Researchers at Thomas Jefferson University have developed an automated machine learning (AutoML) model capable of distinguishing between two common types of brain tumors using preoperative MRI scans. The findings, published in the December 2025 issue of Otolaryngology–Head and Neck Surgery, mark the first use of AutoML technology specifically trained to classify pituitary macroadenomas and parasellar meningiomas.
“Our automated machine learning model achieved over 97% accuracy in distinguishing between two common types of skull base tumors (pituitary macroadenomas and meningiomas of the parasellar region) using preoperative MRI scans. This work is significant because it demonstrates that automated machine learning can streamline model development for medical imaging classification, reducing barriers to implementing artificial intelligence-based diagnostic support in otolaryngology,” said Gurston G. Nyquist, MD, Professor of Otolaryngology and Neurological Surgery, and Chief, Division of Rhinology and Skull Base Surgery at Thomas Jefferson University.
“While multi-institutional validation and careful integration into clinical workflows are warranted, this study represents an important step in the development of reliable tools that may improve skull base tumor diagnosis in both community and tertiary care settings,” he continued.
Accurate diagnosis before surgery is important because these tumors require different surgical methods. Biopsies are rarely performed on brain masses prior to surgery, so correct interpretation of imaging is essential. Misdiagnosis can result in inadequate preparation or less effective outcomes for patients.
The research team analyzed 1,628 MRI images from 116 patients. Their AutoML model achieved an overall accuracy rate of 97.55%. For pituitary macroadenomas, sensitivity was 97% with specificity at 98.96%. For parasellar meningiomas, sensitivity reached 98.41% with specificity at 95.53%. External validation using an additional set of 959 images confirmed the reliability of these results.
The ability to adjust confidence thresholds after developing the model allows flexibility for different clinical environments. High-sensitivity settings could be used where specialist access is limited; high-specificity settings might help reduce false positives in larger care centers.
The technology may assist with initial evaluations, expedite referrals to specialists, improve surgical planning before operations, and provide educational resources for medical trainees.
Future plans include expanding the model by adding more imaging types and clinical data such as hormone levels. The researchers also aim to adapt their approach for other uses beyond skull base surgery.
The study was conducted through a collaboration between Thomas Jefferson University’s Department of Otolaryngology-Head and Neck Surgery and Department of Neurological Surgery. It received exemption approval from the institutional review board and was presented at the AAO-HNSF 2025 Annual Meeting & OTO EXPO held October 11-14 in Indianapolis.
Otolaryngology–Head and Neck Surgery is a peer-reviewed journal published by the American Academy of Otolaryngology–Head and Neck Surgery Foundation (AAO-HNS/F). The AAO-HNS/F represents about 13,000 specialists worldwide who treat disorders related to ears, nose, throat, head, and neck structures.
