Figure 1

Dataset Characteristics. Entire dataset comprised of Non-COVID (n = 284,904) and COVID-19 (n = 29,138) cohorts. Images in the non-COVID cohort (collected between February 2008 and December 2019) were used to pre-train all classifiers and generate Projection and Body Part classifiers, as well as the Lung and Heart Segmentation model. Images in the COVID-19 cohort (collected between March 2020 and September 2020) have undergone rigorous Quality Control process and were used to train and test the COVID-19 Ensemble.