The Future of Radiology

A year ago, a group at BIMIT Harvard created a data set of chest x-rays. They tried out our labeler and it worked really well, much better than what the NIH's neck bio labeler was doing at the time. And so they were able to replicate the ChexNext study in which we're again able to show that we're at least close to the performance of radiologists. But just at the time this was going on over the past year, we had had a lot of learnings. We can still do very good training with these labels. It's just that we can't trust the ground truth when we come to evaluate it.