MRI Evaluation

Two readers had full access to all images on a standard radiological workstation (SIENET MV 1000, Siemens, Erlangen, Germany). They were blinded to the respective PSA levels, GS and histopathological classification of malignant tumors (TNM classification) but were aware that all subjects had biopsy-proven cancer. Reader A (M.R.-Z.), a radiologist with 6 years’ experience in general MRI diagnostics, had already read 70 endorectal prostate 3-T MRI cases while reader B (J.G.), a radiology resident with 2 years of general MRI experience, had read 160 such cases. For each patient, both MR images sets were anonymized and randomized into two equally sized groups. Images of the same patient were read in two separate sessions at least 6 weeks apart to reduce rater bias.

Image quality was evaluated by nine items following established criteria originally described by Heijmink et al.¹⁰: Five of them (a–e) focused on cancer localization and involved (a) discrimination between transitional and peripheral zone (TZ/PZ), (b) visibility of the PZ itself, (c) visibility of the TZ itself, (d) visibility of the lesion, (e) visualization of the internal architecture of the TZ. Four items (f–i) were related to tumor staging and included (f) delineation of the prostatic capsule, (g) visualization of the neurovascular bundle as well as (h) visualization of the rectoprostatic angle. For the last staging item, (i) presence of minimal capsular penetration, reader confidence was only assessed in the positive case. Reader confidence was generally rated on a five-point scale as very poor (1), poor (2), moderate (3), good (4), and very good (5).

The presence of image artifacts caused by motion or other sources (aliasing, susceptibility) was also evaluated on a five-point scale with image quality being severely affected (1), moderately affected (2), acceptable (3), hardly affected (4), or not affected (5). Lesions were classified according to PIRADS version 1. Both readers rated the presence of clinically significant disease as either highly unlikely (1), unlikely (2), unclear (3), likely (4), or highly likely (5) using PIRADS criteria for T2WI and DWI⁴. Circular regions of interest (ROI) were drawn within the boundaries of all cancer-suspicious lesions and the mean ADC was measured. Tumor localization was assigned by using a standardized 27-sector prostate MRI reporting scheme following the recommendations of the European Consensus Meeting in 2010¹¹.