Idea
Prima is a vision language model platform that improves MRI diagnosis accuracy and workflow efficiency for healthcare providers and radiologists
Research Paper
Core Innovation
This paper introduces Prima, the first vision language model trained on health system-scale MRI data, enabling generalizable and explainable neuroimaging diagnosis. Prima outperforms existing AI models across diverse neurologic conditions and supports clinical workflows with fairness and bias mitigation. Its hierarchical vision architecture allows transferability across MRI systems and patient demographics.
Market Size (TAM)
$20–50B TAM for AI-driven medical imaging diagnostics; $2–10B SAM from hospitals and health systems adopting AI for radiology workflow optimization. Driven by increasing MRI demand and need for diagnostic accuracy improvements.
Potential Customers & Pain Points
- Hospitals needing faster and more accurate MRI diagnoses
- Radiology departments facing high workload and burnout
- Health systems aiming to reduce diagnostic disparities in low-resource populations
Business Model
Subscription-based SaaS platform offering API access and enterprise licensing to hospitals and radiology providers
Competitive Landscape
- Aidoc
- Zebra Medical Vision
- Viz.ai
Implementation Challenges
- Integration with diverse hospital IT systems
- Regulatory approval for clinical use
- Data privacy and security concerns
Validation Strategy
- Conduct multi-center clinical trials to validate diagnostic accuracy
- Partner with health systems for pilot deployments
- Collect user feedback to refine workflow integration and fairness features
Research Paper Overview
Learning neuroimaging models from health system-scale data
Summary
Neuroimaging is widely used for neurological disease evaluation but faces challenges like rising MRI demand, long turnaround times, and physician burnout, especially in low-resource settings. This paper presents Prima, a vision language model trained on over 220,000 MRI studies from a large academic health system. Prima uses a hierarchical vision architecture to generate generalizable MRI features and was validated on 30,000 MRI studies across 52 neurologic diagnoses, achieving a mean diagnostic AUC of 92.0, outperforming other AI models. It provides explainable differential diagnoses, prioritizes radiologist worklists, and offers clinical referral recommendations while demonstrating fairness across sensitive groups and helping reduce health system biases.