Original ResearchFree Access
Performance of a Chest Radiograph AI Diagnostic Tool for COVID-19: A Prospective Observational Study
Published Online:Jun 1 2022https://doi.org/10.1148/ryai.210217
Supplemental Material
© RSNA, 2022
APPENDIX AND TABLES
SUPPLEMENTAL FIGURES
Figure E1: Schematic of COVID-19 Diagnostic Model Implementation. ECSA = Epic client systems administrator, DMZ = demilitarized zone, PACS = picture archiving and communication system.
Figure E2: COVID-19 CXR Diagnostic Model Temporal Validation. To investigate how AUC and AUPRC vary in respect to prevalence, each figure represents a differing prevalence ratio of COVID-19 cases. As there were more negative controls available than positive controls, each colored line represents one of ten random subsampling (of the negative controls) result. Mean AUC and AUPRC is displayed with each figure. CXR = chest radiograph, ROC = receiver operating characteristic, PRC = precision-recall curve.
Figure E3: Distribution of COVID-19 Diagnostic Scores (X-axis) for participants with PCR confirmed positive COVID-19 (purple bars) and non-COVID-19 (green bars) during the month of July 2020, prevalence 25.4% (1777/7005). PCR = polymerase chain reaction.
Figure E4: External validation COVID-19 Diagnostic AI Scores for COVID-19 positive and negative participants. Box-and-whiskers plot of (A) COVID-19 Diagnostic Scores (y-axis) for non-COVID-19 versus PCR confirmed COVID-19 from 10,002 CXR from Indiana University, and (B) box-and-whiskers plot of COVID-19 Diagnostic Scores (y-axis) for non-COVID-19 versus PCR confirmed COVID-19 from 2,002 CXR from Emory University. Boxes represent the interquartile range (25%–75%) with the median denoted by the horizontal line within each box. AI = artificial intelligence, PCR = polymerase chain reaction.
