CHIEF (Harvard, Yu Lab)

CHIEF is a weakly supervised pathology foundation model from the Yu Lab at Harvard that learns at the level of the whole slide, combining tile features with anatomical-site text embeddings so one model can be pointed at tissue from any organ. It was pretrained on 15M tiles and then on 60,530 slides, and the Nature 2024 paper validated it on 19,400 slides from 24 hospitals across 19 cancer types for cancer detection, tumour origin, genomic prediction and prognosis. It is intended for research pathologists and computational groups who want one backbone rather than a model per task. Broad external validation is its main strength; the open question is whether a research release can move into regulated clinical workflows and how its predictions fare prospectively. For a newcomer: it is a general slide-reading model checked in many hospitals, but still a research tool.