AI Lifts Novice Readers More Than Experienced Ones on Prostate Biparametric MRI

AI-assisted biparametric MRI (uAI, United Imaging) raised novice-reader AUC from 0.684 to 0.744 and sensitivity from 0.71 to 0.79 at PI-RADS ≥3 in 646 men; experienced-reader sensitivity gains were smaller and nonsignificant (p=0.344/0.291). Retrospective.

• Retrospective study; 646 men (297 with prostate cancer confirmed on histopathology), 3.0 T bpMRI; 4 readers — 2 novice (≤50 prior cases) and 2 experienced (8–10 years) — each read all cases twice with and without AI, counterbalanced with a 4-week washout.
• For clinically significant cancer (ISUP grade group ≥2 at PI-RADS ≥3), AI improved sensitivity across all readers (0.76→0.82) and specificity (0.47→0.57); novice inter-reader agreement rose meaningfully (κ 0.582→0.700 at PI-RADS ≥3; 0.654→0.741 at PI-RADS ≥4).
• Key limitations: retrospective design, only two readers per experience tier limits generalizability of reader-level conclusions, single AI system tested, and no external or prospective validation.

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