Artificial intelligence as diagnostic aiding tool in cases of Prostate Imaging Reporting and Data System category 3: the results of retrospective multi-center cohort study

Abstract

Purpose: To study the effect of artificial intelligence (AI) on the diagnostic performance of radiologists in interpreting prostate mpMRI images of the PI-RADS 3 category.

Methods: In this multicenter study, 16 radiologists were invited to interpret prostate mpMRI cases with and without AI. The study included a total of 87 cases initially diagnosed as PI-RADS 3 by radiologists without AI, with 28 cases being clinically significant cancers (csPCa) and 59 cases being non-csPCa. The study compared the diagnostic efficacy between readings without and with AI, the reading time, and confidence levels.

Results: AI changed the diagnosis in 65 out of 87 cases. Among the 59 non-csPCa cases, 41 were correctly downgraded to PI-RADS 1-2, and 9 were incorrectly upgraded to PI-RADS 4-5. For the 28 csPCa cases, 20 were correctly upgraded to PI-RADS 4-5, and 5 were incorrectly downgraded to PI-RADS 1-2. Radiologists assisted by AI achieved higher diagnostic specificity and accuracy than those without AI [0.695 vs 0.000 and 0.736 vs 0.322, both P < 0.001]. Sensitivity with AI was not significantly different from that without AI [0.821 vs 1.000, P = 1.000]. AI reduced reading time significantly compared to without AI (mean: 351 seconds, P < 0.001). The diagnostic confidence score with AI was significantly higher than that without AI (Cohen Kappa: -0.016).

Conclusion: With the help of AI, there was an improvement in the diagnostic accuracy of PI-RADS category 3 cases by radiologists. There is also an increase in diagnostic efficiency and diagnostic confidence.

Keywords: Deep learning; Magnetic resonance imaging; Prostate Imaging Reporting and Data System; Prostatic neoplasms.