A Whole-Slide Foundation Model for Digital Pathology from Real-World Data

Digital pathology is entering a new era with Prov-GigaPath, a whole-slide pathology foundation model pretrained on a massive dataset from Providence health network. Imagine analyzing gigapixel slides with tens of thousands of image tiles, all seamlessly integrated to capture both local and global patterns. Prov-GigaPath leverages the power of the novel GigaPath vision transformer architecture, adapted from the cutting-edge LongNet method, to achieve this feat. With an impressive training set comprising 1.3 billion image tiles from 171,189 whole slides, this model is designed to handle the vast complexity of real-world pathology data. From cancer subtyping to prognostic predictions, Prov-GigaPath outperforms existing models across 25 out of 26 tasks, thanks to its innovative approach that combines large-scale pretraining with ultra-large-context modelling. Not just a technical marvel, Prov-GigaPath also highlights the importance of community collaboration in advancing computational pathology. By making this model open-weight, researchers worldwide can tap into its capabilities, fostering further innovations in healthcare and life sciences. Whether it's simulating the dynamics of high-energy physics experiments or revolutionizing cancer diagnostics, Prov-GigaPath stands at the forefront of a quantum leap in digital pathology.