AI Tool Shows Strong Potential for Accurate Brain Tumor Diagnosis Without Surgery: Study - Video

What if artificial intelligence could diagnose brain tumors more accurately than ever before-before a surgeon even touches a scalpel? Researchers at Thomas Jefferson University have developed an automated machine learning (AutoML) model that can do just that, distinguishing between two look-alike brain tumors with over 97% accuracy. Their findings, published in Otolaryngology–Head and Neck Surgery, could transform how doctors plan surgeries and improve outcomes for patients with skull base tumors.

The study focused on two benign but often-confused tumors that grow near the base of the brain: pituitary macroadenomas and parasellar meningiomas. Although both appear similar on standard MRI scans, they require very different surgical approaches, and pre-surgery biopsies are risky or rarely performed. Misidentifying one can lead to suboptimal operations or longer recovery times.

Led by Dr. Gurston G. Nyquist, a professor of Otolaryngology and Neurological Surgery, the team used a dataset of 1,628 MRI images from 116 patients to train their AutoML model. Unlike traditional AI systems that demand expert programming, AutoML automatically selects, trains, and optimizes algorithms, making it easier to integrate into hospitals. The model’s performance was then validated across 959 additional MRI images, proving its reliability.

Results were striking: Overall accuracy: 97.55%; Pituitary macroadenomas: 97% sensitivity, 98.96% specificity; Parasellar meningiomas: 98.41% sensitivity, 95.53% specificity

The researchers designed the system to adjust its confidence thresholds depending on clinical needs. For example, a high-sensitivity mode (99.39%) could help community hospitals catch more potential tumors, while a high-specificity mode (99.31%) could reduce false alarms in specialist centers.

Beyond diagnostics, the model could assist surgeons in planning procedures, expedite referrals, and even serve as an educational tool for trainees. Looking ahead, the Jefferson team plans to enhance the model by combining MRI data with clinical information and hormonal profiles, potentially expanding its use to other head and neck conditions like thyroid nodules or vocal cord lesions.

This breakthrough brings AI one step closer to the operating room—simplifying tumor diagnosis, saving time, and paving the way for more precise, customized care.

REFERENCE: Elliott M. Sina et al, Automated Machine Learning Differentiation of Pituitary Macroadenomas and Parasellar Meningiomas Using Preoperative Magnetic Resonance Imaging, Otolaryngology–Head and Neck Surgery (2025). DOI: 10.1002/ohn.70034