Using AI for pan-cancer diagnosis

Pathologists, take note: If this Biorxiv preprint holds up through peer review, it's another reminder of the potential of artificial intelligence to help identify types of cancer. Moritz Gerstung at the European Bioinformatics Institute, along with several colleagues, used deep learning to analyze more than 17,000 images of tumor cells — and was able to accurately classify 28 types of cancer. 

The pan-cancer computational histopathology tool was also able to match up the cellular images with their respective genomic aberrations. And, importantly, it provided prognostic information — determining how a patient might fare, based on the composition of the cells on their respective slides. 

As Scripps seer Dr. Eric Topol points out, “computational pathology is teaching us about all the information that can be gleaned from images that humans can’t see.”