Artificial intelligence in prostate MRI: Comparative diagnostic performance in a high-prevalence cohort

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Artificial intelligence (AI) is increasingly integrated into prostate cancer diagnostic workflows.
• Validation of AI in high-prevalence academic referral centers remains limited.

Purpose of the Study:

• To compare the diagnostic performance of licensed AI software against radiologist readings for prostate cancer detection.
• To utilize histopathology as the reference standard for evaluating diagnostic accuracy.

Main Methods:

• Retrospective analysis of 959 patients who underwent prostate MRI for suspected prostate cancer.
• Assessment of diagnostic performance metrics including sensitivity, specificity, PPV, NPV, and accuracy across PI-RADS thresholds.
• Utilized ROC analysis, Cohen's kappa, and paired McNemar's test for statistical evaluation.

Main Results:

• AI assigned fewer PI-RADS 3 scores compared to radiologists (κ = 0.388).
• Radiologists demonstrated higher specificity (20.9%) than AI (37.2%) at PI-RADS ≥3 threshold, though AI had higher PPV (90.3% vs 88.7%).
• AI showed a higher false-negative rate in the prostatectomy subgroup (8.4% vs 4.1%).

Conclusions:

• AI diagnostic performance for clinically significant prostate cancer (csPCa) is comparable to radiologists, with similar AUC.
• AI effectively reduced PI-RADS 3 indeterminate scores but exhibited a higher rate of missed significant cancers.
• Further validation and refinement of AI tools are necessary for optimal clinical integration in prostate cancer diagnostics.