The analysis yielded 6 sets of diagnostic parameters (SNS, SPC, AUC) characterizing each group in each protocol. Table 2 shows the values of TP, SNS and AUC obtained by the three groups when working with the biparametric protocol compared with those obtained with the multiparametric protocol.

Specificity was always around 0.80 irrespective of the protocol and the readers’ experience (range 0.79-0.84). Significant differences were instead found for Sensitivity and AUC, both related to strong variations in the number of true positives (TP). For the bpMRI protocol, the decreasing experience reduced significantly TP (and consequently SNS and AUC) from Group A to Group B (p<0.0001 for SNS and p=0.03 for AUC) and to a lesser extent from Group B to Group C (p=0.02 for SNS and p=0.40 for AUC), whereas for the mpMRI protocol, the only significant reductions in performance are from Group B to Group C.

With the mpMRI protocol, each reader of Group A reclassified two false negative patients into two true-positives, increasing slightly (p=0.13) the number of TP from 82/90 to 86/90. As a consequence, the performance of Group A was essentially similar with or without DCE sequences: SNS 0.96 vs 0.91 (p=0.25), SPC 0.84 vs 0.78 (p=0.5) and AUC 0.93 vs 0.86(p=0.10).

Conversely, for Group B, the mpMRI reading mode witnessed an increase in the TP number by 30 units, from 52/90 in the previous session to 82/90 (p<0.0001): the two readers’ log showed that the increase was due to 13 upgrading of the PI-RADSv2 attribution from 3 to 4 and to the entry of 17 lesions which had escaped detection in the biparametric evaluation. Eight lesions remained undetected. (Fig.1; Fig.2)

Finally, for Group C, the mpMRI reading mode witnessed an increase in the TP number from 45/90 in the previous session to 64/90 (p=0.006): the two readers’ log showed that the increase was due to 5 upgradings of the PI-RADSv2 attribution from 3 to 4 and to the entry of 14 lesions which had escaped detection in the biparametric evaluation. The sensitivity increased from 0.50 to 0.71 (p=0.04), and so did AUC, even if not significantly (p=0.14). Twenty-six lesions remained undetected.

All readers detected the two lesions in the transitional zone, with volumes of 1.1 and 1.2 cm³.

The greatest differences in diagnostic parameters between the two protocols were seen for Group B. The performances of its two readers were thus used to understand which features of the prostate images impaired the detection of lesions when the DCE sequences were not considered. Nineteen lesions were detected by both readers (overall 38 TP) and 14 lesions by only one of them (14 TP), with 15 lesions missed by both (overall FN=30) and 8 by either one of them (FN=8).