Optimizing prostate volume estimation: the role of artificial intelligence compared with conventional methods

Objectives The absence of standardization in measuring prostate volume (PV) limits the utilization of prostate-specific antigen density. This study aimed to assess the accuracy of artificial intelligence (AI), MRI, and transrectal ultrasound (TRUS) in PV evaluation. Methods A total of 2037 consecutive participants who underwent prostate MRI and TRUS-guided prostate biopsy were enrolled and divided into classified subgroups based on MRI prostate morphology: Subgroup 1, ellipsoid-shaped; Subgroup 2, protruding into the bladder; Subgroup 3, bulging beyond the capsule. The consistency of PV measurements obtained from AI automatic segmentation (PV_AI), MRI-based ellipsoid formula (EF) (PV_MRIAVG, calculated as the mean of PV_MRI1 and PV_MRI2 by two radiologists), and TRUS-based EF (PV_TRUS) was evaluated against the reference standard by radiologists’ manual planimetry (PV_ref), using the intraclass correlation coefficient (ICC) and Bland-Altman analysis. Results Both the ICC and Bland-Altman analysis demonstrated superior agreement between PV_AI and PV_ref (ICC 0.999) compared with PV_MRIAVG and PV_TRUS. AI exhibited satisfactory accuracy in PV estimation in all three subgroups, while the consistency of both PV_MRIAVG and PV_TRUS with PV_ref was lower in Subgroup 2 and Subgroup 3 than in Subgroup 1, suggesting AI has a superior advantage in estimating the volume of irregularly shaped prostates compared with MRI and TRUS. Conclusion AI outperforms MRI and TRUS in PV estimation, particularly in cases of irregularly shaped prostates. These results hold the potential to empower the promotion of prostate-specific antigen density.