Learning neuroimaging models from health system-scale data.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Machine Learning for Healthcare

Background:

• Growing demand for magnetic resonance imaging (MRI) strains healthcare systems, causing delays and physician burnout.
• These challenges disproportionately affect patients in underserved rural and low-resource areas.
• Existing AI models often lack generalizability for real-world clinical neuroimaging data.

Purpose of the Study:

• To develop Prima, an AI foundation model for neuroimaging analysis.
• To support clinical magnetic resonance imaging (MRI) studies using real-world data.
• To improve diagnostic accuracy and efficiency in evaluating neurological diseases.

Main Methods:

• Trained Prima on over 220,000 MRI studies from a large academic health system.
• Utilized a hierarchical vision architecture for general and transferable MRI features.
• Tested Prima in a system-wide study involving 29,431 MRI studies across 52 diagnoses.

Main Results:

• Prima achieved a mean diagnostic area under the curve (AUC) of 92.0% across 52 radiologic diagnoses.
• Outperformed state-of-the-art general and medical AI models in diagnostic accuracy.
• Demonstrated algorithmic fairness across sensitive patient groups.

Conclusions:

• Prima shows transformative potential for AI-driven healthcare through health system-scale training.
• The model offers explainable differential diagnoses, worklist prioritization, and referral recommendations.
• Prima can significantly advance the application of AI in clinical neuroimaging and neurological disease management.