As the authors rightly call out in the abstract obtaining large amounts of annotated data poses a challenge for training deep learning models for this purpose But doesn't appear that the data they've collected and annotated is made available from my read of the paper, I get that this is from a company (DeepHealth) but it seems like an opportunity for NIH to push for more broadly available data sets.

Anyone have a good reference point for the reader selection of 5 specialists with 5.6yrs avg experience? That population seems small. Another opportunity for licensing bodies or national institutions to grow a publicly available dataset -- including annotations from a wider selection of imaging specialists.