Factors Impacting Performance and Reproducibility of PI-RADS.

Prostate cancer (PCa) is the most common noncutaneous cancer type among American men. In the last decade, prostate magnetic resonance imaging (MRI) and MRI-guided biopsies have been documented to increase the diagnostic yield of clinically significant PCa, which is generally defined as Gleason Grade >1. With the increased use of prostate MRI, a need for standardization of its acquisition, interpretation, and reporting had emerged, and to address this, American College of Radiology and European Society of Genitourinary Radiology had released prostate imaging and reporting system version 2 (PI-RADSv2) in late 2014. Upon its release, PI-RADSv2 was commonly adopted by practicing radiologists and urologists. Moreover, its prospective use to guide prostate biopsies was proven to increase cancer detection yield in level 1A evidence series. While PI-RADSv2 was documented to bring an improvement in standardization for prostate MRIs, its reproducibility and cancer detection rate performance consistency have been reported to be widely variable. Several factors can impact the variable performance and reproducibility of PIRADS. Among the key factors are quality of prostate MRIs, experience level of the practicing radiologists, subjective aspects of PI-RADS categories which currently do not implement quantitative measures, quality of MRI-guided prostate biopsies, and variation in interpretation of prostate biopsy specimens. In this editorial, we are briefly covering the pressure points related to prostate MRI acquisition and interpretation. Quality of prostate MRI: As it applies to every imaging modality, prostate MRI requires high quality for optimum results for cancer detection. The PI-RADS document aims to achieve the quality assurance by implementing pre-image acquisition patient preparation maneuvers such as applying bowel preparation using enema, use of antispasmodic medicine, or removal of rectal gas via catheters. The use of these methods has shown variable results on image quality in the literature, and currently, the ideal one to achieve a good quality MRI is still unknown. In addition to that, the PI-RADS document provides minimum technical requirements for image acquisition. The adherence level to these minimum standards is again variable, and recently, it is shown that having good quality prostate MRIs is not always associated with strictly following them. There is currently an ongoing discussion on further extending these minimum standards and some research groups are trying to utilize artificial intelligence (AI) for the evaluation of prostate MRI quality and alert the technologists and radiologists on time about this. Experience of practicing radiologists: it is apparent that for each radiologist who wants to read prostate MRIs, a certain learning curve should be covered. However, the exact number of studies to cover this learning period is yet to be determined. Additionally, certification based on stringent criteria and performance metrics is still under discussion to ensure the needed qualifications are owned by practicing radiologists who read prostate MRIs. Subjective aspects of PI-RADS categories: it is evident that in the absence of absolute quantitative criteria, radiologic evaluation is impacted by subjectivity and is prone to significant variation, which is documented to apply to prostate MRIs. In order to cover this issue, the PI-RADS committee has made edits in the PI-RADSv2 document and released PI-RADSv2.1 in March 2019. This new version aims to improve some of the documented subjective parts of PI-RADSv2 by clarifying diffusionweighted imaging criteria for categories 2 versus 3, by further detailing the distinction between positive and negative enhancement on DCE, by revising the criteria for T2 weighted (T2W) scores of 1 versus 2 in transition zone (TZ), and by revision in determination of overall assessment category in TZ. The impact of these edits released in PI-RADSv2.1 is yet to be reported, and future studies are needed to document it. In addition to those factors, the performance variability of PI-RADS can also depend on the referral pattern of the practice and the pretest probability of the referred patient population. It is apparent that PI-RADS-associated cancer detection rates will be different in biopsy naive versus previous negative biopsy versus previous positive biopsy (staging) populations. Potential solutions for the abovementioned challenges related to PI-RADS include the education of radiologists, and technologists with workshops and courses, and development of certification programs with objective metrics. Additionally, recent research revealed that AI may contribute to the solution of these problems. Few groups have developed AI solutions for automated PI-RADS categorization to maintain objectivity in image interpretation. Although the results are promising,