Fig. E2.

Examples of success with our method when the previous severity quantification models fail. (A) Annotated severity score is 2, but other models fail to make the correct prediction (CheXNet, Cohen, PXS scores are 4, 5, 3). On the contrary, our model predicts a correct severity score while providing a severity map with high agreement. (B) Also in the severe case with a score of 5, our model makes a correct prediction of severity while other models fail (CheXNet, Cohen, PXS scores are 6, 4, 2).