Teaching AI to speak the language of pathology
microsoft ·
Researchers from Microsoft Research and Paige have created PRISM2, a new pathology foundation model designed to bridge the gap between medical imaging and clinical language. Unlike traditional artificial intelligence systems that require complete redevelopment for each specific diagnostic task, this model learns by analyzing both tissue photographs and corresponding pathology reports. By generating millions of question-and-answer pairs from real-world clinical documents, the system successfully connects visual tissue patterns with the terminology pathologists use during diagnoses. In recent evaluations published in Nature Medicine, PRISM2 performed on par with or better than dedicated cancer-detection algorithms across multiple benchmarks, including prostate and breast cancer analysis. The architecture allows clinicians and researchers to interact with the tool using natural language prompts rather than relying on rigid, single-purpose software. Because the underlying weights are openly accessible, developers can potentially adapt the framework for various medical applications without building foundational systems from the ground up. This multimodal approach marks a significant shift toward more flexible and scalable artificial intelligence solutions in diagnostic medicine.
microsoft