Hacker News Reader

The reason I'm skeptical of this is that there is no actual comparison to human level performance. I.E. they didn't have radiologist actually read their images to compare against the model. Notice that the title of the paper is "Predicting Breast Cancer by Applying Deep Learning to Linked Health Records and Mammograms" it's only in the press release that they seem to imply a comparison to radiologists was actually done.

View on HN

https://pubs.rsna.org/doi/full/10.1148/radiol.2016161174

This is their comparison point for actual radiologists. Citation number 6. It doesn't look comparable, though. Radiologists are around 90% specificity and sensitivity, which varies a good amount from the model's 77.3% and 87%, respectively.

This is not on this dataset though (right?), so not really a solid comparison point. Plus lik you mentioned, they seem to be doing worse than this benchmark.

Reply on news.ycombinator.com