Population Net Benefit of Prostate MRI with High Spatiotemporal Resolution Contrast-Enhanced Imaging: A Decision Curve Analysis

Abstract

Background:

The value of dynamic contrast-enhanced (DCE) sequences in prostate MRI compared with non-contrast MRI is controversial.

Purpose:

To evaluate the population net benefit of risk stratification using DCE-MRI for detection of high-grade prostate cancer (HGPCA), with or without high spatiotemporal resolution DCE imaging.

Study Type:

Decision curve analysis

Population:

Previously published patient studies on MRI for HGPCA detection, one using DCE with golden-angle radial sparse parallel (GRASP) images and the other using standard DCE-MRI.

Field Strength/Sequence:

GRASP or standard DCE-MRI at 3T.

Assessment:

Each study reported the proportion of lesions with HGPCA in each PI-RADS version 2 category (1–5), before and after reclassification of peripheral zone lesions from PI-RADS 3 to 4 based on contrast-enhanced images. This additional risk stratifying information was translated to population net benefit, when biopsy was hypothetically performed for: all lesions, no lesions, PI-RADS ≥3 (using NC-MRI) and PI-RADS ≥4 on DCE.

Statistical Tests:

Decision curve analysis was performed for both GRASP and standard DCE-MRI data, translating the avoidance of unnecessary biopsies and detection of HGPCA to population net benefit. We standardized net benefit values for HGPCA prevalence and graphically summarized the comparative net benefit of biopsy strategies.

Results:

For a clinically relevant range of risk thresholds for HGPCA (>11%), GRASP DCE-MRI with biopsy of PI-RADS ≥4 lesions provided the highest net benefit, while biopsy of PI-RADS ≥3 lesions provided highest net benefit at low personal risk thresholds (2–11%). In the same range of risk thresholds using standard DCE-MRI, the optimal strategy was biopsy for all lesions (0–15% risk threshold) or PI-RADS ≥3 on NC-MRI (16–33% risk threshold).

Data Conclusion:

GRASP DCE-MRI may potentially enable biopsy of PI-RADS ≥4 lesions, providing relatively preserved detection of HGPCA and avoidance of unnecessary biopsies compared with biopsy of all PI-RADS ≥3 lesions.

INTRODUCTION

Prostate cancer is the most commonly diagnosed non-cutaneous malignancy in men and one of the most common causes of cancer-related mortality.(1,2) Most prostate cancers detected using screening methods (digital rectal exam [DRE] or serum prostate specific antigen [PSA]) are low grade (Gleason score 6), with long-term (10–15 year) prostate cancer-specific mortality ranging from 0.2–0.8%.(3,4) Treatment for indolent prostate tumors can cause significant patient morbidity, including side effects of urinary and sexual dysfunction, and therefore reduced quality of life without increase in life expectancy.(5) Initial treatment of high grade prostate cancer (HGPCA), rather than all prostate cancers, has therefore been recommended in order to improve patient health outcomes.(6,7)

Prostate MRI is often used as supplemental assessment for clinically significant prostate cancer, as it has proven more sensitive and specific than clinical methods alone for the detection of high grade prostate cancer (HGPCA)(8). Targeted biopsy of lesions identified on MRI is now a widely accepted practice; however, biopsy complications such as infection, hemorrhage, and pain can still occur in up to 10% of patients.(9) The competing interests of detection of HGPCA and avoidance of biopsy-related morbidity are weighed differently by individual patients,(1) and there may be a range of probability thresholds for cancer detection at which patients would opt for prostate MRI results to guide the decision for biopsy. Though personal risk thresholds vary by patient race (10) as well as age and other demographic factors,(11) consideration of biopsy is generally recommended at a risk estimate no lower than 5% because of the rate of biopsy complications.(12,13)

Dynamic contrast enhancement (DCE) is a component of standard multi-parametric prostate MRI and typically involves multiple axial images of the prostate obtained after intravenous gadolinium injection. In the Prostate Imaging Reporting and Data System version 2 (PI-RADS v2), focal early enhancement in the peripheral zone on DCE images results in upgrade of a PI-RADS 3 lesion (as already detected on diffusion-weighted imaging) to PI-RADS 4.(14) PI-RADS v2 recommendations include DCE for this risk-stratifying information, which could potentially serve to divert patients at lower risk from unnecessary biopsies. However, the value of DCE imaging as a standard component of prostate MRI has been questioned when compared with lesion detection on non-contrast (NC)-MRI coupled with biopsy of all lesions with a score of at least PI-RADS 3.(15) The comparative benefit of DCE images may also be of interest to patients concerned about receiving gadolinium, for example due to deposition in native tissues.(16)

Innovative DCE techniques may improve risk-stratification through increased accuracy of lesion characterization. High spatiotemporal resolution golden-angle radial sparse parallel (GRASP) imaging offers advantages over standard 3D-T1-weighted fat suppressed imaging.(17,18) The fusion of compressed sensing and parallel imaging into a joint imaging method allows increased acceleration of image acquisition and optimization of both spatial and temporal resolution to characterize potential prostatic tumors, and GRASP has been shown to provide better anatomic clarity, image sharpness, inter-reader correlation for lesion size, and lesion conspicuity.(17)

The purpose of this study was to evaluate the population net benefit of risk stratification using DCE-MRI for detection of high-grade prostate cancer (HGPCA), with or without high spatiotemporal resolution DCE imaging. We used a decision curve analysis (19) to summarize the population net benefit of each MRI technique across a range of risk thresholds at which decision makers would choose to undergo MRI. We hypothesized that a comparative net benefit would be seen with the reported performance of GRASP-DCE MRI, compared with NC-MRI for detection of HGPCA across a clinically relevant range of risk thresholds, while the reported performance of routine DCE may show minimal comparative net benefit versus NC-MRI for the same probability thresholds.

MATERIALS AND METHODS

This analysis applied data from prior publications only and was therefore compliant with the Health Insurance Portability and Accountability Act (HIPAA) without requirement for review by our Institutional Review Board.

Overview of Decision Curve Analysis and Risk Models Compared

Decision curve analysis, first described by Vickers et al in 2006, is a graphical statistical method that plots a population’s net benefit (on the y axis) from a diagnostic test over a variety of probabilities for the test detecting true disease (on the x axis).(19) Decision curve analysis is useful in situations where there is no consensus on when to perform a diagnostic test in a population and decision makers weigh the benefits and risks of performing the diagnostic test differently; this may be because of variable weighting of consequences of the diagnostic test (e.g. risk of infection from biopsy). As a result, decision makers will have variable probabilities of disease detection above which they would pursue the diagnostic test and below which they would not pursue the diagnostic test; this value is termed the probability threshold. The probability threshold is informative of how a decision maker weighs the relative benefit of appropriate treatment compared to the benefit of avoiding unnecessary treatment.(20)

The present study applies to the clinical scenario of a urologist or patient (the decision maker) deciding upon whether or not to undergo a prostate MRI, which will then determine the need for biopsy based on the presence of lesions suspicious for high grade tumor. The decision maker’s probability threshold would be the risk for eventually finding HGPCA above which they would always want to do the imaging test (with subsequent biopsy of suspicious lesions). Assuming a patient comes from the population used to calculate a decision curve, a decision maker can select a risk-stratifying approach by finding their probability threshold and then identifying the approach with the highest net benefit at that threshold.(20) Clinically, consideration of biopsy for HGPCA is recommended when the probability threshold is no lower than 3–5%, due to potential complications from biopsy (12,13). Thus, undergoing prostate MRI may be most relevant to patients who have an estimated pretest risk of HGPCA in the range of 5–20%, where the risk of unnecessary biopsy is weighed against detection of HGPCA.(21,22)

The decision curve method allows for examination of multiple thresholds for biopsy simultaneously. Each set of decision curves employs control-risk models of “biopsy all” and “biopsy none”, with a variable number of additional decision risk models (such as MRI with or without contrast) added. The “biopsy all” and “biopsy none” curves intersect at the disease prevalence. The upper limit of the net benefit is the disease prevalence, while there is no defined boundary for the lowest possible net benefit (−∞). Dividing net benefit by disease prevalence gives a standardized net benefit, with a maximum value of 1, allowing comparison of how the risk model compares to this maximum value of the disease prevalence. At a given standardized net benefit value (e.g. 0.4), the risk model would offer the same standardized net benefit as a policy that intervened for that proportion of actual cases (e.g. 40%) and no controls. We compared the net benefits and standardized net benefits of four distinct PI-RADS score-based risk models for data using two patient-based series:

• 1)Biopsy no lesions
• 2)Biopsy all lesions
• 3)Biopsy PI-RADS ≥3 lesions (as determined using non-contrast sequences)
• 4)Biopsy PI-RADS ≥4 lesions with the addition of DCE imaging.

HGPCA was defined as a Gleason score ≥7 using the reference standards defined above. All reported lesions were included in the analysis. Because of substantial differences in study design, comparisons were made among the biopsy strategies within each study, and not directly between the two techniques. Statistical analysis for decision curve analysis was performed using the dca package in R software, version 3.4.2.(19)

Overview of Analyzed Studies

Data for decision curve analysis were applied from the two large, contemporary studies on the diagnostic test accuracy of prostate MRI using PI-RADS v2. Both studies examined 1) how frequently PI-RADS 3 peripheral zone lesions, as determined on non-contrast MRI sequences, were reclassified as PI-RADS 4 lesions when DCE-images were added for interpretation, and 2) how the reclassification changed the rate of HGPCA in each PI-RADS category. High grade cancer was defined as presence of Gleason ≥3+4 adenocarcinoma. Both studies were also performed at large academic institutions using 3T prostate MRI; the details of these studies are included in Table 1.(23,24)

Table 1.

Characteristics for studies utilized in the decision curve analysis.

Prevalence of High Grade Prostate Cancer in Study Population	Date	Number of patients and lesions (n)	Mean age (years), +/− standard deviation	Reference Standard	Pre-contrast MRI Protocol Parameters	Dynamic Contrast- Enhanced (DCE) MRI Parameters
Average prevalence (Rosenkrantz et al(24))	2017 (patients from 2013-2015)	343 consecutive patients who underwent MRI and subsequent MRI/ultrasound fusion biopsy of 343 lesions detected on MRI	64 +/− 8	MRI-ultrasound fusion targeted biopsy	-3T scanners (Siemens Healthineers) -T2-weighted images -Diffusion-weighted images (acquired up to b1000, calculated b 1500) -Apparent diffusion coefficient maps based on b-values up to 1000	Continuously acquired golden angle radial acquisition with parallel imaging and compressed sensing reconstruction (GRASP) imaging: -Temporal resolution: 2.3 sec -FOV: 240 × 240 mm -Acquisition Matrix: 224 × 224 -Repetition time: 4.1 msec -Echo time: 1.89 msec -Flip angle: 16° -Section thickness: 3 mm -Image Reconstruction Matrix: not given -Reconstruction voxel imaging resolution not given -Spatial resolution (derived): 240/224 × 240/224 × 3 pixels -Acquisition time: 5 min 38 sec
High prevalence (Greer et al(23))	2017 (patients from 2012-2015)	163 patients with 651 lesions detected on MRI: −110 consecutive cases underwent prostatectomy −53 controls had negative MRI and negative biopsy.	62 (standard deviation not provided)	Whole-mount radical prostatectomy	-3T scanners (Philips Healthcare) -T2-weighted images -Diffusion-weighted images (up to 2000) -Apparent diffusion coefficient maps based on b-values up to 750	Routine DCE imaging: -Temporal resolution: not given -FOV: 262 × 262 mm -Acquisition Matrix: 188 × 96 -Repetition time: 3.7 msec -Echo time: 2.3 msec -Flip angle: 8.5° -Section thickness: 3 mm -Image Reconstruction Matrix: 256 × 256 pixels -Reconstruction voxel imaging resolution: 1.02 × 1.02 × 3.00 mm/pixel -Spatial resolution (derived): 262/188 × 262/96 × 3 pixels -Acquisition time: 5 min 16 sec

The first study applied data from a cohort of 343 patients with 343 lesions from 2013–2015 using MRI with GRASP DCE-MRI, with MRI-ultrasound fusion biopsy serving as the reference standard.(24) All patients in this study were referred for prostate MRI to localize suspicious lesions before all biopsies with MRI-ultrasound fusion biopsy: factors such as patient and physician preference, MR imaging findings, clinical risk factors including family history, prior biopsy results, DRE findings, PSA level, and other serum and urine biomarkers indicated the need for testing. This study had two reviewers who evaluated all lesions; the interpretation results of reader 1 were used for the purposes of the present study given reporting of separate reader statistics and this reader’s more established clinical experience with prostate MRI. Of the 343 lesions in this population, 4, 128, 79, 75, and 57 lesions were PI-RADS 1 through 5 of which 0 (0%), 2 (2%), 9 (11%), 37 (49%), and 42 (74%) were HGPCA, respectively. The HGPCA prevalence was 26%, reflecting patients with variable levels of risk and indication for MRI that is similar to prevalence in epidemiologic studies of the general population with suspected prostate cancer.(25) Ten peripheral zone lesions initially scored as PI-RADS 3 were upgraded to PI-RADS 4 based on GRASP-DCE; 3 (30%) of these were high grade cancer. Therefore, a threshold of PI-RADS ≥3 using non-contrast sequences for performing biopsy yielded sensitivity and specificity of 97.8% and 51.6%, respectively, for detecting HGPCA. A threshold of PI-RADS ≥4 using GRASP-DCE for detection of HGPCA provided sensitivity and specificity of 87.8% and 79.1%, respectively, for HGPCA.(24)

The second study involved 163 patients with 651 lesions from 2012–2015, and used standard DCE-MRI techniques, with prostatectomy as the reference standard.(23) Cases were consecutive patients who had undergone MRI before prostatectomy while controls were consecutive patients who no lesions on MRI and a negative biopsy. Multiple readers each reviewed different lesions after establishing agreement using PI-RADS v2. Of the 651 lesions in the study, 70, 115, 305, and 161 lesions were scored as PI-RADS 2 through 5, respectively. Among these lesions in each PI-RADS category, 11 (16%), 38 (33%), 215 (71%), and 147 (91%) were high grade cancer, respectively. The prevalence of high grade cancer was 63%. Eighty-seven peripheral zone lesions initially given a PI-RADS 3 using non-contrast sequences were upgraded to a PI-RADS 4 based on DCE characteristics; 47 (54%) of these were high grade cancer. Thus, a threshold of PI-RADS ≥3 for detection of HGPCA yielded an overall sensitivity and specificity of 97.32% and 24.48%, respectively, while using PI-RADS ≥4 for detection of HGPCA provided sensitivity and specificity of 88.0% and 56.3%, respectively.(23)

RESULTS

Net Benefit and Standardized Net Benefit of DCE-MRI Risk Assessment Strategies

The most favorable of the four strategies using prostate MRI with PI-RADS v2 to determine biopsy depended upon the DCE technique used, differing for each study population across the same range of risk thresholds. The net benefits of each risk model are displayed numerically in Table 2 and graphically in Figures 1a and 2a. Summarizing findings in the clinically relevant range of risk thresholds, GRASP DCE-MRI risk stratification using PI-RADS ≥4 as a biopsy threshold yielded the highest net benefit in an overall average risk population. Meanwhile, the sensitivity and specificity reported for routine DCE-MRI did not enable higher net benefit with a threshold of PI-RADS ≥4 than PI-RADS ≥3 lesions for any clinically relevant risk thresholds, as tested in a high prevalence population.

Table 2.

Probability threshold ranges at which various biopsy options have the greatest net benefit. GRASP = golden-angle radial sparse parallel imaging. The strategy of “Biopsy None” was not relevant within clinically relevant risk thresholds (e.g. greatest net benefit at >62% risk).

Test and Study Author	Biopsy All Lesions	Biopsy All PI-RADS ≥3 lesions	Biopsy All PI- RADS ≥4 lesions using DCE
DCE-MRI with GRASP (Rosenkrantz et al(24))
Range of Probability Thresholds with Model showing highest NB	0-1%	2-11%	>11%
Standard DCE-MRI (Greer et al(23))
Range of Probability Thresholds with Model showing highest NB	0-15%	16-33%	>33%

Figure 1.

Net benefit (1a) and standardized net benefit (1b) of various biopsy strategies based on study data using GRASP DCE-MRI, with average HGPCA prevalence. Over probability thresholds (>11%), GRASP DCE-MRI with biopsy for lesions scored PI-RADS ≥4 (red line) was most beneficial. Biopsy of no lesions did not provide highest net benefit within a clinically relevant range of risk thresholds.

Figure 2.

Net benefit (2a) and standardized net benefit (2b) of various biopsy strategies based on a study population undergoing standard DCE technique, with higher HGPCA prevalence. Over moderate-to-high probability thresholds (15–33%), biopsy of lesions scored PI-RADS ≥ 3 yielded the greatest net benefit, while DCE-MRI for biopsy lesions PI-RADS ≥4 (red line) was most beneficial only at high risk thresholds (>33%). Biopsy of no lesions did not provide highest net benefit within a clinically relevant range of risk thresholds.

GRASP DCE-MRI with a threshold of PI-RADS ≥4 showed the highest net benefit of any biopsy threshold at probability thresholds of >11%. Patients at a lower range of probability thresholds (2–11%) gained the highest population net benefit from biopsy of all PI-RADS ≥3 lesions (without use of DCE information). As expected, patients with a very low probability threshold for HGPCA (0–1%) benefitted most from a biopsy-all strategy, maximizing the possibility of detecting HGPCA regardless of biopsy related harms.

In the study population undergoing standard DCE-MRI, patients with probability thresholds for HGPCA of 0–15% had the highest net benefit from biopsy of all patients, while for the range of probability thresholds of 16–33%, the highest net benefit came from biopsy of all PI-RADS ≥3 lesions without characterization on DCE imaging. Thus, risk stratification using standard DCE and PI-RADS 4 as a threshold for biopsy did not provide a comparative net benefit within the clinically relevant range of risk thresholds, and only maximized the net benefit at very high threshold ranges (>33%) that few decision makers would choose.

The standardized net benefit curves (Figures 1b and 2b) display the net benefit of each strategy divided by the prevalence of disease. Over the range of risk thresholds (>11%) for which GRASP-DCE MRI showed the highest net benefit with a biopsy threshold of PI-RADS ≥4, the standardized net benefit of this risk model ranged from 63–80%, or the same standardized net benefit as a policy that intervened in 63–80% of cases (true positive results) and no controls. For example, at a chosen risk threshold of 15%, the standardized net benefit was 77.4%. Using the reported sensitivity and specificity, projections of standardized net benefit would further increase with increasing prevalence (as may be found in subpopulations with other markers increasing the pretest probability of HGPCA); using the same mathematical formula for standardized net benefit and a hypothetical HGPCA prevalence of 0.4, the net benefit at a risk threshold of 15% would be 82.2%.(26)

Potential Impact of GRASP DCE-MRI on Reduction of Unnecessary Biopsies

The potential reduction in unnecessary biopsies was projected, based on false positive MRI results that would occur with use of each PI-RADS score risk model. For a population undergoing GRASP DCE-MRI at a selected probability threshold of 15%, use of PI-RADS ≥4 as a threshold would result in a reduction of 5.5 unnecessary biopsies for every 100 men compared with biopsy of PI-RADS≥3 lesions. Compared with biopsy of all patients, 40 unnecessary biopsies would be prevented with the use of GRASP DCE-MRI with biopsy threshold of PI-RADS ≥4.

For the study population undergoing standard DCE-MRI, biopsy of only PI-RADS≥4 lesions as based on DCE images did not maximize net benefit in the clinically relevant range of thresholds of 5–20%. Still, compared with biopsy of all patients at a risk threshold of 5%, biopsy of PI-RADS≥3 lesions would provide reduction of 23 unnecessary biopsies.

Sensitivity Analysis for Clinical Utility of DCE-MRI with GRASP

We used alternative reader data from the DCE-MRI study using GRASP to assess the potential effect of reader variability on the clinical utility of the technique. When the sensitivity and specificity of a less experienced reader were applied to the decision curve analysis, the probability thresholds for which PI-RADS ≥3 provided greatest net benefit were 4–27%, while the thresholds for which PI-RADS ≥4 offered greater net benefit than biopsy of PI-RADS ≥3 lesions was >27%.

DISCUSSION

Clinical risk assessment for high grade prostate cancer is increasingly supplemented with prostate MRI, in which the likelihood of clinically significant cancer is scored using PI-RADS v2.(21,22,27,28) Because contrast-enhanced MRI techniques are only used to re-categorize peripheral zone lesion risk from PI-RADS 3 to 4 in the current scoring system, the value of performing contrast-enhanced imaging has been questioned. Without contrast-enhanced imaging, a threshold of PI-RADS 3 would determine the need for biopsy without further risk-stratifying information. We found that a broad recommendation to image with versus without contrast enhanced sequences would not adequately capture the profile of population level benefits of MRI-based risk stratification. Some patients with very low risk thresholds for detecting HGPCA may benefit most from a binary test result for biopsy, but others benefit most from additional risk stratification. Prostate cancer detection typifies the clinical decision where personal weighing of benefits and harms guides testing and treatment, supporting the assessment of testing technique in this context.

Our results support the broad clinical utility of PI-RADS ≥3 as a biopsy threshold in reducing unnecessary biopsies compared with requisite biopsy of all patients, but also that the net benefit of MRI is further improved when DCE imaging provides sufficiently accurate risk stratification. GRASP DCE-MRI was associated with sufficient lesion reclassification to support a threshold of PI-RADS ≥4 by maintaining similar sensitivity while reducing unnecessary biopsies in a patient population with a broad range of risk for HGPCA. Using PI-RADS 3 as a threshold for biopsy was favored only in the low range of personal risk thresholds (2–11%) representing maximized detection of HGPCA with little consideration for the risks of biopsy. We note that the favorability of PI-RADS ≥4 threshold for biopsy reflects the sensitivity and specificity of a highly experienced reader using DCE-MRI with GRASP, and therefore further prospective study is needed to establish whether high spatiotemporal resolution techniques, in combination with routine non-contrast sequences (diffusion weighted imaging, T2 weighted imaging), can facilitate risk stratification with fewer biopsies and preserved oncologic outcomes in patients who are deciding whether to undergo prostate biopsy.

GRASP utilizes a combination of compressed sensing and parallel imaging to reconstruct images at very high temporal resolution. In an early study of 20 patients with known prostate cancer, GRASP DCE-MRI had better anatomic delineation of the capsule, peripheral/transition zone boundary, urethra, and peri-prostatic vessels as well as improved image sharpness, inter-reader correlation for lesion size, and lesion conspicuity.(17) In a subsequent study of 58 men with prostate cancer, lesion conspicuity was higher for HGPCA Gleason score ≥ 3+4 lesions at very high temporal resolution compared with lower temporal resolution (1.4 vs. 14.9 sec, p=0.029).(29) A direct comparison of GRASP to standard DCE-MRI test performance has not been performed. We analyzed the available evidence for each technique and found that a comparative net benefit could not be shown using the data on standard DCE imaging, but high spatiotemporal resolution DCE-MRI is likely to yield clinically impactful risk-stratification for HGPCA in the patient population deciding upon biopsy.

Commonly employed test performance metrics such as sensitivity, specificity, and area under the receiver operating characteristic curve lend limited support for decision makers weighing consequences of the test. Once the risk-stratifying ability of the test is established, the benefits and harms of undergoing a test can be personalized and weighed in context of the patient’s circumstances, including degree of risk and personal preferences.(19,20,30) Decision curve analysis is grounded in a decision-theoretical framework of the benefits and harms of undergoing or avoiding an intervention across various risk thresholds. Aside from showing that the highest net benefit of “biopsy all” and “biopsy none” regardless of MRI occur only at extreme ends of risk thresholds, MRI-based risk models (dependent upon DCE technique) differ in relevance according to the underlying disease prevalence and the patients’ individual risk thresholds.

A recent study by Kuhl et al evaluated interpretation of an abbreviated non-contrast protocol versus contrast-enhanced prostate MRI in men with elevated PSA.(15) Radiologists sequentially evaluated and interpreted non-contrast MRI images followed by the addition of contrast-enhanced images in 542 men; contrast-enhanced images identified only one additional HGPCA but also added 11 false positives. All men with PI-RADS ≥3 lesions underwent biopsy, and DCE images were used only to aid detection, rather than re-classify PI-RADS 3 lesions already detected on diffusion weighted imaging to higher suspicion PI-RADS 4 scores. Our study assessed the potential of DCE-MRI to separate lesions needing immediate biopsy from those that can be observed, based on reclassification of PI-RADS 3 peripheral zone lesions using DCE images.

There are several limitations to our study. First, we included two studies and populations who underwent specific MRI protocols, which may differ slightly across institutions despite meeting technical specifications of PI-RADS. The decision curve analysis results for each study do not represent a head-to-head comparison of the techniques. The differences may be attributable to multiple factors, including the non-contrast sequences as well as the differences in DCE technique (GRASP versus standard), and potentially reader experience or inter-reader variability in scoring based on PI-RADS v2 criteria. While we standardized the decision curve analysis results for varying prevalence, differences in study design may affect true positive versus false positive rates. First, the use of whole-mount prostatectomy in the study with high prevalence may possibly differ in likelihood of correctly identifying HGPCA in a given location than the fusion biopsy standard in the standard risk group, but whole-mount histologic correlation with a visible lesion on MRI may still be challenging given the changes from surgical resection and formalin fixation. While the study sample with standard HGPCA prevalence included all consecutive patients who had a fusion biopsy, the sample population in the high HGPCA prevalence study only included patients who underwent prostatectomy, meaning the cases in the latter study were more likely to have HGPCA on biopsy that would warrant surgical management. Decision curve analysis does not predict an individual patient’s favorable MRI technique but reflects population level benefits and harms. We focused on the added value of intravenous-contrast based on the PI-RADS v2 upgrade and did not evaluate effects of DCE on lesion detection outside of the PI-RADS system (as in scoring of transition zone lesions). Given these differences in study design, a direct comparison of DCE MRI techniques was not possible and further prospective studies are needed to establish the contribution of GRASP or other high-resolution DCE techniques to the clinical outcomes of prostate lesion characterization, and such data may update decision analyses on prescribing decision rules for biopsy.

In conclusion, reclassification of peripheral zone lesions using GRASP DCE-MRI that provides relative preservation of sensitivity and improved specificity for detection of HGPCA can maximize the population net benefit across clinically relevant risk thresholds. Non-contrast MRI characterization is likely sufficient to inform the decision for biopsy only at low personal risk thresholds for detection of HGPCA (5–11%), where the risks of biopsy are given minimal consideration. Thus, prospective study is warranted to further assess the avoidance of unnecessary prostate biopsies and oncologic outcomes of using GRASP DCE-MRI to guide prostate biopsy decisions.