I have been doing some work in recent months with Dr. Sophia Wang (who also happens to be my wife) at Stanford applying deep learning/AI techniques to make predictions using notes written by doctors in electronic medical records (EMR).
Because of the sensitivity of the information, a lot of what we’re working on can be difficult to share. So, I put together a fun project based on public data and some of the lessons I’ve picked up from working on these projects (lessons on working with Tensorflow, Keras, BeautifulSoup, tf.data, etc.) to see if, given a paper abstract and title, you can predict if a paper is going to make it into a top-tier ophthalmology journal 😇. Surprisingly it did pretty well (87% accuracy, 91% AUROC — higher than I expected when I set out to do this!) and so I’m releasing the code on Github as well as a tutorial explaining the code to help people out there who want to try experimenting with these new powerful AI tools on something similar (but are having trouble starting due to a lack of simple documentation and tutorials for some of these features).
If you’re curious, check it out.