Researchers at Mass General Brigham have developed advanced AI foundation models, UNI and CONCH, that have the potential to revolutionize the medical field. These models, trained on large-scale datasets, show promise in enhancing diagnostic accuracy, prognostic insights, and predicting therapeutic responses in computational pathology (CPath).
UNI, a foundation model focusing on understanding pathology images, has the ability to recognize diseases in histology region-of-interests and whole slide imaging. Trained on a vast database of tissue patches and whole slide images, UNI showcases remarkable adaptability and transfer learning capabilities across 34 clinical tasks, including cancer classification and organ transplant assessment. On the other hand, CONCH, another foundation model developed by the research team, is trained to understand pathology images and language. With a database of histopathology image-text pairs, CONCH excels in identifying rare diseases, tumor segmentation, and interpreting gigapixel images.
The research team at Mass General Brigham aims to make the code publicly available for other academic groups to address clinically relevant problems using these foundation models. These models represent a new era in medical artificial intelligence, offering adaptability to various downstream tasks. The publication of research papers on UNI and CONCH in Nature Medicine marks a significant step towards the integration of AI systems in pathology for improved patient outcomes.
Supported by funding from various sources, including the BWH president’s fund and NIH grants, the groundbreaking work by Mass General Brigham researchers highlights the potential of AI in reshaping medical practices. The co-authors of the UNI and CONCH papers include experts in the field of pathology and computational sciences, indicating a collaborative effort to advance healthcare through innovative technologies. By pushing the boundaries of AI in healthcare, these foundation models open doors to new possibilities for disease detection, diagnosis, and treatment, paving the way for a more personalized and effective approach to patient care.
