Multi-parametric MRI (mp-MRI) facilitates identification of transition zone (TZ) cancers, yet its overall diagnostic accuracy is likely lower in this part of the prostate compared with the peripheral zone. Prior studies [1] have built a TZ diagnostic model based on quantitative mp-MRI parameters for three different cancer definitions and validated it on a temporal cohort of patients, scanned at the same 1.5T Siemens MR scanner at a different time period. In this study we validate the diagnostic ability of this model on a truly independent cohort of patients scanned on a 3T Philips MR scanner.The previously reported Logistic Regression (LR) model was derived from 1.5T (Avanto, Siemens, Erlangen, Germany) mp-MRI studies [1]. All patients within the new independent cohort underwent a 3T (Achieva, Philips Medical Systems, Best, Netherlands) mp-MRI. All images were acquired with a pelvic-phased array coil; 0.2 mg/kg (maximum 20 mg) of spasmolytic (Buscopan; Boehringer Ingelheim, Ingelheim, Germany) was administered intravenously to reduce peristalsis. The mp-MRI included axial and coronal small field of view T2-weighted imaging; and was supplemented by axial diffusion weighted imaging (DWI) and dynamic contrast enhanced (DCE) imaging as reported in the PICTURE dataset [2].All men within the PICTURE study underwent a mp-MRI followed by full template mapping biopsy of the prostate. For this study, the PICTURE database was retrospectively searched for (i) a cohort of men with histologically confirmed transition zone prostate cancer; (ii) a cohort of men negative for transition zone prostate cancer. Sixty-five men (mean age 64 years, range 48–83) with a mean PSA of 8.68 ng/ml (range 0.5–32.4 ng/ml) and mean prostate gland volume of 37.7 ml (range 3–90 ml) comprised the full independent cohort. Cancer was classified by three definitions of significance: (i) anycancer; (ii) definition-1 (≥Gleason 4+3 or ≥6mm cancer core length (CCL)) [high risk-significant]; and (iii) definition-2 (≥Gleason 3+4 or ≥4mm CCL) cancer [intermediate-high risk-significant]. Previous work1 has showed that the three mp-MRI variables that best classified TZ PCa are the Apparent Diffusion Coefficient (ADC), normalized T2 signal intensity and Maximum Enhancements (ME), and trained a LR model based on these parameters for each cancer definition. These LR models are now validated on the PICTURE independent cohort using receiver operating characteristic (ROC) analysis for each cancer definition. In the PICTURE independent validation patient cohort, 31/65 patients were positive for any-cancer, of these 19/31 for definition-2 cancer, and of these 14/19 for definition-1 cancer. 34/65 patients had a negative biopsy of the transition zone. Applying the previously derived LR model [1] to the PICTURE dataset, the ROC area under curve (ROC-AUC) for prediction of patients with any-cancer, definition-2 and definition-1 cancer model was 0.69 (95% CI 0.56–0.82), 0.66 (95% CI 0.53–0.79) and 0.63 (95% CI 0.50–0.77). To achieve >90% sensitivity on the PICTURE independent cohort, the specificities of the models were 32% (thre=0.65; any-cancer), 52% (thre=0.96; definition-2), and 45% (thre=0.35; definition-1), where thre is the probability threshold. Figures 1 and 2 compare the performance of the LR model on the PICTURE independent cohort of patients scanned on a 3T scanner with the previously published performance1 on a temporal validation cohort scanned on a 1.5T scanner. The ROC-AUC performance on the temporal validation cohort for any-cancer, definition-2 and definition-1 cancer model was 0.76 (95% CI 0.66–0.87), 0.67 (95% CI 0.55–0.79) and 0.70 (95% CI 0.55–0.85). To achieve >90% sensitivity on the temporal validation cohort, the LR model had specificities of 25% (thre=0.52; any-cancer), 30% (thre=0.27; definition-2), and 33% (thre=0.27; definition-1). The specificities achieved on the temporal validation cohort for high sensitivities are lower than the respective specificities for the PICTURE independent cohort for all cancer definitions.Clinically it will be more important not to miss patients with significant cancer. A high sensitivity is required to achieve this. We demonstrate that overall performance of TZ specific mp-MRI LR model based on a 1.5T MR scanner can be applied on an independent cohort of patients scanned at a 3T MR scanner. Furthermore, at a high sensitivity (>90%) the models provide a moderate specificity (approx. 50%) for significant cancer (definition-2).