When surgeons remove brain tumors, they rarely get all the cancerous cells, since they so closely resemble healthy brain tissue. Technology such as MRI and fluorescent imaging agents are often inaccurate.
Scientists from the University of Michigan and the University of California, San Francisco, found a way to overcome those limitations through artificial intelligence. Using more than 11,000 surgical specimens and four million microscopic images, they taught a computer how to detect the cancer cells left over after the main mass of a glioma is removed. (A glioma is a growth of cells which starts in the brain or spinal cord.)
(Public domain)
The technology, called FastGlioma, takes only seconds and has an accuracy rate of 90% or better, compared to traditional techniques that miss residual cancer about 25% of the time.
Source: Kondepudi, A., Pekmezci, M., Hou, X. et al. “Foundation models for fast, label-free detection of glioma infiltration,” Nature (2024).
