mpMRI is a powerful technology that provides insight into which patients may harbor clinically concerning tumors. As with all risk assessment methodologies, mpMRI has strengths and limitations. Comparing mpMRI performance across studies is difficult due to variations in study design, equipment (e.g., magnet strength), inconsistent incorporation of the three individual phases (T2 weighted, diffusion, and dynamic contrast-enhanced), as well as reader expertise. Similarly, biopsy methodology varies greatly as do the types of biopsy samples evaluated (4–7). These ‘mixed use’ inclusion criteria impact disease prevalence, which affects both positive and negative predictive values (8). Inconsistencies in the definition of HGPC across mpMRI studies can also skew results. While many use the HGPC definition provided by the International Society of Urological Pathology (ISUP) of ≥GG2, other definitions include ISUP grade ≥GG3, core length, positive cores percentage or some combination (9–15). How HGPC definitions and biopsy methodology vary and impact HGPC incidence has been summarized previously (16).

The mpMRI HGPC detection metrics depend upon the selected biopsy method (transrectal ultrasound scan (TRUS)-guided, targeted, etc.) ( Table 1 ). Studies still vacillate on the most appropriate use of TRUS-guided biopsy versus targeted biopsy (7, 17, 23). Though often debated, it is accepted that mpMRI imaging provides equal or superior detection over TRUS-guided biopsy (18, 24). NCCN guidelines call for the ‘routine’ use of image-guided biopsy, but also highlight the potential value of a systemic biopsy in addition to image guided biopsy (1). Recently updated American Urology Association (AUA) guidelines call for a TRUS-guided vs targeted biopsy based on Prostate Imaging Reporting & Data System (PI-RADS) scores (25).

The most widely acknowledged factors that impact mpMRI-specific tumor detection are reader interpretation, reporting, and biopsy methodology (26).

PI-RADS scores are a group-based risk assessment that provides the probability, not guarantee, of a biopsy result (10). A key foundation of PI-RADS is a critical mpMRI limitation, reader-dependent variation. Although high concordance (78%) is claimed in studies that utilize ‘expert’ readers, the definition of ‘expert’ raises concerns surrounding the use of generalizations in clinical practice (18, 27, 28). A recent study spanning 26 sites highlighted that generalization of mpMRI led to varying positive predictive values (PPV) between sites due to reader variation, poor targeting, and inconsistent disease prevalence (29). Thus, there is a need for mpMRI standardization.

Analysis of studies with mpMRI negative results (PI-RADS <3) demonstrate that mpMRI NPV is quite variable (1, 4, 5, 7, 9–15, 26). Of particular importance, Otti et al. found that 17% of men with ‘normal’ mpMRI readings had palpable disease (26). Chung et al. examined ‘invisible’ tumors (PI-RADS <3) with biopsy and RP, finding at biopsy that 24% were ≥GG2 and 6.6% were ≥GG4 (41). Finally, men initially classified as mpMRI negative at 2.4 years post-imaging identified the false negative rate to be 23% for HGPC (42).

Smaller cohort studies frequently present NPV and PPV, often combining different biopsy sample types (biopsy naive, prior negative, active surveillance) without sufficient regard to the impact on disease prevalence (4–8, 15). Although some studies conduct sub-analysis to examine metrics separately based on biopsy type, many do not (5, 9, 16). In 2017, Moldovan et al. (8) performed a thorough review and documented the impact of clinical heterogeneity and disease prevalence on NPV. Therefore, with the exception of the landmark PROMIS trial (9), only studies published since Moldovan et al. (8) are reviewed here ( Figure 1C ). Unfortunately, if a study did not biopsy mpMRI negative results (<PIRADS 3) it could not be included in the analysis (including the otherwise well-executed PRECISION trial (18)).

Since the Moldovan review (including PROMIS), there have been 2,035 mpMRI negative cases (PI-RADS 1/2s) across these studies. As shown in Figure 1C , a number of studies with mixed biopsy types (4, 6, 7, 15, 16), small sample sizes, and/or solitary experienced readers trend towards higher NPVs (4, 5, 12, 43). The majority (68%) of these studies resulted in NPV <90% with the landmark PROMIS trial observing 76% NPV for ≥GG2. Only 32% (N=653) of the data generated NPVs ≥90%. In fact, the best study for initial biopsy with NPV above 90%, comprising 47% (N=309) of the cases with NPV >90%, was the 4M study (96%) (44). This prospective trial employed trained radiologists, centralized image review and consensus assessment resulting in an atypical low number of PI-RADS 3 (6%). Due to the expertise in this study, the generalizability of the results to routine practice is a legitimate question. Indeed, the authors believe a key limitation of their data is its performance reproducibly outside their ‘expert’ sites (44).

In addition to sample size and range of mpMRI protocols, biopsy methodology also limits mpMRI performance metrics. In particular, the widely used TRUS-guided biopsy provides an imperfect window into mpMRI performance. Studies that utilize more holistic assessment methods such as template biopsy mapping, saturation biopsy, or whole mount RP offer the most comprehensive pathologic ground truth for assessing performance metrics (9, 10, 14, 32, 45, 46). The landmark PROMIS study arguably provides the most comprehensive assessment of mpMRI metrics because it employed template biopsies for all PI-RADS categories (9). Although the 1.5 Tesla (T) magnet strength in the PROMIS study was less than the often employed 3T magnet strength, much of the essential mpMRI literature utilizes a mix of magnet strengths (both 1.5T and 3.0T) (18, 20, 24). Furthermore, some studies have found no difference in PPV or NPV when comparing 1.5T or 3T generated data (11, 29).

The PROMIS study combined high quality and standardized MRI, in-depth reporting, dedicated and experienced urologic radiologists, centralized reader training, TRUS biopsies, and high-quality targeted mapping biopsies every 5 mm. The primary definition of clinically significant cancer was defined as ≥GG3 or cancer core length ≥6 mm, but two additional cancer definitions were also measured (1) ≥GG2 or (2) cancer core length ≥4 mm. As such, each definition resulted in different performance metrics. The primary definition of ≥GG3 or cancer core length ≥6 mm had a PPV of 51% and NPV of 89%. The more widely used definition, ≥GG2, had a PPV of 65% and an NPV of 76% (9). Other studies utilizing mapping biopsies have demonstrated similar NPV for ≥GG2 or related definitions. Simmons et al. noted an NPV of 68.6 for ≥GG2 and/or tumor length of ≥4 mm (14). Mortezavi et al. used template saturation biopsy to measure mpMRI performance metrics for ≥GG2 and noted overall NPVs of 74.2% and 68.5% for the saturation biopsy and targeted fusion biopsy, respectively. The authors conducted a subgroup analysis according to biopsy type (naïve, prior negative biopsy, or positive biopsy), resulting in changed predictive values due to changes in disease prevalence (16).

Considerations for adding biomarker information to mpMRI has clinical benefit, with the reduction of potential side effects, ease of use, cost, and ability for generalizations. Evaluating biomarker data can be challenging because there is no one-size-fits-all approach and biomarkers will fit each case differently. The advantages are that biomarkers for early detection of PCa can be analyzed in urine, blood, or post-needle biopsy samples (47–50). One test already has a home collection kit further simplifying sample collection even further (51). When considering deferred HGPC as a potential side effect of early detection risk assessment, the percentage of deferred HGPC depends upon the biomarker threshold utilized.

Biomarker durability and generalizability vary due to biomarker study design ranging from observational and retrospective to prospective clinical trials. In prospective studies for the intended use population, biomarker durability and clinical benefit are evident (52). However in retrospective analyses, when cohorts do not represent the intended population, it is difficult to assess the actual performance or generalizability of the biomarker (48, 49, 53). Furthermore, most biomarkers include clinical information combined with genomic data making it difficult to determine the specific value the unique test components provide when not supported by the clinical features (48, 49, 53). However, stand-alone biomarker assays do exist that do not integrate clinical features and have been used in studies with prospective trial design, tailored to intended-to-use population (47, 54).

Non-invasive tests such as PSA density, risk calculators, genomic testing, and commercial biomarker tests all appear to provide some degree of enhanced clinical risk assessment when appropriately layered into a clinical pathway with mpMRI (6, 13, 15, 37, 45). PSA density (PSAD), for example, is often found to be complementary to mpMRI (6, 12, 13, 37, 55, 56). In one study, 63% of men with abnormal mpMRI results and HGPC had a PSAD of ≥0.15 ng/mL, compared to 29% of men with normal mpMRI readings and benign or insignificant PCa who had PSAD of ≤0.15 ng/mL (26). Another study observed that PSAD >0.1 ng/mL provides complementary risk assessment value to mpMRI for HGPC defined as ≥GG2. Furthermore, an increased frequency of PI-RADS 4 and 5 lesions with HGPC was observed when PSAD was above the 0.15 ng/mL threshold (56). Conversely, other research has shown that when PI-RADS is low-risk (≤2) a PSAD <0.15 ng/mL may improve the NPV (12). In a mixed group of biopsy naïve men (36%) and men with a prior negative biopsy and a low-risk mpMRI (PI-RADS <3), mpMRI NPV improved from 82% to 90% with a PSAD <0.15 ng/mL. In solely biopsy naive men with negative mpMRI and PSAD <0.15 ng/mL, the NPV improved from 71% to 80% (6). Moreover, the mpMRI NPV for prostate biopsy naïve men with PI-RADS 1 or 2 was 80% that increased to 91% when PSAD <0.1 ng/mL was included in a prospective study (57). Other studies have specifically focused on the clinical benefit of combining PSAD and mpMRI for PI-RADS 3 scores with HGPC detection across PI-RADS scores gated by PSADs of <0.15, 0.15-0.29, or ≥0.3 ng/mL. The highest HGPC detection rate (97%) was found when PI-RADS 4 and 5 had a PSAD of ≥0.3 ng/mL. In contrast, the lowest HGPC detection rate (0%) was found in PI-RADS 1 and 2 scores with PSAD <0.15 ng/mL and no HGPC was detected in PI-RADS 3 with PSAD <0.15 ng/mL (N=6). Although the numbers were small, the authors conclude that men with a PI-RADS score ≤3 and PSAD <0.15 ng/mL may be able to avoid unnecessary biopsies (37).

Beyond individual biomarkers, the next step is to assess how integrating multiple biomarkers might complement mpMRI. The current landscape is expanding PSAD to numerous clinical features and mpMRI (15), with clinical feature calculators being integrated to mpMRI interpretation. Initial results of these calculators are highlighting that some clinical risk features are more important to NPV than others (15).

Although published studies have limitations, the biomarker/mpMRI literature suggests that commercial biomarker tests and mpMRI capture independent information that can provide a synergistic benefit (58). Several biomarker studies have demonstrated a correlation between increasing PI-RADS scores and biomarker scores, and suggest that performance for HGPC detection improves when mpMRI and biomarkers are combined (49, 59). An observational study concluded that the SelectMDx biomarker had independent information that improved the PI-RADS area under the curve (AUC). However, the picture was incomplete since PI-RADS <3 biopsies were not included (60). Another biomarker, Proclarix, combines clinical features with thrombospondin-1 and cathepsin, which when combined with mpMRI, the biomarker demonstrated improved performance and increasing mpMRI NPV up to 6% in men with PI-RADS <3 (61). The actual biomarker value versus the clinical feature performance in these studies is unclear as the initial and repeat biopsies were mixed. Moreover, the HGPC prevalence in one study was lower than in an intended-to-use population. The Myprostatescore biomarker improved NPV in men with PI-RADS 3 and performed better than PSA density. However, 57% of the cohort had a prior negative biopsy, lowering the HGPC prevalence and likely artificially inflating NPV (48). Preliminary data with the ExoDx Prostate biomarker and mpMRI has shown an association between rising ExoDx Prostate IntelliScore (EPI) scores and PI-RADS, demonstrating the benefit of modeling EPI and mpMRI together (59). Specifically, the PCA3 biomarker and mpMRI combination in men going for initial biopsy had an improved performance over mpMRI alone (62). A superior performance in AUC was observed when combining the 4K biomarker with mpMRI, each providing independent and complementary information (53). Similarly, a complementary performance was detected when combining PSA density, the prostate health index (PHI) biomarker, and mpMRI for men with a prior negative biopsy (63). Finally, a prospective non-inferiority trial demonstrated that a clinical workflow combining the Stockholm biomarker and mpMRI detected more HGPC and fewer low-grade cancers (64).

The cost of integrating biomarkers with mpMRI is variable based on the biomarker. mpMRI is an expensive procedure with a median cost of $4396 (interquartile range $2,784-$7,127) for MRI-guided biopsy, increasing to $5,832 when anesthesia is used (65). The cost of mpMRI/biomarker care will likely depend on how the two modalities are implemented. In risk assessment for early PCa detection, multiple clinical combinations of commercial biomarkers and mpMRI have been presented (66), with initial data suggesting biomarkers placed before mpMRI will provide the most clinical benefit (50). Additional cost savings may also result from reducing un-needed biopsies and reducing the use of mpMRI for men at low risk for finding HGPC (67).