Researchers at Microsoft, Providence Health System, and the University of Washington have developed a new generative AI model for diagnosing cancer called Prov-GigaPath. This model is based on an analysis of over a billion images of tissue samples from more than 30,000 patients and is already being used in clinical applications. Through AI tools like Prov-GigaPath, researchers hope to uncover novel relationships and insights in pathology slides that go beyond what the human eye can discern. The team behind this groundbreaking work decided to make the model widely available to benefit patients globally.
The development of Prov-GigaPath utilized OpenAI’s GPT-3.5 generative AI platform to identify patterns in 1.3 billion pathology image tiles obtained from 171,189 digital whole slides provided by Providence. This effort was the largest pre-training endeavor to date with whole-slide modeling, utilizing a database five to 10 times larger than other datasets. Whole-slide imaging has become a critical tool in digital pathology, but the large size of gigapixel slides presents challenges for conventional computer vision programs. Microsoft’s GigaPath platform employed AI-based strategies to break up the large-scale images into more manageable tiles and look for patterns associated with various cancer subtypes.
To evaluate the performance of Prov-GigaPath, the researchers set up a digital pathology benchmark that included nine cancer subtyping tasks and 17 analytical tasks. The model achieved state-of-the-art performance on 25 out of 26 tasks, with significant improvement over the second-best model on 18 tasks. The authors of the study believe that AI-assisted digital pathology opens up new possibilities to advance patient care and accelerate clinical discovery. However, they acknowledge that there is much more to explore in key precision health tasks such as modeling tumor microenvironment and predicting treatment response.
The authors of the Nature paper, “A Whole-Slide Foundation Model for Digital Pathology From Real-World Data,” include several individuals from Microsoft, Providence, and the University of Washington. They highlight the potential of AI in digital pathology to significantly advance cancer research and diagnostics. The Prov-GigaPath model has shown promising results in cancer subtyping tasks, but there is still much room for further exploration and improvement in precision health tasks. The researchers emphasize the need to continue exploring the impact of GigaPath and whole-slide pretraining in areas such as predicting treatment response and modeling tumor microenvironment.
Overall, the development of Prov-GigaPath represents a significant step forward in the use of AI for digital pathology. By analyzing a vast dataset of pathology images, the model has demonstrated strong performance in cancer subtyping tasks and analytical tasks. The widespread availability of Prov-GigaPath has the potential to benefit patients globally by providing clinicians with valuable insights that may not be visible to the human eye. Moving forward, the researchers are focused on expanding the application of Prov-GigaPath to other precision health tasks to further enhance patient care and clinical discovery in the field of cancer research.
