Skip to content

CHIEF

A pathology foundation model for cancer diagnosis and prognosis

Visit official site

What CHIEF is

The model is described as a pathology foundation model for cancer diagnosis and prognosis prediction that extracts imaging features with two complementary pretraining methods. Unsupervised pretraining targets tile-level features, while weakly supervised pretraining targets whole-slide patterns.

What you can do with it

  • Detect cancer cells and normal cells
  • Identify the origin site of a tumor
  • Characterize the molecular profile of a tumor
  • Predict the prognosis of cancer patients

Development and validation

  • Development used 60,530 whole-slide images across 19 anatomical sites
  • Pretraining ran on a 44TB high-resolution pathology imaging dataset
  • Validation covered 19,491 whole-slide images in 32 independent slide sets
  • The slide sets came from 24 international hospitals and cohorts

Getting started

  • Install a Linux system with an NVIDIA GPU and a Python environment
  • Clone the code repository to a local environment
  • Install the required libraries and tools from the installation guide
  • Download and install the pretrained weights

Pros & cons

✓ What we like

  • Tile-level and whole-slide features are both extracted
  • Detection, origin, molecular and prognosis tasks are all listed

! What to watch out for

  • Local setup calls for a Linux system, an NVIDIA GPU and a Python environment
  • Pretrained weights must be downloaded and installed separately

FAQ

What was used to develop the model?

The material states 60,530 whole-slide images covering 19 anatomical sites, pretrained on a 44TB dataset.

How was it validated?

It was validated on 19,491 whole-slide images from 32 independent slide sets across 24 international hospitals and cohorts.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review