AI Tool Predicts Meningioma Recurrence from Routine Pathology Slides with 97% Accuracy
Key Insights
A new AI tool developed by University of Toronto and Mayo Clinic researchers can classify meningioma (search) molecular groups from routine H&E pathology slides with up to 97% accuracy.
The deep learning models provide prognostic information in under an hour using only a laptop, eliminating the need for costly and time-consuming DNA methylation profiling.
The study, published in The Lancet Digital Health, analyzed tissue samples from over 600 patients and demonstrated AI-based recurrence predictions remained useful beyond traditional clinical factors.
Researchers at the University of Toronto and Mayo Clinic have developed an artificial intelligence tool capable of analyzing routine pathology slides to classify meningiomas into molecular groups and predict a patient's risk of tumor recurrence — all in under an hour using only a laptop. The findings, published in The Lancet Digital Health, represent a significant step toward making advanced tumor insights accessible in settings where costly molecular testing is unavailable.
The study, funded in part by The Brain Tumour Charity (search) and the Canadian Institutes of Health Research, applied deep learning models to hematoxylin and eosin (H&E) stained tissue images from more than 600 patients. These are the same tissue images already generated during routine post-surgical care.
Classifying Meningiomas Without Molecular Testing
Meningiomas are the most common primary brain tumor in adults and are usually low grade. However, some grow more aggressively and are more likely to recur after treatment. DNA methylation profiling can provide valuable diagnostic and prognostic information about these tumors, but the technology is expensive, requires specialized expertise, and can take several weeks to return results. It is also unavailable in many hospitals worldwide.
The AI tool achieved up to 97% accuracy in reliably classifying meningiomas into molecular groups that, according to a growing body of research, can help clinicians deliver more accurate prognoses and personalized treatment plans than traditional approaches alone.
"This is one of the many studies where we can harness the strength of digital pathology by capturing the last two decades of genomic and molecular knowledge into AI algorithms," said Gelareh Zadeh, M.D., Ph.D., chair of the Department of Neurologic Surgery at Mayo Clinic in Rochester and senior author on the paper.
Prognostic Value Beyond Traditional Clinical Factors
The study found that AI-based recurrence predictions remained useful even after accounting for traditional clinical factors such as tumor grade, the extent of surgical resection, and patient age. The models were also able to identify patterns of tumor heterogeneity — differences within the same tumor — that may help explain why some tumors behave more aggressively or respond differently to treatment.
For patients, recurrence risk can directly influence follow-up care decisions, including imaging frequency and whether adjuvant radiation therapy should be considered after surgery.
Making the Technology Accessible
The models developed in this study are now being made publicly available, with researchers hoping they can serve as a useful decision support tool for clinicians. Validation for wider clinical use is ongoing, and the authors note that additional prospective studies are needed before the AI models can be used routinely in clinical care.
"The aim is to make these algorithms readily and simply accessible for use globally, improving patient care across many healthcare settings," said Dr. Zadeh.
Andy Tudor, a retired defence systems engineer from Hampshire, UK, who was diagnosed with a grade 2 meningioma (search) in 2017 and experienced a recurrence five years later, welcomed the research: "It's fantastic to see technologies like AI being used in such a positive way. I very much welcome its potential to help tailor meningioma treatment, which could prove a massive benefit for people like me in the future."
The researchers suggest the approach may also lay the groundwork for similar AI applications in other cancers, extending the potential impact beyond meningioma (search) care.