Researchers from Microsoft, Providence Health System, and the University of Washington have developed a new generative AI model called Prov-GigaPath for diagnosing cancer. The model is based on an analysis of over a billion images of tissue samples from more than 30,000 patients. The open-access model has already been used in clinical applications and is aimed at advancing cancer research and diagnostics. The researchers utilized OpenAI’s GPT-3.5 generative AI platform to analyze 1.3 billion pathology image tiles and create the model.
Whole-slide imaging, a process that transforms a microscope slide of tumor tissue into a high-resolution digital image, is widely used in digital pathology. However, handling the large gigapixel slides poses challenges for conventional computer vision programs. Microsoft’s GigaPath platform used AI-based strategies to break up the large images into manageable 256-by-256-pixel tiles and identify patterns associated with various cancer subtypes. To evaluate the performance of the Prov-GigaPath model, the researchers set up a digital pathology benchmark consisting of nine cancer subtyping tasks and 17 analytical tasks.
Prov-GigaPath achieved state-of-the-art performance on 25 out of 26 tasks, surpassing the second-best model on 18 tasks, according to study authors Hoifung Poon and Naoto Usuyama from Microsoft. They believe that the AI-assisted approach to digital pathology can enhance patient care and clinical discovery, but emphasize that further research is needed. The potential impact of GigaPath and whole-slide pretraining on precision health tasks such as modeling tumor microenvironment and predicting treatment response has yet to be explored.
The Nature paper, “A Whole-Slide Foundation Model for Digital Pathology From Real-World Data,” was authored by a team including Carlo Bifulco, Hoifung Poon, Naoto Usuyama, and other collaborators. The research drew upon a database of 1.3 billion pathology image tiles and 171,189 digital whole-slide images provided by Providence Health System. The model aims to uncover novel relationships and insights in pathology slides that may not be discernible to the human eye, ultimately benefiting patients globally.
The researchers highlight the importance of making Prov-GigaPath widely accessible to drive advancements in cancer research and diagnostics. The study leveraged a large dataset for whole-slide modeling, which was significantly larger than previous datasets used in similar research efforts. By using AI to analyze and interpret pathology images, the Prov-GigaPath model offers new possibilities for precision medicine and accelerating clinical discovery. Despite achieving impressive results, the researchers acknowledge the need for further investigation into the model’s potential in key precision health tasks.
