A new AutoML model from Thomas Jefferson University differentiates pituitary macroadenomas from parasellar meningiomas using preoperative MRI scans, achieving over 97% overall accuracy.

Performance metrics show 97% sensitivity and 99%+ specificity for macroadenomas, and 98% sensitivity with 95% specificity for parasellar meningiomas, validated on a large external image set.

Overall accuracy for the model stands at about 97.55%, with macroadenoma metrics around 97% sensitivity and 99% specificity, and parasellar meningioma metrics around 98% sensitivity and 95% specificity; these were validated on 1,628 images from 116 patients and externally on 959 additional images.

Experts caution that multi-institutional validation and seamless integration into clinical workflows are needed before widespread adoption to ensure reliability across diverse care environments.

The research was unveiled at the AAO-HNSF 2025 Annual Meeting & OTO EXPO and published in Otolaryngology–Head and Neck Surgery in December 2025, with plans to expand the model to include additional imaging modalities, clinical data, and multi-label classification.

The tool is designed to act as a second pair of eyes for surgical teams, potentially speeding decision-making, triaging complex cases for specialist review, and improving preoperative planning.

Clinically, the model could support preliminary evaluations, streamlined triage and referrals, and enhanced preoperative planning, with configurable high-sensitivity or high-specificity modes to fit different clinical settings.

The study emphasizes that improving preoperative diagnosis and tailored surgical planning is particularly valuable given imaging overlap and diagnostic challenges between these tumors.

By offering adjustable sensitivity and specificity, the model could streamline workflows for preliminary assessments, triage, referrals, and planning.

This marks the first AutoML application trained specifically to classify these two skull base tumors, addressing diagnostic challenges from overlapping imaging features and varying clinician interpretation accuracy.

The work could streamline AI-based diagnostic support in otolaryngology, contingent on multi-institutional validation and workflow integration.

This summary derives from the Otolaryngology–Head and Neck Surgery article and related press coverage.

Looking ahead, researchers plan to add more imaging modalities and clinical data (e.g., hormone levels), pursue multi-label classification for coexisting pathologies, and explore applications beyond skull base surgery.