An FDA-cleared artificial intelligence platform detected 39% more intracranial aneurysms than radiologists working without AI assistance, according to a prospective study published in the Journal of the American College of Radiology. The research, conducted at Northwell Health across 3,856 consecutive brain CTA examinations, identified 55 additional aneurysms that radiologists initially missed—a finding that could reshape how hospitals deploy AI tools in neuroradiology workflows.

AI Outperforms Radiologists on Sensitivity While Maintaining Specificity

The study, led by Dr. Shlomit Goldberg-Stein, Director of Artificial Intelligence in Northwell Health's Department of Radiology, evaluated Aidoc's AI algorithm across inpatient, emergency, and outpatient settings during November and December 2023. The AI achieved 84.6% sensitivity compared to 71.8% for radiologists alone, while maintaining comparable specificity at 98.7% versus 98.5% for physicians.

Overall concordance between AI and radiologists reached 96.3%, with agreement on 3,714 of the 3,856 examinations. The study found that approximately 70 examinations were needed to identify one additional aneurysm across all settings. More than half of the AI-exclusive detections—31 of 55 aneurysms—measured less than 3 mm in diameter, suggesting the algorithm complements rather than simply replicates human interpretation patterns.

Inpatient Settings Show Strongest Clinical Benefit

Performance varied significantly by care setting, with the most dramatic impact observed in high-acuity inpatient environments. In inpatient settings, the AI demonstrated 78.3% enhanced detection, identifying one additional aneurysm per 28.9 examinations. Emergency department performance showed 37.1% enhanced detection, while outpatient settings yielded only 14.1% improvement—a context where false positives actually exceeded true positive findings.

This setting-specific variation suggests that hospitals should strategically target AI deployment rather than implementing blanket coverage across all radiology workflows. The strongest benefit appears in environments where missed diagnoses carry the highest morbidity and mortality risk, and where radiologists face the greatest time pressure and case complexity.

Implications for Clinical Practice and Market Competition

The prospective, real-world nature of the Northwell Health study provides stronger evidence than controlled trial environments, offering concrete data for hospital procurement decisions. The finding that AI predominantly detects small aneurysms raises important questions about clinical significance, given ongoing debates about management of sub-3mm findings and the balance between early intervention and potential overtreatment of incidental discoveries.

For Aidoc, the results strengthen its competitive position against rivals including Viz.ai, iSchemaView, and RapidAI in the neuroradiology AI market. The 39% enhanced detection rate provides a measurable value proposition, though the number-needed-to-screen metric of 70 examinations per additional aneurysm will inform return-on-investment discussions and reimbursement strategies as health systems evaluate deployment costs against clinical benefits.

Sources

  1. AI Tool Detects 39% More Intracranial Aneurysms Than Radiologists, Study Finds — Digital Health News