The paper details the development and validation of FastGlioma, an advanced artificial intelligence system designed to detect residual tumor infiltration during brain surgery. By utilizing stimulated Raman histology (SRH) and foundation models, this technology provides rapid, high-resolution imaging that identifies cancerous cells within surgical margins in near real-time. The research addresses a critical public health issue, as leftover tumor tissue significantly reduces patient survival rates and adds billions in healthcare costs annually. Through self-supervised learning and ordinal metric learning, the system achieves high diagnostic accuracy across various glioma subtypes and shows potential for future application in lung, breast, and prostate cancers. Ultimately, the sources highlight a major shift toward precision surgery, offering an accessible and affordable tool to improve global cancer outcomes.

References:

• Kondepudi A, Pekmezci M, Hou X, et al. Foundation models for fast, label-free detection of glioma infiltration[J]. Nature, 2025, 637(8045): 439-445.