Independent validation of three AI algorithms for intracranial hemorrhage shows Aidoc achieves clinically relevant accuracy

Of three AI tools for intracranial hemorrhage detection on non-contrast head CT, Aidoc achieved 90.3% sensitivity and 99.0% specificity. Simulated AI+radiologist reading boosted sensitivity to 96.0% (specificity 99.4%, P<0.001), matching two-radiologist performance.

• Among the three commercial algorithms, performance varied substantially; only Aidoc demonstrated clinically relevant accuracy.
• Eight intracranial hemorrhages missed by both radiologists were detected by at least one AI system.
• The combination model assumed radiologists perfectly dismissed all false-positive AI flags, which may not reflect real-world practice.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.