A clinically-oriented foundation model for intraoperative pathology.

CRISP, a clinically oriented foundation model trained on over 100,000 frozen sections, achieves 92.6% diagnostic accuracy in intraoperative pathology across diverse cases.

• Why it matters: Intraoperative pathology is essential for precision surgery but faces challenges from diagnostic complexity and limited high-quality data, hindering routine clinical adoption of AI tools.
• What they did: Developed using extensive data from ten medical centers, CRISP was evaluated on more than 15,000 slides across nearly 100 diagnostic tasks, demonstrating strong generalization across institutions, tumor types, and sites.
• The result: CRISP supports surgical decisions in real-world settings, reduces diagnostic workload by 35%, and improves detection of micrometastases, paving the way for AI integration into clinical intraoperative pathology.