MRI imaging was performed for all patients at 3 Tesla (MAGNETOM PrismaFit, Siemens Healthineers, Erlangen, Germany).

The mpMRI protocol included T2w imaging in 3 orientations, Diffusion Weighted Imaging (DWI), dynamic contrast-enhanced T1-weighted (same orientation as axial T2-weighted and DWI) and pre-contrast T2M (0.7 × 0.7 × 3.0 mm3, 16 echoes with ΔTE of 10.8 ms, TR 5000 ms, acceleration factor = 10; Fig. 3). Image reconstruction parameters are listed in the Supplementary material. Notably, T2M, a research application sequence, “GRAPPATINI”, that accelerates a multi-echo spin-echo sequence is used to achieve acquisition times that are feasible in clinical routine [9]. Particularly in this study, a tenfold acceleration is achieved by combining a twofold GRAPPA acceleration with a fivefold model-based acceleration resulting in an acquisition time of 4:37 min [9, 10]. GRAPPATINI has been evaluated in various body parts including the knee, brain, spine, pancreas, cervix, and prostate [3, 6, 11,12,13,14,15]. The method has been compared to various other methods across organs [13, 16].

Example of region of interest (ROI) measurement of malignant lesion. T2w sections demonstrating the height of measurements for T2M sequence. Region of interest (ROI) measurements were drawn with the polygonal measurement tool an apical (A), midbase (B) and base (C). Suspicious lesions were additionally measured on the slice with its largest diameter (D dotted line). This image represents a confirmed prostate cancer of the left peripheral zone (arrowhead)

All patients included in our study underwent a systematic prostate biopsy. Before the biopsy, rectal swabs and/or urine cultures were performed if clinically indicated. A periprostatic local anesthesia was injected under ultrasound-guidance. We took 12 cores, 6 from each prostate lobe, with a length of 15–22 mm. If a targeted fusion biopsy was performed in addition, 2 cores were taken from each suspicious lesion (defined as PIRADS ≥ 3).

Targeted biopsy was performed with a high-end ultrasound-machine (HiVison, Hitachi Medical Systems, Tokyo, Japan) [17,18,19,20].

Imaging abnormalities were defined as confirmed malignancies based on the interpretation of the biopsy results for the corresponding prostate regions. Additionally, the regions of the corresponding lesions were noted and compared with the imaging findings.

Measurements were performed by an independent assessor trained in the use of PACs measurement tools on a GE Workstation (Universal Viewer, GE, Boston).

For each MRI study, three regions of interest were drawn at representative axial sections: at the level of apex (pTa), mid-base (pTm) and base (pTb). Each ROI was drawn in three different regions per slice, including the right peripheral zone, left peripheral zone and transitional zone. Thus, a minimum of 9 ROIs were drawn per subject.

A differentiation was made between tumor-free (pT) and tumor-containing tissue (pCA). Tumor-containing tissue was defined as PIRADS ≥ 3 and Gleason Score ≠ 0. Prostate tissue without suspicious lesions (as determined by mpMRI and biopsy report) was measured bilaterally in the peripheral zone, unless the zone was completely affected such that a representative ROI could not be set. In this case, only malignant lesion measurements were recorded for this region. The slice with the largest tumor extent was also included to mark a representative ROI for the tumor-tissue (Fig. 3). Image measurements were performed in a structured manner suitable to inexperienced readers.

MRIs were originally marked in the T2w reconstruction of the T2M sequence. All corresponding measurements were mapped to other sequences by table position and synchronization of image stacks. If the automatic matching failed, corrections to the mapping between sequences could be made manually. ROIs were always copied from the first sequence measured to all other corresponding sequences. Therefore, identical ROI shapes and sizes were ensured to match measurements between sequences as closely as possible. Mean and minimum transverse relaxation time (T2) values in each ROI were recorded.

To ensure data quality, incomplete data, studies of poor image quality and inconsistent data was rigorously sorted out.

In addition, the ROIs were marked by an independent reader. ROIs were drawn leaving a margin to the edge of each structure to avoid including other tissue. ROIs were made as large as possible to get a representative average value per region. ROIs were not placed in areas containing artifacts.

All measurements and the markings of the ROIs were reviewed by an expert reader (AMB) with 6 years of experience in reading prostate MRI. ROIs were however only redrawn if the desired prostate region was not accurately included, in order to test the suitability of the approach for an inexperienced reader.

Statistical analysis was conducted using commercially available statistical software, including SPSS, version 21.0 (© 1989–2012, IBM, Armonk, NY, USA; MedCalc Software bv, Ostend, Belgium; RStudio, PBC, Boston, MA, USA). The normal distribution was determined using the Shapiro-Wilk test. Categorical variables are presented as percentages, continuous variables as mean ± standard deviation or median and interquartile ranges if the distribution is not normal. The non-parametric were assessed using the Mann-Whitney-U-test. To determine any correlation, the Spearman test was performed. A receiver operating characteristic (ROC) was used to identify the cut-off value that achieved the best balance between sensitivity and specificity. Using a linear mixed-effects model (LMM) fit by Restricted Maximum Likelihood (REML), we assessed the differences in transverse relaxation times (T2) between cancerous and healthy prostate tissues. The model included fixed effects for tissue type (cancerous vs. healthy) and random intercepts for subjects to account for inter-individual variability. P-values less than 0.05 were considered statistically significant.