Stanford-led study finds multimodal AI diagnostic models outperform human radiologists in key cancer screenings.
A research consortium led by Stanford Medicine reports that multimodal AI models now outperform human radiologists in detecting early-stage lung and breast cancers. By integrating imaging data with clinical and genomic records, the model achieved 96% sensitivity and 94% specificity—surpassing average clinician performance.
The breakthrough lies in cross-modal context: combining CT scan interpretation with EHR-derived risk factors. Rather than replace radiologists, AI systems are being deployed as triage tools, flagging high-risk scans for human review. Trials across six hospitals show improved workflow efficiency and earlier diagnosis rates.
Regulators are evaluating guidelines for AI co-signature in diagnostics—a legal recognition that could streamline adoption nationwide. The line between technology and clinician continues to blur, but the result is clear: earlier detection, faster treatment, and a new benchmark for medical precision.