This single-institution study was part of an evaluation of transperineal prostate biopsies with the need for informed consent for data analysis waived by the local ethics committee. From April 2013 to February 2016, 158 consecutive patients referred for transperineal prostate biopsies from regional hospitals and meeting inclusion criteria were included for analysis. Minimum MRI criteria were based on the European Society of Urogenital Radiology (ESUR) 2012 guidance (below) . Patients were referred through an agreed pathway either for active surveillance for repeat surveillance biopsy, or in patients with negative systematic biopsy but ongoing suspicion of cancer based on prostate-specific antigen (PSA), symptoms or prior high risk biopsy, based on UK guidance. The clinical objective was to identify significant cancer necessitating treatment, defined as Gleason score 7-10. Thirty-three patients on active surveillance for Gleason 3+3 disease were included for retrospective analysis; 125 patients had previous negative systematic transrectal ultrasound (TRUS)-guided biopsies (Supplementary Fig. 1). Patients on active monitoring for Gleason score 7 cancer were therefore excluded from the analysis, as their disease already met our criteria for clinically significant cancer. The Standards of Reporting for MRI-targeted Biopsy Studies (START) were used to describe the study population, the conduct and reporting of the MRI, and the conduct of the biopsy and the Standards of Reporting of Diagnostic Accuracy (STARD) were used to describe and discuss the results [21, 22].
Magnetic resonance imaging
Patients underwent prostate mpMRI at seven different regional hospitals on either 1.5-T (87/158) or 3.0-T (71/158) MRI scanners with surface coil and no endorectal coil. Minimum sequence requirements for inclusion in the study were axial T1-weighted images (T1WIs) of the pelvis, high resolution axial T2-weighted images (T2WIs) of the prostate and diffusion-weighted imaging (DWI), with a minimum of two b values, with a high b value of 800-1,000 . Apparent diffusion coefficient (ADC) maps were calculated for all patients. Eight percent (12/158) of mpMRIs included dynamic contrast-enhanced sequences.
All mpMRIs were performed and first-read at seven regional referral centres by 28 different radiologists. Data on the referring radiologists’ experience in reporting mpMRIs of the prostate were not available. All mpMRI images were prospectively second-read at our own tertiary centre, with external reports available. Second reads were performed by one of two subspecialist uroradiologists with 6 (over 1,500 cases) and 4 years (over 1,000 cases) of experience in reading prostate MRI. In the second-read, T2WI and DWI sequences were evaluated using a Likert scale of tumour probability, based on the Prostate Imaging Reporting and Data System (PI-RADS version 1) structured scoring criteria developed by the ESUR  and a final score was defined by combining all scores for T2WI and DWI sequences as recommended in PI-RADS version 2 . This second read was performed contemporaneously, soon after initial MRI acquisition and before the biopsy procedure, therefore prospectively affecting the biopsy targeting. For patients in whom more than one MRI had been performed, all studies were available to both internal and external readers. The Likert-based scoring system was as follows: 1 = cancer highly unlikely, 2 = cancer unlikely, 3 = equivocal for cancer, 4 = cancer likely, 5 = cancer highly likely. Lesion probability only, rather than quantifiable measures such as ADC value and measurements were recorded. In cases where external reports did not explicitly state a PIRADS/Likert score or location, the reports were reviewed by two research radiologists in consensus, blinded to the clinical details and retrospectively grouped for suspicion of cancer as no suspicion (Likert 1-2), equivocal (Likert 3) or suspicion (Likert 4-5) for cancer.
The BiopseeTM transperineal MRI/TRUS fusion biopsy system version 1 or 2 (Medcom, Darmstadt, Germany) was used for all biopsies. All patients had 18-24 systematic biopsies taken according to the Ginsburg protocol, using a spring-loaded biopsy gun with an 18-gauge needle [5, 24]. In patients with MRI lesions prospectively called by the subspecialist reader, two biopsy cores were taken from each Likert 3-5 lesion before the systematic biopsies. In the systematic biopsy, two biopsy cores were sampled from each of 12 sectors, starting with the anterior sectors. All procedures were undertaken by one of three urologists with several years’ experience of transperineal biopsy using the Biopsee MRI/TRUS fusion biopsy system.
All biopsies were reported by a specialist uropathologist and were reviewed a second time, by another uropathologist, prior to discussion at a multidisciplinary team meeting. Biopsies were reviewed according to the ISUP 2005 recommendations . The final Gleason score was used as data for this study.
The absolute and relative agreements of initial and tertiary centre reads were calculated. Kappa coefficient calculation was used to compare distribution into the different probability subgroups. It was also analysed in how many cases lesions were called in both reads, how many reports called different lesions and how often there was no corresponding lesion at all with the other read being rated as non-suspicious.
Benign histopathology, prostatitis and Gleason score (GS) 3 + 3 = 6 were considered as negative histopathology. GS 7-10 cancer detection rate, all cancer detection rate, and positive (PPV) and negative (NPV) predictive values were calculated for each probability group and both readings, including targeted and systematic biopsy cores in the area that the index lesion was located in. For example: if an index lesion was called in the right anterior, the results for the targeted cores and the systematic cores in the right anterior were used for analysis. If the index lesion was called in the initial read but not targeted after the second read, the systematic cores in the respective area were used for analysis. Fisher’s exact test was used to test for statistically significant difference of cancer proportions.