Histopathology Foundation Model

Histopathology foundation models (FMs) are large, pre-trained deep learning models designed to analyze digital pathology images, aiming to improve the accuracy and efficiency of disease diagnosis and biomarker prediction. Current research focuses on developing and benchmarking these models, often using Vision Transformers (ViTs) and exploring various self-supervised learning techniques like DINOv2, with a strong emphasis on addressing data biases and improving generalization across different institutions and datasets. These models show promise for improving diagnostic accuracy in various cancers, particularly when combined with multiple instance learning (MIL) strategies, and are increasingly being used for tasks such as cancer subtyping, treatment response prediction, and even spatial transcriptome prediction.