Stanford AI tool lets scientists search ordinary tissue slides by typing in plain English
SpatialWhisperer predicts which cell types are where on cheap, common lab slides.
Stanford researchers have built an AI tool that lets scientists type a phrase such as T cells and see where those immune cells sit on an ordinary tissue slide, according to the Stanford Cancer Institute. The work was presented this summer at ICML, a major AI research conference.
Most tissue samples are stained with two cheap dyes, called H&E, that turn cells purple and pink. Figuring out exactly which cells are present usually takes costly tests that measure which genes are switched on. SpatialWhisperer learned the link between the two using 921,000 tissue samples with gene measurements and 1.8 million single-cell gene profiles, so it can estimate cell types from a plain slide.
The team said it beat the best existing models by 13.9% to 15.9% on three tests covering 19 tissue types. It is a research tool, not for diagnosing patients, and the code is public.
Our goal was to democratize access to pathology data interpretation across science.
Source: Stanford Cancer Institute, Nov. 3, 2026