Looking at tissue under a microscope to identify disease takes on a whole new life with an AI system that comes up with its own research ideas.

The system, called SPARK, behaves like a tiny pathology lab that works at lightning speed and has the ability to decipher patterns linked to cancer risk and treatment response.

The study, published in Nature Medicine, introduces the system using more than 5,400 cancer cases. SPARK scanned routine tissue images from cancers including lung, breast and colorectal tumors and generated and tested hundreds of biologically inspired ideas of its own.

The system found hidden tissue patterns related to essential cancer biomarkers like PD-L1 and MSI, along with clues tied to patient outcomes. Notably, some of the strongest signals came not from the tumor itself, but from nearby immune cells and support cells that were “hanging out” around it.

Researchers even let pathologists type in questions in plain English, which SPARK turned into measurable image analyses — almost like having an AI research assistant.

The tool is still experimental, but scientists say it could lead to a future where AI doesn’t just identify diseases — it helps make discoveries too.

More like this: Cancer can run, but it can no longer hide/a>