six readers and by the AI system. Diagnostic performance was performed by receiver operating characteristic curve analysis. Results For the test set, the mean age of the patients was 67.3 (+/-14.4) years (56% male). Using RT-PCR test results as the reference standard, the AI system correctly classified CXR images as COVID-19 pneumonia with an AUC of 0.81. The system significantly outperforms each reader (p < 0.001 using McNemar test) at their highest possible sensitivities. At their lowest sensitivities, only one reader can significantly outperform the AI system (p=0.04). Conclusions An AI system for detection of COVID-19 on chest radiographs was comparable to six independent readers.