Figure 4
From: Assessing clinical applicability of COVID-19 detection in chest radiography with deep learning
ROC of each of the models on each of the datasets. Average model performance across all folds is shown as a line (full line: full dataset used for testing; dashed line: separate test set for each fold) and shaded region corresponds to the 95% confidence interval. Left: model performance on all CXRs; Right: model and radiologists performance on annotated CXRs. Rightmost plots are a zoomed version of the gray shaded region of the left plot. Radiologists performance is shown when considering as positives only CXRs marked as Indicative of COVID-19 (\(\bullet \)) or Indicative of COVID-19 and Undetermined (\(\blacklozenge \)).
