The second abstract that I’m going to discuss today is from Gil Shamai and his group in Israel. And what they’ve done is take H&E analysis, or they analyze images of the H&E. That’s the hematoxylin and eosin pathology slide. And that’s the slide on which the pathologist made the original diagnosis. That is, nobody has cancer until the pathologist says so...
The second abstract that I’m going to discuss today is from Gil Shamai and his group in Israel. And what they’ve done is take H&E analysis, or they analyze images of the H&E. That’s the hematoxylin and eosin pathology slide. And that’s the slide on which the pathologist made the original diagnosis. That is, nobody has cancer until the pathologist says so. And the pathologist looks at that H&E slide, and that’s how they make their decision as to what the disease is. Well, that slide, that H&E slide contains a lot of other information as well. And it’s been harvested, and people have been doing H&E analysis and built foundation models, which is the sort of result of analysis of tens of thousands or even millions of H&E slides so that they can have a basis. And then they use that basis to further hone the value of the H&E. And this group that has used the H&E to be able to prognosticate, that is to be able to determine whether or not patients are low risk or high risk, and has used the H&E to determine whether their estrogen receptor is negative or positive, and other molecular markers that they can use the H&E instead of actually doing the molecular test. But what they find is neither one is perfect. That is, if you compare either the H&E or the molecular test to outcome, they don’t do that well. And their breakthrough this time is by using both informations, they can actually increase the likelihood of accurate prediction of outcome for whichever variable they’re looking at at this time. And so the breakthrough is that usually when you train H&Es, you needed the outcome to train the H&E. And now they would be training the H&E on the H&E and the molecular information as opposed to the outcome information.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
