CHIEF

Outperforms state-of-the-art deep learning methods by up to 36.1% on independent international datasets, effectively handling domain shifts from diverse slide preparations and populations.

Encodes entire whole-slide images into a single 768-dimensional feature vector, simplifying downstream tasks like survival analysis and clustering without fine-tuning the aggregator.

Pretrained on 44 terabytes of data from 19 anatomical sites, providing a robust and diverse foundation for various cancer evaluation applications, as validated in the Nature paper.

Available as a Docker container on Docker Hub, facilitating easy setup and consistent inference environments, reducing deployment overhead.