An artificial intelligence (AI)-based system called "FastGlioma" makes it possible to analyze the tissue removed from a brain tumor during a difficult operation. The US-Austrian team presents its latest development in the journal "Nature". In just a few seconds, it allows precise estimates to be made as to whether the tissue that has just been removed is tumor cells from a glioma or already healthy tissue. The system was developed with significant collaboration from MedUni Vienna.
Determining where a tumor ends and healthy tissue begins is particularly important in the brain. If too much is removed, this can impair central processes such as the ability to speak or move. If cancer cells remain, this increases the likelihood of the disease returning early with a correspondingly reduced survival rate. The increasing duration of brain surgery also increases the risk of complications for patients. It is therefore crucial to arrive at reliable assessments more quickly or as promptly as possible.
A team led by Todd Hollon from the University of Michigan (USA), together with researchers from the University of California in San Francisco, New York University, the Department of Neurosurgery and the Division of Neuropathology and Neurochemistry at the Department of Neurology at the Medical University of Vienna, has been working for years on the development of such an approach.
Georg Widhalm and Lisa Körner (both from the Department of Neurosurgery), who are driving the project forward with Thomas Rötzer-Pejrimovsky (Division of Neuropathology and Neurochemistry) and others, explained that the Vienna General Hospital has already performed more than 500 operations using AI histopathology since 2020. Instead of taking around half an hour to analyze a sample using conventional methods, the tandem human-machine learning system at the Medical University of Vienna is much faster.
System was trained on around four million images
In the current study, numerous glioma samples were analyzed using the new technology. In total, the AI was trained with around four million images, as the researchers write in their paper. Neuropathological and molecular biological analyses were also carried out by the MedUni Vienna research group: "This data contributed significantly to the development of the new AI ’FastGlioma’," say Körner and Widhalm. The system can now estimate within a few seconds how much a freshly surgically removed sample has been penetrated by cancer cells - the experts refer to this as "infiltration" by the tumor. According to the research results, "FastGlioma" outperforms the standard methods used in surgery, which enable differentiation based on images or fluorescent contrast agents.
The authors of the study are convinced that what has now been demonstrated for gliomas can also be transferred to other brain tumor diagnoses in children and adults. After all, there are currently around 120 different types of tumor that can affect the brain. Supported in this way, the chances of removing a maximum of the tumor with pinpoint accuracy would increase accordingly: "We expect this to improve the prognosis of brain tumor patients in the future," say the co-developers from MedUni Vienna. The new findings would show the great potential that AI has in the care of cancer patients as a whole, the researchers write in "Nature".
Publication: Nature
Foundation models for fast, label-free detection of glioma infiltration
Akhil Kondepudi, Melike Pekmezci, Xinhai Hou, Katie Scotford, Cheng Jiang, Akshay Rao, Edward S. Harake, Asadur Chowdury, Wajd Al-Holou, Lin Wang, Aditya Pandey, Pedro R. Lowenstein, Maria G. Castro, Lisa Irina Koerner, Thomas Roetzer-Pejrimovsky, Georg Widhalm, Sandra Camelo-Piragua, Misha Movahed-Ezazi, Daniel A. Orringer, Honglak Lee, Christian Freudiger, Mitchel Berger, Shawn Hervey-Jumper & Todd Hollon
https://doi.org/10.1038/s41586’024 -08169-3
