External validation of deep learning for prostate cancer detection on MRI shows comparable performance to radiologists

A deep learning algorithm (XProstate) for prostate cancer MRI detection performed similarly to radiologists in external validation: heat map AUC 0.76 (0.70-0.82) vs PI-RADS 0.73 (0.68-0.79), p=0.38. 202 patients, MR-US fusion biopsy gold standard.

• The algorithm's heat map (AUC 0.76) and ROI (AUC 0.71) were not significantly different from radiologists' PI-RADS (AUC 0.73; p=0.38 and 0.43).
• The study used true external data (no training overlap) and MR-US fusion biopsy as gold standard, simulating real-world deployment.

Automated summary

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