Neurology and Neurosurgery

Aug. 14, 2026

Genetic profiling as a tool for molecular classification has revolutionized care for patients with brain tumors. But genetic profiling is resource intensive and not widely available. A multicenter study led by Mayo Clinic has found that deep learning models can identify molecular subtypes and predict outcomes in meningiomas using only routine hematoxylin and eosin (H&E) staining. The results have potential to widen access to genetic information and improve patient care.

In a retrospective study, the researchers created a cohort of patients with meningioma with paired DNA methylation and matched digitized H&E images. The cohort included a training dataset consisting of a cross-section of real-world cases from the National Cancer Institute and an independent test cohort from Canada's University Health Network. A second test dataset consisting of World Health Organization (WHO) grade 2 meningiomas that were completely surgically removed at Mayo Clinic was used for additional clinical validation.

The researchers then trained and validated five deep learning models to use H&E slides alone to predict molecular classification of meningiomas, three relevant chromosome arm aneuploidies and DNA methylation-based prognosis for five-year progression-free survival.

Key findings:

• The deep learning classifier achieved balanced accuracies of 87% to 97% for predicting meningioma molecular group from H&E staining.
• Area under the curve scores for chromosome-level alterations were 0.86 for chromosome 1p loss, 0.86 for chromosome 22q loss and 0.79 for chromosome 1q gain.
• The dedicated outcome prediction model could confidently stratify patients whose tumors recurred after surgery from patients whose tumors didn't — even after adjusting for WHO grade, extent of resection and patient age.

"This is the first study to our knowledge to apply deep learning models that can identify molecular subtypes and predict outcomes in meningioma using H&E staining alone," says Gelareh Zadeh, M.D., Ph.D., chair of Neurosurgery at Mayo Clinic in Rochester, Minnesota, and the study's senior author. "This approach provides an alternative to genomic profiling that can substantially democratize access to modern tumor subclassification tools."

Dr. Zadeh notes that genomic profiling requires considerable time to complete, with results typically unavailable until weeks or even months after surgery. "By contrast, our H&E-based models require less than one hour to run," she says. "They can immediately inform clinical decision-making and allow for personalized care."

For more information

Landry AP, et al. Deep learning for H&E-based meningioma molecular classification and outcome prediction: A retrospective cohort study. The Lancet Digital Health. 2026;8:100986.

Refer a patient to Mayo Clinic.