Supplementary Figure for the reply to the commentary “Fast prostate MRI learning curves in urology and radiology residents: beware of overinterpreting”

Supplementary material for the reply to the commentary “Fast prostate MRI learning curves in urology and radiology residents: beware of overinterpreting” by Russo and Villeirs (European Radiology, https://doi.org/10.1007/s00330-025-12298-2), concerning the original article “Prostate MRI learning curves: establishing training benchmarks for radiology and urology trainees” (European Radiology, https://doi.org/10.1007/s00330-025-12177-w). This supplementary figure provides a side-by-side comparison of learning curves based on two reference standards: consensus PI-RADSv2.1 scores from two expert uroradiologists and the presence or absence of clinically significant prostate cancer (csPCa; defined as ISUP Grade ≥ 2) on combined targeted plus systematic biopsy performed within 6 months after prostate multiparametric MRI (mpMRI). The full cohort comprised 200 consecutive prostate mpMRI cases interpreted by 14 readers (10 radiology trainees and 4 urology trainees). The biopsy subcohort comprised 85 men with available biopsy reference data. In the biopsy-based analysis, a correct classification was defined as true positive (PI-RADS ≥ 3 with csPCa present) or true negative (PI-RADS < 3 with csPCa absent). The figure shows learning trajectories for radiology and urology trainees across the 200-case sequence. Solid curves represent agreement with the consensus expert reference across the full cohort, whereas dashed curves represent biopsy-referenced diagnostic accuracy in the 85-case biopsy subcohort. × markers indicate cases with available biopsy data; rug ticks along the lower x-axis indicate cases with biopsy-confirmed csPCa. Across both reference standards, the overall curve shape is similar, with rapid early improvement followed by a plateau. The biopsy-referenced curves show a slightly more pronounced late decline, likely because case order was balanced by PI-RADS-based difficulty rather than biopsy yield. This led, by chance, to a later clustering of biopsy-negative but PI-RADS-positive cases and, consequently, to more false-positive classifications in the final part of the sequence, whereas a case distribution optimized for biopsy outcomes would likely have spread these cases more evenly across the case sequence. Curves were modeled using natural cubic splines with boundary knots fixed at cases 1 and 200 and internal knots placed at the tertiles of the case sequence, allowing direct visual comparison of trajectory shape across the full study range. Abbreviations: Bx, biopsy; csPCa, clinically significant prostate cancer (defined as ISUP Grade ≥ 2); mpMRI, multiparametric magnetic resonance imaging.