DeepGlioma AI classifies brain tumors within 90 seconds—and with 93% accuracy, study finds

An artificial-intelligence-powered screening tool could make it easier for doctors to decide on the best treatment route for patients with certain types of brain tumors, according to a newly published study.

Though surgical removal may seem like the obvious choice to treat cancerous tumors, when it comes to diffuse gliomas—among the most common and dangerous brain tumors—doctors and researchers have found that the risks of the major surgery may outweigh its potential benefits among certain molecular subtypes of gliomas.

That’s where the AI tool comes in. Currently, molecular tests for brain tumors aren’t widely available and may also take days or weeks to return results, according to Todd Hollon, M.D., a neurosurgeon at University of Michigan Health and the creator of the tool. In contrast, the DeepGlioma system aims to churn out molecular analyses in less than two minutes.

“DeepGlioma creates an avenue for accurate and more timely identification that would give providers a better chance to define treatments and predict patient prognosis,” Hollon said in a Michigan Medicine press release this week.

DeepGlioma’s machine learning AI works by applying deep neural network algorithms to scans of brain tumor tissue as surgeons are collecting tissue samples. Those scans are generated through a new real-time imaging technique called stimulated Raman histology, another University of Michigan creation.

The algorithms identify the molecular subtype of each tumor based on a classification system that was developed by the World Health Organization (WHO) and most recently updated in 2021.

In the study, which was published last week in Nature Medicine, a research team representing Michigan Medicine, New York University and the University of California, San Francisco plus a handful of other institutions applied the DeepGlioma AI to 153 patients with diffuse gliomas who had undergone the real-time imaging process.

Altogether, the researchers concluded, the system was able to correctly classify the tumors to their WHO-designated subgroups with an average accuracy of 93.3%—in less than 90 seconds each.

Though the study results suggest DeepGlioma could significantly improve doctors’ ability to decide on a surgical or chemotherapy treatment regimen, the researchers noted in the release that currently available treatments for diffuse gliomas still fall short, citing data showing a median survival time of just 18 months for patients with the malignant tumors.

The AI tool’s creators are hoping, then, that the promise of DeepGlioma could encourage more people with gliomas to enroll in clinical trials that may then produce new treatments specific to tumor subtypes.

“Progress in the treatment of the most deadly brain tumors has been limited in the past decades—in part because it has been hard to identify the patients who would benefit most from targeted therapies,” said Daniel Orringer, M.D., a senior author of the study and one of the creators of stimulated Raman histology. “Rapid methods for molecular classification hold great promise for rethinking clinical trial design and bringing new therapies to patients.”