Recently, American scientists published significant research findings in the journal Nature: the artificial intelligence model FastGlioma, jointly developed by the University of Michigan and the University of California, San Francisco, can quickly determine the presence of cancerous tumor remnants within 10 seconds during brain tumor surgeries, bringing a revolutionary breakthrough to neurosurgery.
This innovation combines microscopic optical imaging with AI foundational models. The research team utilized over 11,000 surgical samples and 4 million microscopic images for pre-training, employing a high-resolution imaging technique developed by the University of Michigan called stimulated Raman tissue imaging.
The outstanding advantage of FastGlioma lies in its exceptional detection capability. In practical applications, the model has a mere 3.8% rate of missing high-risk tumor remnants, far superior to the 25% omission rate of traditional imaging and fluorescence-guided surgeries. Even in "fast mode," its average accuracy rate remains at 92%.
Research shows that FastGlioma can also reduce reliance on traditional methods such as radiographic imaging, contrast enhancement, or fluorescence labeling. This groundbreaking technology not only aids surgeons in making quick decisions during operations but can also be applied to the diagnosis of other types of brain tumors.
It is noteworthy that complete removal of brain tumors has always been a significant challenge in neurosurgery, as distinguishing between residual tumors and healthy brain tissue can be difficult. The emergence of FastGlioma offers a new solution to this clinical dilemma, marking an important step forward for artificial intelligence in the field of precision medicine.