AI matches radiologists for prostate cancer MRI detection but misses more significant cancers in high-prevalence cohort

AI matched radiologists for csPCa detection on prostate MRI (AUC 0.739 vs 0.763, P=.28) but had more false negatives (8.4% vs 4.1% in prostatectomy subgroup). AI reduced indeterminate PI-RADS 3 scores.

• At PI-RADS ≥3, AI sensitivity was 91.3% vs radiologists' 96.0%, specificity 37.2% vs 20.9%; AUCs comparable (0.739 vs 0.763, P=.28).
• In the radical prostatectomy subgroup, AI's false-negative rate was higher (8.4% vs 4.1%; OR 3.83, 95% CI 1.52–11.51).
• Retrospective single-center study of 959 patients (mean age 69.7 years) with 86.5% csPCa prevalence; AI assigned fewer PI-RADS 3 scores (κ=0.388).

Automated summary

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