PI-RADS and Likert scales for structured reporting in multiparametric MR imaging of the prostate

Abstract

Multiparametric MRI (mpMRI) plays a critical role in the detection, staging and risk stratification of prostate cancer (PCa). There are two widely accepted structured reporting systems used for interpretation of mpMRI of the prostate - PI-RADS v2.1 and Likert. Both these systems demonstrate good diagnostic performance with high cancer detection rates however have key conceptual differences. In this commentary, the authors highlight the individual strengths and areas of potential improvement as well as emphasize the need for continued clinical validation for these interpreting and reporting systems.

Introduction

Multiparametric MRI (mpMRI) plays a vital role in the detection, staging and risk stratification of prostate cancer (PCa). The term multiparametric refers to the integration of information obtained from different MRI sequences including T ₂-weighted (T2W), diffusion-weighted (DWI) and dynamic contrast-enhanced (DCE) MR imaging.

The imaging protocol, acquisition and diagnostic interpretation are critical elements for a good-quality mpMRI. The radiology report, however, is the single most important deliverable for proper communication with referring physicians and patients. Historically, imaging reports were crafted with a narrative, free text approach frequently resulting in lack of clarity, missing expected elements, or a gap between the intended and the received message. More recently, structured reporting has been proposed as a strategy to ensure adherence to the expected elements of a good report such as clarity, accuracy, concision, completeness and standardization. Potential advantages of structured reports include a checklist and systematic approach that decrease cognitive biases (e.g. “satisfaction of search”) and speech recognition errors, facilitates comparison with prior studies, billing compliance and data mining for quality initiatives and research. ^1,2

Two-structured reporting systems widely used for the interpretation of mpMRI of the prostate are the prostate imaging- reporting and data system (PI-RADS) and the Likert scale.

PI-RADS (Prostate imaging-reporting and data system)

The origin of PI-RADS dates back to 2012, when the European Society of Urogenital Radiology proposed a standardized reporting system for mpMRI examinations ³ , which later became known as the PI-RADS version 1. Despite the limitations of its initial iteration, it contributed to promote the discussion of the importance of mpMRI report standardization. Major changes were introduced in collaboration with the American College of Radiology by version 2 in 2015, ⁴ including a refined zone-dependent scoring system and established risk assessment score categories. This substantially improved version resulted in broader adoption and clinical adoption. As stated by the PI-RADS Steering Committee, the system is meant to be a “living” document that will evolve as clinical experience and scientific data accrue. More recently, v.2.1 ⁵ proposed minor modifications including a slightly updated scoring system preserving the overall framework introduced by version 2.

In brief, it provides zone-specific definitions of 1–5 scores for T2W and DWI as well as DCE positivity, elements that are then used to determine an overall PI-RADS risk category. These also range from 1 (clinically significant PCa is highly unlikely to be present) to 5 (clinically significant PCa is highly likely to be present).

Several studies have provided clinical validation of PI-RADS as a tool to serve its original purpose – stratify the likelihood of the presence of clinically significant PCa. Although most literature concerns version 2, ⁶ a recent prospective study by Walker et al ⁷ assessed cancer detection rates stratified by PI-RADS v.2.1 category using targeted biopsies as the reference standard. They found clinically significant cancers in 0%, 6%, 15%, 44% and 80% of the category 1, 2, 3, 4 and 5 lesions, respectively. Notwithstanding, certain limitations continue to be identified with broader utilization of some of the consensus-based PI-RADS changes. Areas for improvement of v.2.1 highlighted by Purysko et al ⁸ in a recent critical review included the lack of validation of some newly introduced criteria, the vague language used in some of the criteria, the need for systematic evaluation and reporting of the background prostate appearance that can negatively affect cancer detection, a new category for lesions that do not fit into the PCa typical appearance, inclusion of quantitative parameters beyond lesion size to evaluate lesion aggressiveness, and standardized assessment of the risk of extraprostatic extension.

In addition to image interpretation criteria, the PI-RADS document includes sections with recommended clinical considerations (e.g. patient preparation), minimum technical parameters (e.g. MRI scanner field strength, coil options, slice thickness, spatial resolution), and reporting elements beyond lesion appearance (e.g. how to measure the prostate volume and lesion size). Although staging is discussed as a separate section, no standardized or quantitative reporting approach is suggested. ⁵

Likert scale

Similar to PI-RADS, the Likert scale is based on a 5-point score meant to reflect the probability of PCa being present. Although image interpretation incorporates the findings considered suspicious by PI-RADS, no specific criteria are provided to define each risk category. Therefore, to some extent, much is left at the discretion of the radiologist and it is considered a gestalt assessment. As such, it can be tweaked according to the experience of the reader as well as parameters beyond imaging appearance, such as age, PSA, PSA density and previous biopsy results. Different from PI-RADS, given its relatively free form, there is not a unified document providing specific guidelines such as the concept of zone-specific dominant MRI pulse sequence. Moreover, a lesion can be classified as risk category 5 regardless of size, whereas PI-RADS requires a minimum size of 15 mm. While PI-RADS has been proposed specifically for treatment-naïve patients, the flexibility resulting from the lack of specific criteria in the Likert scale also allows the use of Likert scales in other settings such as suspected disease recurrence after radiation or focal therapy. Noteworthy is the preference of Likert over PI-RADS by the UK-based National Institute for Health and Care Excellence. ⁹

Similar to PI-RADS, numerous studies have shown the predictive value of the Likert approach to assessing the likelihood of clinically significant PCa. Shin et al ¹⁰ assessed cancer detection rates stratified by prospectively assigned Likert scores using targeted biopsies as the reference standard. They found clinically significant PCa in 4%, 4%, 12%, 33% and 48% of the category 1, 2, 3, 4 and 5 lesions, respectively. A separate Likert score has also been proposed by some groups to provide the likelihood of extra prostatic extension. ¹¹

Discussion

PI-RADS and Likert scoring systems share similar lexicons and good cancer predictive values. Different than PI-RADS, assessment with a Likert scale does not follow specific criteria and can be used in settings other than initial cancer detection, including suspected disease recurrence after radiation of focal therapy.

One of the challenges in comparing both approaches is the “living” nature of PI-RADS with new iterations that may preclude extrapolating validation data from previous to newer formats. As of the writing of this manuscript, no head-to-head comparison of the cancer detection rates between PI-RADS v.2.1 and Likert assessment has been published. Most of the literature is focused on comparing v.2.1 against v.2. However, studies comparing v.2 and Likert scales demonstrated similar diagnostic performance for tumour localization ¹² – some with slightly higher accuracy for Likert scales in the hands of experienced operators ¹³ mostly related to improved specificity. While the relatively objective criteria for PI-RADS v.2.1 could improve the agreement in interpretation among readers, particularly those of varying experience across different institutions and practice setting, direct comparisons are lacking. Studies comparing inter-reader agreement using Likert scales and PI-RADS v.2 have failed to demonstrate a substantial difference between the two approaches. ¹⁴ A combination approach might include using the grading criteria in PI-RADS during the initial learning stages of interpreting mpMRI of the prostate and then refining interpretation using a Likert-like scoring system as one gains more experience.

Current tools aim to convey the likelihood of clinically significant prostate cancer being present, reflecting the major role of mpMRI in patients being evaluated before biopsies. The utility of mpMRI in other scenarios – for example, active surveillance, preoperative and focal therapy planning, suspected disease recurrence after radiation or focal therapy – is not currently addressed by the PI-RADS. This is one of the main advantages of the flexible nature of the Likert approach, however its diagnostic performance in many of these specific scenarios has not been rigorously explored. Future versions of these system may address this through the incorporation of new fields, such as markers of disease progression (e.g. apparent diffusion coefficient values), a systematic assessment of local staging, and the tumour relationship with certain structures of surgical interest. Continued clinical validation, quality metrics, and input from urology and radiation oncology representatives as well as integration of new nuclear medicine studies will remain cornerstones of future developments.

In conclusion, despite the simplified scoring schema proposed by PI-RADS and Likert scales, reporting mpMRI of the prostate is a complex task. The decision to use one system versus the other may depend on the practice setting, the preference among referring urologists and the comfort level amongst interpreting radiologists with each approach. Both structured reporting tools, however, facilitate communication with referring physicians and patients while showing clinically useful positive predictive values and fair inter-reader agreement. Both systems can also be used for internal quality control to continuously monitor the correlation between imaging and biopsy results. Likert has the advantage of being flexible and tends to perform better with experienced radiologists. PI-RADS is a more comprehensive document, by design expected to undergo updates allowing to incorporate refinements that will require continuous validation by real-world clinical data.