Biparametric Quantitative MRI for Prostate Cancer Detection

Supplemental Digital Content is Available in the Text.

Abstract:

Objectives:

This study sought to prospectively investigate a novel quantitative biparametric prostate magnetic resonance imaging (MRI) protocol to detect prostate cancer (PCa) in biopsy-naïve men. Secondarily, this study reports the accuracy of fractional order calculus (FROC) diffusion and quantitative T2 compared with the Prostate Imaging Reporting & Data System (PI-RADS).

Methods:

This prospective pilot study (NCT04175730) enrolled 50 prostate biopsy-naïve men who met eligibility criteria. All men received 3T MRI with T2 and diffusion-weighted imaging (DWI) (b-values: 50–4,000 s/mm²). Men with PI-RADS lesions ≥3 underwent targeted and systematic prostate biopsy, omitting systematic biopsy cores in peripheral zone lesions. DWI series images were fit to signal decay to calculate ADC (mm²/s) and the FROC model for coefficient D_F (mm²/s). The primary end point was detection of Gleason grade group ≥2 (GG≥2) PCa. Receiver operating characteristic regression and area under the curve (AUC) were reported.

Results:

Forty-eight men underwent MRI and biopsy. Mean age was 61.5 years (56–68), 29% were White, 52% were African American, mean PSA was 6.0 ng/mL (4.9–8.0), and mean PSA density was 0.14 ng/mL². In total, 61 PI-RADS ≥3 lesions were targeted for biopsy. GG≥2 PC was found in 7% (1/14) of PI-RADS 3 lesions, 28% (10/36) of PI-RADS 4 lesions, and 36% (4/11) of PI-RADS 5 lesions. The AUC for detection of GG≥2 PC was 0.63 (0.5–0.76) for PI-RADS, 0.82 (0.68–0.96) for ADC, and 0.87 (0.77–0.97) for the FROC model.

Conclusion:

This small prospective pilot study demonstrates the feasibility of a novel quantitative biparametic MRI protocol to detect prostate cancer in biopsy-naïve men.

Prostate cancer (PCa) remains the second leading cause of cancer death in US men, but concern for overtreatment remains.¹ This paradox is partly due to limitations of serum prostate-specific antigen (PSA) and digital rectal examination (DRE) screening and the inability to reliably differentiate indolent from aggressive tumors. Multiparametric magnetic resonance imaging (mpMRI) and PI-RADS scoring have helped identify suspicious lesions for MRI/US fusion-targeted biopsy (TBx), improving the detection of clinically significant PCa while simultaneously reducing the number of biopsies needed by approximately 1/3rd.^2,3

These results are despite limitations in current mpMRI acquisition protocols and PI-RADS scoring. For example, the specificity of PI-RADS 3 for clinically significant PCa is as low as 12% and would miss almost 50% of all PCa and 13% of Gleason Grade Group (GG) ≥2 PCa if used as a cut-off for biopsy.^3,4 In addition, prostate MRI interpretation largely relies on subjective analysis to generate PI-RADS scores, conferring only moderate inter-rater reliability which has led to difficulty generalizing and replicating results across larger populations.^5–7 To address these shortcomings, quantitative parameters are being studied in an attempt to enhance PI-RADS's standardization and characterization of lesions.^8–10 One such example can be seen when deploying a fractional order calculus model (FROC), instead of a monoexponential diffusion model, to provide additional quantitative parameters in diffusion-weighted MRI series, which has already shown promise in differentiating benign tissue from prostate cancer in one retrospective study.¹¹

Finally, current mpMRI protocols are “one size fits all” but used in a variety of clinical scenarios including PCa detection in men with prior negative standard transrectal ultrasound-guided prostate biopsies (SBx), staging of known PCa, and suspected local recurrence after treatment—making them costly, time-consuming, and currently not ideal for widespread use in the biopsy-naïve setting. Abbreviated biparametric MRI (bpMRI) protocols (without the use of enhanced or DCE series) have been shown to be comparable with mpMRI in the detection of PCa, but the data are limited and have not been studied in a diverse US population.^12–14 This study sought to design a bpMRI protocol that reduces image acquisition time, provides targets during prostate biopsy, and facilitates additional quantitative analyses of imaging parameters for PCa detection to address some of the limitations of existing MRI protocols. Primarily, this study reports on the feasibility and performance of a novel bpMRI protocol to detect prostate cancer in a racially diverse group of US men with clinical suspicion for PCa. Secondarily, this study reports the accuracy of fractional order calculus diffusion and quantitative T2 compared with the Prostate Imaging Reporting & Data System (PI-RADS).

MATERIALS AND METHODS

Study Design

After institutional review board approval, a single-arm prospective pilot study was conducted with an a priori sample size of 50 men (NCT04175730). All subjects required informed consent and were adult male subjects with a clinical suspicion of PCa (PSA ≥4 ng/mL and ≤20 ng/mL) with no prior PCa diagnosis or biopsy. In addition, men with MRI-incompatible medical conditions (severe claustrophobia not responsive to treatment with oral anxiolytic medication or rectal anatomy incompatible with TRUS) or implants were ineligible to participate. The study flow diagram is shown in Fig. 1. The primary end point was PCa detection. Secondary objectives were to analyze detection of Gleason Grade Group (GG) ≥2 PCa and report the diagnostic accuracy of the quantitative imaging parameters.

FIGURE 1.

CONSORT diagram. Patients assessed and included in this study. PZ, peripheral zone; ROI, region of interest; SOC, standard of care; TRUS, transrectal ultrasound-guided; TZ, transition zone.

Image Data Acquisition

MRIs were performed using a 3T MRI scanner and 32-channel cardiac surface coil (GE Healthcare, Discovery 750 MRI). No intravenous contrast was used. The custom protocol consisted of T2-weighted imaging in the axial and sagittal planes, T2 mapping images at multiple echo times (TE), and DWI (at multiple b-values) in the axial plane. The axial acquisition plane was defined as perpendicular to a line fit to the anterior rectal wall at the midsagittal plane. The DWI phase was acquired with a separation between 2 diffusion gradient lobes ∆ = 38.6 ms and duration of each diffusion gradient δ = 32.2 ms, while using both a single-shot echo planar imaging pulse sequence and Stejskal-Tanner diffusion gradients, fixed for each b-value. Detailed acquisition parameters are displayed in Table 1. After acquisition, all images were deidentified and transferred to a secure server.

Table 1.

MR Acquisition Parameters

Parameter	Sequence	TR	TE	FOV	Acquisition matrix	b-values (# of averages)	Slice thickness	Acquisition Time
T2-weighted	SE	9,789 ms	97.6 ms	18 cm ×18 cm	352 × 224	—	3 mm	03:51 min
DWI	EP\SE	1,638 ms	84.7 ms	28 cm × 28 cm	128 × 128	50(1), 100(1), 400(1), 700(1), 1000(2), 1300(2), 1600(4), 2000(4), 3000(8), 4000(16) s/mm2	3 mm	06:36 min
T2 mapping	SE	1,900 ms	14, 28, 41, 55, 69, 83, 97, 110 ms	28 cm × 28 cm	128 × 128	-	3 mm	08:13 min

EP, echo planar; SE, spin echo; TE, echo time; TR, repetition time.

Image Data Processing

Diffusion-Weighted Imaging

Diffusion-weighted imaging data acquired for each b-value were normalized with respect to the minimum b-value (50 mm²/s). Cross-sectional voxels were fit to all the tested models using a nonlinear least-squares estimation with an iterative Levenberg-Marquardt method in MATLAB (MathWorks, v2017b, Natick, MA).

For consistency of model comparison, ADC values were calculated by fitting all b-values of the acquired diffusion data to a monoexponential model, using the following equation:

$$\mathrm{S}\left(\mathrm{b}\right)={\mathrm{S}}_0\cdot \exp \left(-\mathrm{b}\cdot \text{ADC}\right)$$

where S(b) represents the signal intensity at a particular b-value, S₀ is the signal intensity at b = 0 s/mm², and b is the b-value.

In addition, diffusion data from all b-values were fit to the FROC model,^15,16 using the following equation:

$$\mathrm{S}\left(\mathrm{b}\right)={\mathrm{S}}_0\cdot \exp \left(-{\mathrm{D}}_{\mathrm{f}}\cdot \mu *\cdot {\mathrm{b}}^{\beta }\right)$$

where S₀ is the signal intensity without diffusion weighting, D_f (mm²/s) is the anomalous diffusion coefficient, β is the fractional order derivative in space, and µ (mm) is a spatial parameter.

T2 Mapping

T2 maps are obtained from all TE (14, 28, 41, 55, 69, 83, 97, 110 ms) values using the monoexponential nonlinear least-square fit analysis with the following equation:

$$\mathrm{S}\left(\text{TE}\right)={\mathrm{S}}_0\cdot \exp \left(-{\rm T}{\rm E}/\mathrm{T}2\right)$$

where S(TE) represents the signal intensity at a particular TE and S₀ is the signal intensity at the first TE = 14 ms.¹⁷

Image Interpretation

MRIs were reviewed by a single board-certified radiologist (Karen Xie, 16 years of experience) with subspecialty training in body imaging and significant prostate MRI experience using OsiriX software (PIxmeo, Geneva, Switzerland). After evaluation of all collected and processed images, the radiologist segmented all suspicious regions of interest (ROI) on each of the axial T2-weighted and ADC calculated images and assigned each ROI a PI-RADS score (version 2.1).¹⁸ Each ROI was measured in its greatest dimension and categorized as left/right and anterior/posterior relative to the urethra, as transition or peripheral zone, assessed for extracapsular extension (ECE), and labeled with its' ROI specific median quantitative parameters. All annotated images were exported to DICOM format as well as an MPEG movie for use by the surgeon during biopsy.

Biopsy Protocol

PI-RADS scores ≥3 were considered a positive ROI and subsequently targeted for biopsy using 3 different strategies. Men with no suspicious lesions on MRI (“negative MRI”) underwent a 12 core SBx. Men with a transition zone ROI alone underwent 6 core SBx and TBx of each ROI. Men with peripheral zone ROI underwent TBx of each ROI and no SBx. All TBxs were performed by a urologist using the GE logiq E9 Ultrasound platform (GE Healthcare, Chicago, IL). Two to 4 biopsies were obtained for each ROI at the discretion of the surgeon using an 18-gauge, 15-mm length core needle biopsy instrument (Monopty, Bard Medical, Covington, GA).

Biopsy cores were paraffin-embedded and prepared on 5-micron slices stained with hematoxylin and eosin. Additional immunostaining was performed at the discretion of the pathologist. Slides were digitized on an Aperio AT2 whole slide scanner at 20× magnification (Leica Biosystems Inc. Buffalo Grove, IL). Biopsy cores were evaluated by a board-certified genitourinary pathologist using the International Society for Urologic Pathology Gleason grade group (GG) and Aperio's Image Scope software (Leica Biosystems Inc. Buffalo Grove, IL).¹⁹ An ROI was considered positive if any TBx core contained cancerous prostate tissue.

Statistical Analysis

Patient-level descriptive parameters were compared based on biopsy outcome (cancerous vs. benign) using the Kruskal-Wallis test for continuous variables and chi-square tests for categorical variables. At the 3-dimensional ROI level, after the maps of ADC, D_f, β, and μ were produced, we examined histograms of the voxel-wise estimates of ADC, D_f, β, and μ parameters. Each PI-RADS ≥3 ROI was summarized with mean, median, minimum, maximum, and range of values of all model parameters (ADC, D_f, β, and µ). Parameters of each ROI were normally distributed (Q-Q plot); therefore, mean parameter values were used for analysis. All parameters were compared between cancerous and benign prostate ROIs by a nonparametric Kruskal-Wallis H test. The discriminatory accuracy for the detection of PCa using the individual model parameters and combined FROC (D_f, β, and µ) model parameters were computed with receiver operating characteristic (ROC) regression for each group. The area under the ROC (AUC) was reported. Sensitivity, specificity, and diagnostic accuracy of each parameter was calculated at the Youden's index point of the ROC, as well as the point at 90% fixed sensitivity. Multivariate logistic regression models for patient-level cancer diagnosis and generalized estimation equations (GEE) models for ROI biopsy cancer outcomes were built using a forward variable selection method (2-sided alpha of 0.10) combining patient-level and imaging characteristics. Odds ratios (OR) and 95% confidence intervals were estimated for selected patient and imaging-level (ROI) factors that predicted malignancy. Predicted probabilities of malignancy from these multivariate logistic or GEE models were then used to generate ROC curves. All statistical analyses were performed using SAS (SAS Institute, v9.4, Cary, NC) and MATLAB (MathWorks, v2017b Natick, MA) at a significance level of P < 0.05.

RESULTS

Patient and Protocol Characteristics and Biopsy Outcomes

Target enrollment was met, and 48 men completed biopsy. PCa was diagnosed in 19/48 men (40%). SBx diagnosed 5 of the 19 men with PCa (3 GG1, 2 GG ≥2), representing a false-negative MRI. The characteristics of the overall study cohort stratified by benign vs. PCa biopsy outcome are shown in Table 2. The cohort was racially diverse (71% African American or other race) with a median age of 62 years. Prostate volume was higher and PSA density lower in the benign patients (both P < 0.05). The remaining characteristics were similar between groups, including PI-RADS score.

Table 2.

Study Participants Demographic Information

	Any PCa (n = 19)	Benign (n = 29)	Total (n = 48)	P
Age (years)	61.3 (57.0–68.0)	61.0 (56.0–66.0)	61.5 (56.8–68.0)	0.50
Race				0.98
White	6 (43%)	8 (57%)	14 (29%)
AA	10 (40%)	15 (60%)	25 (52%)
Other	3 (33%)	6 (67%)	9 (19%)
Family history of PCa?				0.48
Yes	5 (56%)	4 (44%)	9 (19%)
No	14 (36%)	25 (64%)	39 (81%)
PSA (ng/mL)	7.1 (5.3–11.0)	5.7 (4.7–7.9)	6.0 (4.9–8.0)	0.13
Prostate volume (cc)	33.9 (28.2–41.5)	53.5 (35.1–67.7)	39.0 (32.0–60.8)	0.006
PSA density (ng/mL/cc)	0.21 (0.15–0.26)	0.13 (0.09–0.17)	0.14 (0.10–0.22)	0.003
IPSS	6 (3–10)	8 (3–13)	7 (3–13)	0.68
SHIM	18 (12–23)	17 (14–22)	17 (13–23)	0.83
Any PI-RADS ≥3?				0.78
Yes	14 (42%)	19 (58%)	33 (69%)
No	5 (33%)	10 (67%)	15 (31%)

Patient-level descriptive parameters were compared based on biopsy outcome (cancerous versus benign) using the Kruskal-Wallis test for continuous variables and chi-square tests for categorical variables.

AA, African American; IPSS, International Prostate Symptom Score; PCa, prostate cancer; PSA, prostate-specific antigen; SHIM, Sexual Health Inventory for Men.

The total acquisition time for this bpMRI protocol was 18 minutes and 40 seconds. The DWI sequence alone took 6 minutes and 36 seconds to complete (Table 1). Of the 50 study subjects, 33 (66%) had PI-RADS ≥3 ROI on MRI and underwent TBx (n = 61 ROIs). 21 (44%) had negative MRIs and underwent SBx (supplemental materials Fig. 1, http://links.lww.com/TMRI/A20). ROI level results are displayed in Table 3. Notably, ROIs with ECE and peripheral zone location were associated with malignancy (both P < 0.05). The rate of malignancy of transition zone ROIs was 17% versus 47% in peripheral zone. While PI-RADS distribution did not differ significantly between malignant and benign biopsies, there was a low percentage of malignant biopsies from ROIs with PI-RADS score 3 (Supplemental materials Fig. 2, http://links.lww.com/TMRI/A21). The monoexponential ADC was significantly lower in malignant versus benign ROI.

Table 3.

MRI Lesion Characteristics

	Any PCa (n = 22)	Benign (n = 39)	Total (n = 61)	P
PI RADS				0.14
3	2 (14.2%)	12 (85.7%)	14
4	16 (44.4%)	20 (55.6%)	36
5	4 (36.4%)	7 (63.6%)	11
Maximum size (mm)	10.1 (8.3–13.4)	9.3 (7.1–14.7)	10.0 (7.6–13.6)	0.35
Location				0.14
Anterior	9 (26.4%)	25 (73.5%)	34
Posterior	13 (48.1%)	14 (51.9%)	27
Prostate zone				0.04
Transition	4 (17.4%)	19 (82.6%)	23
Peripheral	18 (47.4%)	20 (52.6%)	38
ECE				0.01
Yes	8 (72.7%)	3 (27.3%)	11
No	14 (28.0%)	36 (72%)	50
T2 (ms)	118 (107–127)	120 (104–142)	119 (104–139)	0.51
ADC (mm2/s × 10−3)	0.78 (0.68–0.98)	1.02 (0.86–1.15)	0.95 (0.77–1.11)	0.009
DF (mm2/s × 10−3)	1.2 (1.1–1.4)	1.4 (1.4–1.5)	1.4 (1.2–1.5)	0.002
β	0.48 (0.43–0.53)	0.49 (0.41–0.54)	0.49 (0.41–0.53)	1
μ (10−3)	7.3 (7.1–7.5)	7.3 (7.1–7.4)	7.3 (7.1–7.5)	0.41

Men who underwent biopsy for suspicious regions of interest on MRI. Parameters were compared based on biopsy outcome (cancerous versus benign) using the Kruskal-Wallis test for continuous variables and chi-square tests for categorical variables.

β, the fractional order derivative in space; µ, the fractional order calculus model spatial parameter; ADC, apparent diffusion coefficient; D_f, fractional order calculus model diffusion coefficient; ECE, extracapsular extension; PCa, prostate cancer.

Quantitative Model Performance

Image maps were processed for each MRI image and summarized within the ROI segmented by the radiologist using the clinically standard sequences (T2-weighted and monoexponential DWI/ADC: representative map shown in Fig. 2). As shown in Table 3, the quantitative T2 did not differ between benign and malignant ROIs. After fitting the FROC model, only a low Df parameter was statistically associated with a malignant biopsy result.

FIGURE 2.

Representative MRI image heat maps. Sample images from a single patient. β, the fractional order derivative in space; µ, the fractional order calculus model spatial parameter; ADC, apparent diffusion coefficient; D_f, fractional order calculus model diffusion coefficient.

ROC regression was used to compare the quantitative T2 and FROC models with ADC for PCa presence in the ROI, displayed in the supplemental materials Table 1 (http://links.lww.com/TMRI/A18). Secondary analyses were performed for PCa (GG ≥2) and restricted to ROIs in the peripheral zone (supplemental materials Table 2, http://links.lww.com/TMRI/A19). When analyzed for the presence of any cancer on biopsy, the T2 parameter reported modest accuracy (AUC 0.55), good sensitivity (86%), and poor specificity (41%). Quantitative T2 accuracy was slightly improved (AUC 0.66) in the peripheral zone. Regarding DWI, the 3 parameter FROC model had slightly improved discrimination over monoexponential ADC (AUC 0.75 vs. 0.70) overall. Both diffusion models performed better in the peripheral zone and in detecting higher-grade PCa. The FROC model had improved specificity and AUC compared with monoexponential ADC in all analyses with an AUC of 0.87 for GG ≥2 PCa.

Finally, ROC regression was performed to compare the DWI models (ADC and FROC) with PI-RADS. These data are displayed in Fig. 3. PI-RADS showed modest accuracy in the overall cohort (AUC 0.59); however, this was improved in the peripheral zone (AUC 0.70). Both diffusion models outperformed PI-RADS on all analyses, but only the FROC model statistically outperformed PI-RADS for high-grade PCa detection as shown in Fig. 3B (P < 0.05).

FIGURE 3.

Receiver operator curves of quantitative models and PI-RADS for detection of prostate cancer. (A) All ROI—any cancer, (B) all ROI—high grade cancer, and (C) peripheral zone ROI—any cancer. AUC, area under the curve; ROI, region of interest.

Multivariable (Clinical and Imaging) Models

Patient-level clinical factors were added to the imaging parameters to determine how the diffusion models independently predicted cancerous ROIs. In this adjusted model, smaller prostate volume (OR 0.61, 95% CI 0.41–0.91), peripheral zone ROI location (OR 11.19, 95% CI 2.87–43.63), and the FROC diffusion parameter (OR 0.48, 95% CI 0.27–0.84) were significantly associated with cancer on TBx. The AUC of this model for any cancer was 0.88 (95% CI 0.80–0.97).

DISCUSSION

Prostate MRI, in conjunction with TBx, has become a valuable tool to diagnose PCa, but its role in screening paradigms remains limited. This prospective pilot study tested the accuracy of a novel biparametric protocol designed with several features to limit acquisition time and facilitate reproducible quantitative analysis. This study's results indicate that the use of this biparametric protocol increased the PCa detection rate and decreased the detection of GG1 PCa compared with those undergoing SBx after negative MRI despite omitting SBx in patients with peripheral zone lesions on MRI. These results are consistent with those reported with the use of mpMRI and TBx in a similar setting.²⁰ In addition, of the 21 men (44%) with negative bpMRIs, 5 (24%) were diagnosed with prostate cancer (3 GG1, 2 GG ≥2). Overall, the rates of any PCa and GG≥2 PCa diagnosed after negative bpMRI are lower than those reported after negative mpMRI in a similar setting (any PCa: 24% vs 38%, GG ≥2 PCa: 10% vs 18%).²¹ Furthermore, this study reports that the FROC model statistically outperformed PI-RADS for high-grade PCa detection and that a low Df parameter was statistically associated with a malignant biopsy result. This finding supports those from a previously published study reporting that Df is inversely correlated with PCa grade in PZ lesions.¹¹ In addition, while not statistically significant, these data indicate that the FROC model may outperform the monoexponential ADC model in detecting PCa, indicating that FROC parameters should continue to be investigated as a quantitative imaging technique in the assessment of prostate cancer.

The PI-RADS consortium has made significant improvements in the standardization of acquisition and interpretation of prostate MRIs. This study did not use IV contrast, and therefore, the reported PI-RADS scores are a modified version. There are conflicting data on the value of DCE for characterization of PCa, with some data showing results in reclassification from PI-RADS 3 to 4 in approximately 20% of peripheral zone lesions.²² In this study, the specificity of PI-RADS 3 lesions was lower than other published studies. Therefore, the generalizability of these results to studies using DCE imaging is cautioned.³ Furthermore, this study's specificity of PI-RADS 4 and 5 ROIs was lower than expected. This was primarily driven by transition zone lesions, with improved performance of PI-RADS demonstrated in the peripheral zone. As such, the accurate characterization of transition zone lesions continues to be a challenge.

While T2-weighted imaging is a cornerstone of prostate MR, the T2 mapping sequence included in this study has been sparsely studied.¹⁷ The potential advantage of this approach is in adding a quantitative T2 parameter per ROI that may be directly compared across scanners, centers, and clinical settings. In this study, the accuracy of the T2 time with regard to PCa on biopsy was unexpectedly poor. It is possible that using T2-weighted imaging by the interpreting radiologist to select ROIs resulted in a selection bias that removed higher T2 values from biopsy evaluation. Furthermore, it is possible that T2 times may vary in normal tissues such that T2 relative to “normal” tissues may be needed to demonstrate better accuracy. While these are areas for future study, these data suggest that T2 mapping as measured does not warrant the 8.5-minute acquisition time in a short screening protocol.

In this study, the DWI varied from most standard clinical protocols in several ways. Most importantly, very high b-values were acquired (up to 4000 s/mm²) to facilitate analyses other than ADC. Several models of diffusion signal decay suggest that these data deviate from a monoexponential model (ADC) at higher b-values, which may have biological significance.^{11,15,23–25} The value of acquiring high b-values as opposed to imputing these data has been studied extensively.^26–28 We chose to acquire high b-values to compare the FROC diffusion model with traditional ADC. We found that this model outperformed ADC for PCa detection, achieving excellent accuracy for GG ≥2 PCa and in the peripheral zone. How this model performs with computed b-values to further shorten image acquisition time was not tested in this study.

Despite the modest sample size of this pilot study, we attempted to control for patient-related clinical factors such as prostate volume, PSA, and age along with the MR imaging because these are known predictors of prostate biopsy outcome. These results corroborate those of previous studies, showing that smaller prostate volume and higher PSA density were strongly associated with biopsy outcome in biopsy-naïve men undergoing TBx.^13,29–31

There are several limitations of this study. As a pilot study, the sample size was modest and MRI interpretation was limited to a single reader. In addition, as a detection study of biopsy-naïve men, we relied on biopsy histology instead of whole mount prostatectomy during analysis. Similarly, the protocol used a TBx-only strategy, resulting in the inability to report the sensitivity of MRI in regions outside of ROI. Finally, we lack long-term follow-up and therefore an unknown rate of PCa diagnosis in negative biopsy patients.

Despite these limitations, this study describes a novel short biparametric prostate MR protocol designed for PCa detection and quantitative analysis in biopsy-naïve men. Furthermore, this is the first report using the FROC diffusion model to differentiate benign tissue from PCa on TBx. If the accuracy is confirmed in larger studies, a similar protocol might be a cost-effective way to perform prostate MRIs in a screening capacity in men with clinical suspicion of PCa.

CONCLUSION

This short biparametric protocol using T2 weighting and FROC diffusion imaging showed reasonable accuracy in detecting clinically significant prostate cancer in a diverse cohort of biopsy-naïve men with elevated PSA.