Is dynamic contrast enhancement still necessary in multiparametric magnetic resonance for diagnosis of prostate cancer: a systematic review and meta-analysis.

BACKGROUND: The purpose of this study is to systematically review the literatures assessing the value of dynamic contrast enhancement (DCE) in the multiparametric magnetic resonance imaging (mpMRI) for the diagnosis of prostate cancer (PCa).
METHODS: We searched Embase, PubMed and Web of science until January 2019 to extract articles exploring the possibilities whether the pre-biopsy biparametric magnetic resonance imaging (bpMRI) can replace the position of mpMRI in the diagnosis of PCa. The sensitivity and specificity of bpMRI were all included. The study quality was assessed by QUADAS-2. Bivariate random effects meta-analyses and a hierarchical summary receiver operating characteristic plot were performed for further study through Revman 5 and Stata12.
RESULTS: After searching, we acquired 752 articles among which 45 studies with 5,217 participants were eligible for inclusion. The positive likelihood ratio for the detection of PCa was 2.40 (95% CI: 1.50-3.80) and the negative likelihood ratio was 0.31 (95% CI: 0.18-0.53). The sensitivity and specificity were 0.77 (95% CI: 0.73-0.81) and 0.81 (95% CI: 0.76-0.85) respectively. Based on our result, pooled specificity demonstrated little difference between bpMRI and mpMRI [bpMRI, 0.81 (95% CI, 0.76-0.85); mpMRI, 0.82 (95% CI, 0.72-0.88); P=0.169]. The sensitivity, however, indicated a significant difference between these two groups [bpMRI, 0.77 (95% CI, 0.73-0.81); mpMRI, 0.84 (95% CI, 0.78-0.89); P=0.001].
CONCLUSIONS: bpMRI with high b-value is a sensitive tool for diagnosing PCa. Consistent results were found in multiple subgroup analysis. 2020 Translational Andrology and Urology. All rights reserved.

Introduction

Prostate cancer (PCa) is the most commonly diagnosed disease in male around the world and its incidence and mortality have been increasing (1,2). In last several years, multiparametric magnetic resonance imaging (mpMRI) has emerged as a valuable tool for several aspects of PCa management, including detection, staging, and treatment (3,4). In order to standardize and diminish the variation in acquisition, interpretation, and reporting of prostate mpMRI, the European Society of Urogenital Radiology proposed the Prostate Imaging Reporting and Data System (PI-RADS) in 2012 (5). In December 2014, the updated and simplified PI-RADS version 2 (PI-RADSv2) was introduced to address the limitations and issues derived from the old version (3). It summarized the level of suspicion of PCa in a five-point scale based on mpMRI findings considering the combination of T2-weighted (T2W), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced MRI [dynamic contrast enhancement (DCE)] (5). It is notable, however, in PI-RADSv2, DCE-MRI is considered to play only a minor role in the detection of prostate tumors, and has a secondary role to T2W and DW MRI. Recent studies have demonstrated good accuracy of biparametric-MRI (bpMRI)—the combination of T2-weighted imaging and DWI, used for tumor detection when evaluated with PSA (6-8).

DCE-MRI serves to show the perfusion parameters of tissues. It gathers information about the vascularity of tissues by assessing the signal intensity of overtime after administration of gadolinium contrast material. Greer et al. (9) indicated that DCE-MRI added extra benefits to the application of PI-RADSv2 because abnormal DCE-MRI findings increased the cancer detection rate in every PI-RADSv2 categories 2, 3, 4, and 5. Puech et al. (10) considered DCE as one of the cornerstones of mpMRI for its improvement in detection and evaluation of PCa aggressiveness. On the other hand, those who advocated the nonuse of DCE suggested that bpMRI has several advantages over mpMRI, such as shorter examination time, lower risk of allergy associated with gadolinium-based contrast agents (7,11). Aydin et al. (12) indicated both highly vascularized BPH nodules and prostatitis can lead to increased vessel enhancement, which may cause low specificity of mpMRI. Although the updated version of PI-RADS maps out guidelines of the interpretation of DCE-MRI and acquisition processing for imaging, Berman et al. (13) pointed out there were still sources of variability, such as the application of 3T scanners thus it is difficult for DCE-MRI to reproduce results across centers. In our current study, based on quantitative data, a comparison has been drawn between bpMRI and mpMRI through systematic review and meta-analysis.

Methods

Literature search

The protocol for systematic review was written according to the Cochrane Handbook for Systematic Review of Interventions version 5.1.0 (14). We searched PubMed, Embase, Web of Science to make a head to head comparison between bpMRI and mpMRI in the diagnosis of PCa, and our search strategy was as follows: (prostate cancer OR prostatic cancer OR prostate neoplasm OR prostatic neoplasm OR prostate tumor OR prostatic tumor OR prostate carcinoma OR prostatic carcinoma OR PCa) AND (magnetic resonance imaging OR MRI OR MR) AND (biparametric OR bp OR T2-weighted image and DWI OR T2-weighted imaging and DWI) until January 2019. Hand-searching of the reference lists of included studies was also performed to identify other relevant articles.

Study selection

The original studies can only be included in our network meta-analysis by meeting all the following requirements: (I) the study is published in English; (II) the available data is sufficient enough to calculate the diagnostic sensitivity and specificity of bpMRI; (III) the pathology results were provided by prostatectomy or prostate biopsy; (IV) the reported data is adequate for constructing 2×2 contingency tables with at least 10 patients. Narrative reviews, observational studies, editorials, letters comments, opinion pieces and methodological reports were all excluded. The relevant articles were selected by two researchers independently and disagreements were resolved by discussion.

Methodological quality of the included studies was evaluated by two authors independently using the same criteria as described in the Cochrane Manual for Systematic Intervention Reviews 5.2 to guarantee the quality of studies. Each item was scored as either low, high or unclear risk of bias.

Statistical analysis

Collection of results data for the quantitative synthesis was processed through Open Meta-analyst (15). All statistical analyses were conducted with the Midas module in Stata 13.1 (Stata Corporation, College Station, TX, USA). The sensitivity rate TP/(TP + FN) ×100% and specificity rate TN/(TN + FP)×100% were calculated and two forest plots were generated side by side: one for specificity and the other for sensitivity. A bivariate random effects regression was performed to calculate several primary outcomes, including diagnostic likelihood ratio positive (DLR+), diagnostic likelihood ratio negative (DLR–), and diagnostic OR (DOR) pooled sensitivity, specificity, with corresponding 95% CIs (16). The summary receiver operating characteristic curve (SROC) was used to evaluate the predictive value of each scoring system. Deek’s funnel plot was conducted to detect publication bias, with P<0.05 suggesting publication bias. Heterogeneity was valued with the Higgins-Thompson I method and the Chi-square. The significant heterogeneity was indicated by P value <0.05 and I2>50% (17). Subgroup analysis was accomplished if there was significant heterogeneity.

Results

The electronic databases search yielded 752 titles and abstracts, among which 602 studies were selected to be fully reviewed; after excluding 362 duplicates and 240 conference abstracts, reviews, case reports and letters to journal editors, 71 studies were assessed for eligibility. The details of study selection are demonstrated in . A total of 45 studies were included in the final analysis.

Figure 1

Flowchart summarizes selection process toward final group of studies analyzed.

The sample size ranged from 20 to 1,063, with a total of 5,217 patients included in our study. The involved 45 cohorts were carried out in the United States, Egypt, Switzerland, Germany, Denmark, France, Korea, Canada. Belgium, Japan, Finland, Austria, United States, Brazil, Italy, Spain and Turkey respectively. Among them 15 were (8,18-31) prospective studies and 30 were retrospective studies. The publication period of these studies was from 2005 to 2018. The characteristics of included studies are presented in . The age range of men was from 41 to 87 years (average 65.8). Across all studies, the PSA value ranged from 0.1–935.5 ng/mL. The definition of clinically significant prostate cancer (csPCa) is also varied considerably.

Table 1

Basic characteristic of included studies

Author	Period	Patient number	No. of Patients with PCa	Pre-or post-biopsy MRI	MRI-reference standard interval	Mean/median age (years)	Age range (ng/mL)	Mean/median PSA	PSA range	Mean/median prostate volume (cm)	Prostate volume range	Repeat setting	Definition of clinically significant cancer	Gleason score
Afifi et al. (18)	2016	61	54	NR	NR	NR	NR	NR	>4	NR	BR	FB	–	NR
Agha et al. (19)	2014	20	15	Post	NR	NR	NR	NR	NR	NR	NR	FB	–	NR
Rais-Bahrami et al. (32)	2015	143	84	Pre	NR	60.7	41–80	6.8	0.1–51.1	48.1	NR	FB	–	NR
Barth et al. (20)	2017	63	38	Pre	1–161 days	65.2	51.2–78.2	9.2	0.3–32.4	NR	NR	FB	Diameter ≥10 mm or GS ≥7 (3+4)	5–9
Baur et al. (33)	2014	55	55	Pre (some)	1–118 days	66	54–78	10	2.9–65.2	65	49–88	FB & PNB	–	6–10
Boesen et al. (8)	2018	1,020	655	Pre	NR	67	61–71	8	5.7–13	NR	NR	FB	GS ≥3+4	6–10
Brock et al. (21)	2015	45	41	Post	NR	66	NR	66	NR	37.5	NR	NR	–	≥6
Delongchamps et al. (22)	2011	58	58	Post	NR	62	49–74	6.8	4–9.9	35	20–120	NR	–	NR
Delongchamps et al. (23)	2011	57	57	Post	NR	63	54–76	7	2.8–28	NR	NR	NR	–	≤6 to ≥7
Doo et al. (34)	2012	51	51	Post	>3 weeks	63	50–72	11.5	4.23–43.83	NR	NR	NR	GS ≥7	6–10
Fascelli et al. (35)	2016	59	44	Pre	NR	64.3	45.0–84.9	6.8	0.9–43.3	49.1	NR	FB	GS ≥7	NR
Franiel et al. (24)	2011	54	21	Pre	2–120 days	68	49–78	12.1	3.3–65.2	NR	NR	FB & PNB	–	6–10
Haider et al. (25)	2007	44	44	Post	>6 weeks	61	46–75	5.375	0.9–26	NR	NR	NR	Gs ≥6 and diameter >4 mm	6–10
Isebaert et al. (26)	2013	75	75	Pre	1–149 days	66	49–74	10.4	1.5–70.9	NR	NR	NR	≥5 mm	6–10
Iwazawa et al. (36)	2011	178	72	NR	<44 days	68.8	41–86	20.5	4.04–568.5	NR	NR	NR	–	6–9
Jambor et al. (27)	2015	55	37	Post	1–217 days	66	47–76	7.4	4–14	42	17–107	FB	Gs ≥3+3 and diameter >3 mm	6–9
Jung et al. (37)	2013	156	72	Pre	0–189 days	59.2	42–79	5	0.2–78.1	NR	NR	FB	Diameter ≥5 mm	≥6
Junker et al. (38)	2019	236	135	Pre	NR	67.6	NR	6.4	1.89–88.44	45	15–190	FB	–	6–9
Katahira et al. (39)	2011	201	201	Post	RP: >2 months; biopsy: >1 month	70	43–80	8.6	2.61–114	NR	NR	FB	–	4–10
Kitajima et al. (40)	2010	53	30	Pre	10–41 days	69	56–84	11.1	4.2–112.1	NR	NR	NR	–	NR
Kitamura et al. (28)	2014	54	54	Post	24.8–54.5 days	62.7	NR	5.7	4.4–7.6	NR	NR	NR	–	≥6
Kuhl et al. (7)	2017	542	138	Pre	28–169 days	65	42–80	7	3.2–67.5	52	13–196	FB & PNB	PSA ≥10, GS ≥7≥ T2	6–10
Lawrence et al. (41)	2014	39	16	Pre	>9 months	64	47–77	10	1.2–36	NR	NR	PNB	–	6–9
Lee et al. (29)	2017	55	23	Pre	NR	mpMRI: 61.8; bpMRI: 62.0	NR	mpMRI: 6.7; bpMRI: 6.19	NR	mpMRI: 38.6; bpMRI: 40.2	NR	FB	–	6–10
Lim et al. (42)	2009	52	52	Post	2–38 days	65	48–76	10.5	1.2–79.6	NR	NR	FB	–	6–9
Morgan et al. (30)	2007	54	54	Post	<3 months	68	52–80	9.8	2.3–46	NR	NR	NR	–	6–9
Mussi et al. (43)	2017	118	68	Pre	<16 months	NR	NR	4.6	3.8–7.0	45	35–70	FB	NR	≥6
Naiki et al. (44)	2011	35	35	Pre	NR	67.7	49–78	12.8	2.78–67.3	NR	NR	FB	–	5–10
Rinaldi et al. (31)	2012	41	36	Post (some)	48±54 days	69	57–80	15.15	5.98–133	NR	NR	FB	–	NR
Rosenkrantz et al. (45)	2011	42	42	Post	NR	62	47–76	6.2	1.3–32.5	NR	NR	NR	–	6–9
Scialpi et al. (46)	2017	41	41	Post	28–121 days	64.5	53–78	6.8	1.5–39.3	NR	NR	NR	GS ≥7	≥6
Schimmöller et al. (47)	2014	235	115	Post	4–6 weeks	65.7	NR	9.9	NR	57.9	NR	FB & PNB	–	NR
Shimofusa et al. (48)	2005	60	37	Post	<6 months	71	54–82	21.8	4.5–130	NR	NR	FB	–	NR
Stanzione et al. (49)	2016	82	34	Pre	20–30 days	65	43–84	8.8	NR	62.9	NR	FB	–	6–9
Tamada et al. (50)	2011	50	35	Pre	1–87 days	65	45–75	6.68	4.06–9.94	NR	NR	FB	–	6–10
Tanimoto et al. (51)	2007	83	44	Pre	<4 months	67.4	53–87	19.4	4.3–332.1	NR	NR	NR	–	6–10
Thestrup et al. (52)	2016	204	68	Post (some)	<3 months	65	45–75	14	2.2–120	60	23–263	NR	GS ≥7	NR
Ueno et al. (53)	2013	80	80	Post	28±33 days	66.5	50–77	9.51	2.9–49	NR	NR	NR	–	6–9
Ueno et al. (54)	2013	73	73	Post	NR	66	50–77	9.51	2.9–49	NR	NR	NR	–	6–9
Vargas et al. (11)	2011	51	51	Post	<6 months	58	46–74	5.3	0.4–62.2	NR	NR	NR	–	6–8
Vilanova et al. (55)	2010	70	38	Pre	13±9 days	63.5	43–87	7.4	4–17.20	NR	NR	FB	–	6–8
Visschere et al. (56)	2017	245	198	NR	<2 years	66	44–85	9	1.4–935.5	NR	NR	FB	GS ≥7c, ≥0.5 mL, or extraprostatic extension	NR
Yaðci et al. (57)	2011	43	21	Pre	NR	66	49–79	9.4	1.4–120	NR	NR	FB	–	6–10
Yoshimitsu et al. (58)	2008	37	37	Post	6–9 weeks	66	56–75	11.9	0.7–54.8	49.3	19.8–201	FB	–	NR
Yoshizako et al. (59)	2009	23	23	Post	1–7 weeks	68	52–81	NR	NR	NR	NR	NR	–	6–9

PNB, previous negative biopsy; FB, first biopsy; NR, not reported.

Table 2

Basic characteristic of included studies

Author	Study design	Consecutive enrollment	Reference Standard	Blinding	Field strength (T)	b value (s/mm2)	Type of Analysis	Localization	Endorectal coil	ADC map
Afifi et al. (18)	Prospective	Y	TRUSGB and RP	NR	1.5	0, 800, 1,000	Lesion	PZ, TZ, whole	N	Y
Agha et al. (19)	Prospective	Y	TRUSGB	NR	3	0, 1,000	Lesion	Whole	Y	Y
Rais-Bahrami et al. (32)	Retrospective	NR	MRI-TRUSGB	NR	NR	NR	Patient	Whole	NR	NR
Barth et al. (20)	Prospective	N	TTSB	Y	3	0,50, 1,000 or 100, 600, 1,000	Lesion	Whole	Y	NR
Baur et al. (33)	Retrospective	Y	Targeted MRGB	Y	1.5	0, 100, 400, 800	Lesion	PZ, TZ, whole	Y	Y
Boesen et al. (8)	Prospective	NR	TRUSGB	Y	3	0, 100, 800, and 2,000	Patient	Whole	Y	NR
Brock et al. (21)	Prospective	Y	RP	Y	1.5	NR	Lesion	Whole	Y	NR
Delongchamps et al. (22)	Prospective	Y	RP	Y	1.5	0, 800	Lesion	PZ, TZ, whole	Y	Y
Delongchamps et al. (23)	Prospective	Y	RP	Y	1.5	0, 800	Lesion	PZ, TZ, whole	Y	Y
Doo et al. (34)	Retrospective	NR	RP	Y	3	0, 1,000	Lesion	Whole	N	Y
Fascelli et al. (35)	Retrospective	Y	MRI-TRUSGB	NR	NR	NR	Patient	Whole	NR	Y
Franiel et al. (24)	Prospective	Y	TRUSGB and MRGB	NR	1.5	0, 100, 400, 800	Lesion	Whole	Y	Y
Haider et al. (25)	Prospective	NR	RP	Y	1.5	600	Lesion	PZ, TZ, Whole	Y	Y
Isebaert et al. (26)	Prospective	NR	TRUSGB and RP	Y	1.5	NR	Lesion	Whole	N	Y
Iwazawa et al. (36)	Retrospective	Y	TRUSGB	Y	1.5	0, 1,000	Lesion	PZ, TZ, whole	Y	Y
Jambor et al. (27)	Prospective	Y	TRUSGB and MRGB	NR	3	0, 100, 200, 350	Lesion	Whole	Y	Y
Jung et al. (37)	Retrospective	Y	RP	Y	1.5	0, 1,000	Patient	TZ	Y	Y
Junker et al. (38)	Retrospective	Y	TTSB and RP	NR	3	50, 400, 1,000 s/mm2 before 2014 and 50, 500, 1,400 s/mm2 after 2014	Patient	Whole	Y	NR
Katahira et al. (39)	Retrospective	Y	RP	Y	1.5	500	Lesion	PZ, TZ, Whole	N	Y
Kitajima et al. (40)	Retrospective	Y	TRUSGB	Y	3	0, 1,000	Lesion	PZ, TZ, Whole	N	Y
Kitamura et al. (28)	Prospective	Y	TRUSGB and RP	Y	1.5	NR	Lesion	Whole	Y	Y
Kuhl et al. (7)	Retrospective	Y	TRUSGB and RP and TTSB and Targeted MRGB	Y	3	0, 800, 1,000, 1,400	Patient	Whole	N	NR
Lawrence et al. (41)	Retrospective	Y	MRI-TRUSGB	Y	1.5 or 3	0, 1,000, 1,400	Lesion	PZ, TZ	Y	Y
Lee et al. (29)	Prospective	Y	TRUSGB	NR	3	0, 1,000	Lesion	Whole	Y	Y
Lim et al. (42)	Retrospective	NR	RP	Y	1.5	0, 1,000	Lesion	Whole	Y	Y
Morgan et al. (30)	Prospective	Y	TRUSGB	Y	1.5	50, 400, 800, 1,500	Lesion	Whole	N	Y
Mussi et al. (43)	Retrospective	NR	MRI-TRUSGB	Y	3	50, 400, 800, 1,500	Lesion	Whole	N	Y
Naiki et al. (44)	Retrospective	NR	TRUSGB and RP	Y	NR	0, 800	Lesion	PZ, TZ, whole	Y	Y
Rinaldi et al. (31)	Prospective	NR	TRUSGB	NR	1.5	0, 250, 500, 750, 1,000	Lesion	PZ, CZ, whole	Y	Y
Rosenkrantz et al. (45)	Retrospective	Y	RP	Y	1.5	0, 500, 1,000	Lesion	PZ	N	Y
Scialpi et al. (46)	Retrospective	NR	TRUSGB and RP	Y	3	0, 2,000	Lesion	PZ, TZ, whole	N	Y
Schimmöller et al. (47)	Retrospective	Y	MRI-TRUSGB	NR	3	0, 250, 500, 750, 1,000	Lesion	PZ, TZ, whole	Y	Y
Shimofusa et al. (48)	Retrospective	Y	RP	Y	1.5	0, 1,000	Patient	Whole	N	NR
Stanzione et al. (49)	Retrospective	NR	TRUSGB	Y	3	0, 400, 2,000	Patient	Whole	N	Y
Tamada et al. (50)	Retrospective	NR	TRUSGB	Y	1.5	NR	Patient	Whole	Y	Y
Tanimoto et al. (51)	Retrospective	Y	RP	NR	1.5	0, 1,000	Patient	Whole	N	Y
Thestrup et al. (52)	Retrospective	NR	TRUSGB and MRGB	Y	3	0, 100, 800, 2,000	Patient	Whole	N	Y
Ueno et al. (53)	Retrospective	Y	RP	Y	3	0, 1,000, 2,000	Lesion	PZ, TZ, whole	N	Y
Ueno et al. (54)	Retrospective	Y	RP	Y	3	0, 1,000, 2,000	Lesion	PZ, TZ, whole	N	Y
Vargas et al. (11)	Retrospective	Y	RP	Y	3	0, 700, 1,000	Lesion	Whole	Y	Y
Vilanova et al. (55)	Retrospective	NR	TRUSGB and RP	Y	1.5	0, 1,000	Lesion	PZ	N	Y
Visschere et al. (56)	Retrospective	NR	TRUSGB and RP	NR	3	NR	Patient	Whole	N	Y
Yaðci et al. (57)	Retrospective	Y	TRUSGB	Y	1.5	800	Lesion	TZ	Y	Y
Yoshimitsu et al. (58)	Retrospective	NR	TRUSGB	Y	1.5	0, 500, 1,000	Lesion	PZ, TZ, whole	N	Y
Yoshizako et al. (59)	Retrospective	NR	RP	NR	1.5	0, 1,000	Lesion	TZ	N	Y

PZ, peripheral zone; RP, radical prostatectomy; TRUSGB, transrectal ultrasound-guided standard biopsy; NR, not given; Y, yes; N, no; TZ, transitional zone.

A total of 22 (8,18-20,27,29,31,32,35,37-39,42-44,48-50,55-58) studies were performed on biopsy-naive patients, and 4 (7,24,32,48) studies reported on a mixed cohort (patients with previous prostate biopsy or no biopsy experience). The reference standard was based on radical prostatectomy in 23 (11,18,21-23,25,26,28,34,37-39,42,44-46,48,51,53-56,59) studies, transperineal template saturation biopsy in 3 (7,20,38) studies, targeted in-bore MRI-guided biopsy in 2 (7,33) studies, MRI-ultrasound fusion guided biopsy in 5 (32,35,41,43,47) studies. Patients of 24 (18,21-26,28,30,31,33,36,37,39,41,42,45,48,50,51,55,57-59) included studies underwent MRI with a 1.5T scanner, and 19 (7,8,11,19,20,27,29,34,38,40,41,43,46,47,49,52-54,56) studies applied 3.0T scanner. Twenty-three (8,11,19-25,27-29,31,33,36-38,41,42,44,47,50,57) studies used endorectal coil. High b values (≥1,400 s/mm2) were applied in 11 (7,8,30,38,41,43,46,49,52-54) studies and low b values (<1,400 s/mm2) in 34 studies. Per-patient analysis was performed in 12 (7,8,32,35,37,38,48-52,56) studies, and per-lesion analysis in 33 studies.

Assessment of study quality and publication bias

The Quality Assessment of Diagnostic Accuracy Studies tool (QUADAS) was conducted to evaluate the quality of the study. The risk of bias for, index test, patient selection, flow and timing, reference standard, as well as the concerns for applicability were displayed in . As for patient selection, 14 (8,25,26,31,32,34,42-44,46,49,50,52,55) studies had high risk of bias as consecutive enrollment was not applied or mentioned in their articles. Regarding the index test domain, 7 (18,21-23,42,49,55) studies had high risk of bias because instead of prespecifying the cutoff value for diagnosing the presence of PCa, they established the values based on ROC curve analysis. Thirteen (18,19,24,27,29,31,32,35,38,47,51,56,59) studies did not provide enough proof that whether the MRI screening results were interpreted by assessors blinded to the biopsy results. In case of reference standard, radical prostatectomy or MRI-TRUS fusion-guided targeted biopsy were considered as the low risk reference standard. Other methods such as TRUS-guided biopsy or transperineal biopsy were considered to be of high risk. Therefore, the risk of bias in the reference standard was high in 12 (8,19,20,29-31,36,40,49,50,57,58) studies. About flow and timing, 8 (7,18,26,28,38,46,55,56) studies had high risk of bias because all included patients did not undergo the same reference standard, some underwent radical prostatectomy while others underwent TRUS- or MRI-guided biopsy. Twelve (8,19,20,29-31,36,40,49,50,57,58) studies had unclear bias for the interval between the reference standard and MRI was not provided. For applicability, 4 (18,33,36,50) studies have high risk of bias since T2W or DWI sequence was used solely instead of combining them together.

Figure 2

Chart shows summary of results of methodologic quality analysis of 45 studies in meta-analysis according to Quality Assessment of Diagnostic Accuracy Studies 2.

Little publication bias was detected by Begg rank correlation (with continuity correction) and Egger’s linear regression test of funnel plot asymmetry in this meta-analysis with a p value of 0.55 for the slope coefficient ().

Figure 3

Plot results of Deeks funnel plot asymmetry test (P=0.55) show log odds ratios for visualization of publication bias

Overall diagnostic accuracy

The result of the including researches was listed in . The sensitivity of bpMRI for distinguishing cancerous and noncancerous specimen ranged from 45% to 99%, and the specificity ranged from 37% to 100%. The pooled sensitivity was 0.77 (95% CI: 0.73–0.81) with heterogeneity (I2=93.55, P=0.00) and a pooled specificity of 0.81 (95% CI: 0.76–0.85) with heterogeneity (I2=95.73, P=0.00). On the other hand, the sensitivity of bpMRI for distinguishing csPCa and insignificant PCa (insPCa) specimen ranged from 49% to 96%, and its specificity was ranged from 34% to 88%. The pooled sensitivity was 0.78 (95% CI: 0.66–0.87) with heterogeneity (I2=96.14, P=0.00) and a pooled specificity of 0.77 (95% CI: 0.66–0.85) with heterogeneity I2=98.00, P=0.00) (). The performance of bpMRI for carcinoma in different locations was also evaluated in our present study. Concerning the peripheral zone the sensitivity of bpMRI was 75% (95% CI: 0.67–0.82) ranging from 32–91% with heterogeneity (I2=88.64, P=0.00), and the specificity was 81% (95% CI: 0.73–0.87) ranging from 45–98% with heterogeneity (I2=92.76, P=0.00) (). The sensitivity of bpMRI for transition zone was 80% (95% CI: 0.73–0.85) ranging from 72–100% with heterogeneity (I2=70.13, P=0.00), the specificity was 80% (95% CI: 0.70–0.87) ranging from 50–91% with heterogeneity (I2=92.95, P=0.00) (). The summary AUC was 0.86 for overall cancer and 0.84 for csPCa which is similar to the performance of mpMRI (0.90, 0.83 for overall PCa and csPCa respectively) (). For the cancer located at the peripheral zone, the summary AUC of bpMRI was 0.85 (), while the AUC was 0.86 for transition zone cancer (). In addition, the overall positive LR and negative LR for the overall PCa 4.10 (95% CI: 3.30–5.10) and 0.28 (95% CI: 0.24–0.33), respectively. As for csPCa, the positive LR and negative LR were 3.40 (95% CI: 2.4–4.9) and 0.29 (95% CI: 0.18–0.45) respectively, and DOR, 15 (95% CI, 11–20) for PCa, 12 (95% CI, 6–22) for csPCa. The overall positive LR and negative LR for the peripheral zone cancer were 3.90 (95% CI: 2.70–5.60) and 0.31 (95% CI: 0.23–0.40). For the transitional zone cancer, the overall positive LR and negative LR were 3.90 (95% CI: 2.60–5.80) and 0.25 (95% CI: 0.19–0.34) respectively. As for DOR, 13 (95% CI, 8–21) for peripheral zone cancer, 15 (95% CI, 9–27) for transitional zone cancer.

Figure 4

Coupled forest plots show pooled estimates of sensitivity and specificity of biparametric MRI for overall cancer.

Figure 5

Coupled forest plots show pooled estimates of sensitivity and specificity of biparametric MRI for clinically significant cancer.

Figure 6

Coupled forest plots show pooled estimates of sensitivity and specificity of biparametric MRI for cancer located at peripheral zone.

Figure 7

Coupled forest plots show pooled estimates of sensitivity and specificity of biparametric MRI for cancer located at transition zone.

Figure 8

Summary ROC (SROC) curves with prediction and confidence contours of biparametric MRI for overall cancer (A) and clinically significant cancer (B).

Figure 9

Summary ROC (SROC) curves with prediction and confidence contours of multiparametric MRI for overall cancer (A) and for clinically significant cancer (B).

Figure 10

Summary ROC (SROC) curves with prediction and confidence contours of biparametric MRI for cancer located at peripheral zone (A) and transition zone (B).

Subgroup analyses and head-to-head comparison

Subgroup analysis was conducted based on study design, patient enrollment, localization the coil application, magnetic strength, b values, reference standard, blind method application and unit for analysis. Results of all subgroup analysis were summarized in . In accordance with the above results, the distinction among included studies could be explained as a source of the heterogeneity for the diagnosis of PCa, and our result revealed that all the factors mentioned above accounted for the heterogeneity of sensitivity while none of them had an impact on specificity.

Table 3

Subgroup analysis of analysis

Parameter	Category	Number of studies	Sensitivity	P1	Specificity	P2
Coil	Used	20	0.79 (0.73–0.84)	<0.05	0.81 (0.75–0.87)	0.69
	Not used	18	0.72 (0.66–0.79)		0.83 (0.78–0.89)
Magnetic	3	13	0.74 (0.69–0.79)	<0.05	0.85 (0.81–0.88)	0.65
	1.5	23	0.83 (0.77–0.90)		0.66 (0.55–0.77)
Reference	RP or targeted biopsy	24	0.77 (0.72–0.83)	<0.05	0.80 (0.75–0.86)	0.17
	Others	15	0.75 (0.68–0.82)		0.84 (0.77–0.90)
ADC map	Used	35	0.76 (0.72–0.81)	<0.05	0.79 (0.74–0.84)	0.57
	Not used	6	0.79 (0.69–0.89)		0.89 (0.82–0.96)
Enrollment	Consecutive	26	0.76 (0.71–0.82)	<0.05	0.79 (0.74–0.85)	0.77
	Not consecutive	14	0.78 (0.71–0.85)		0.85 (0.78–0.91)
Blinding	Blinded	28	0.74 (0.69–0.79)	<0.05	0.85 (0.81–0.88)	0.97
	Not mention	11	0.83 (0.77–0.90)		0.66 (0.55–0.77)
B-values	High (>1,400)	7	0.79 (0.70–0.87)	<0.05	0.82 (0.72–0.92)	0.96
	Low (≤1,400)	26	0.78 (0.73–0.83)		0.82 (0.77–0.88)

Our studies provided head-to-head comparison between bpMRI and mpMRI. As a result, the pooled specificity demonstrated little difference between bpMRI and mpMRI [bpMRI, 0.81 (95% CI, 0.76–0.85); mpMRI, 0.82 (95% CI, 0.72–0.88); P=0.169]. The sensitivity, however, indicated a significant difference between these two groups [bpMRI, 0.77 (95% CI, 0.73–0.81); mpMRI, 0.84 (95% CI, 0.78–0.89); P=0.001] ().

Figure 11

Coupled forest plots show pooled estimates of sensitivity and specificity of multiparametric MRI for overall cancer

Discussion

Overall, we found very considerable diagnostic accuracy and precision for detection of PCa using bpMRI. Based on our assays, pooled sensitivity of bpMRI was 7% lower than that of mpMRI with statistical difference. Although the high sensitivity means higher confidence that a negative result would be a true negative, thus reducing the likelihood of additional intervention such as prostate biopsy, the 7% lower sensitivity of bpMRI may be an acceptable trade-off for lower potential risk of adverse effects and therapy cost. Besides, the relatively low sensitivity of bpMRI could be fixed through combining with other clinical indicators. Boesen et al. (60) revealed positive potential for a model combining bpMRI and prostate-specific antigen density (PSAD) for detection of PCa among 808 biopsy-naïve men. Knaapila et al. (61) indicated PSAD could improve the NPV among men with equivocal suspicion on bpMRI, this imaging criteria coupled as an adjunct with PSA level and PSAD, could provide even more accuracy in detecting csPCa. Moreover, the issue of access to MRI caused by limited availability may be remedied through the shorter acquisition time (62). Given the impressive specificity and sensitivity of bpMRI, it may be considered as a pre-biopsy test for PCa, in place of mpMRI.

Three systematic reviews (including two meta-analyses regarding) which explored the role of mpMRI in localized PCa have been published recently. In the study by Niu et al. (63) which evaluated 33 studies using a combination of T2WI, DWI, the pooled sensitivity and specificity were 0.81 (95% CI: 0.76–0.85) and 0.77 (95% CI: 0.69–0.84), respectively. In a more recent meta-analysis by Woo et al. (6) which analyzed 20 studies, the pooled sensitivity and specificity were 0.74 (95% CI: 0.66–0.81) and 0.90 (95% CI: 0.87–0.93), respectively. Compared with the former review, the current study is the first meta-analysis to evaluate the performance of bpMRI based on different location of PCa, and assess their discrimination between bpMRI and mpMRI in the detection of csPCa.

From our present study, bpMRI may be sufficient and may not miss csPCa. The pooled specificity demonstrated no significant difference between bpMRI and mpMRI [bpMRI, 0.77 (95% CI, 0.66–0.85); mpMRI, 0.70 (95% CI, 0.50–0.84); P=0.518]. The pooled sensitivity also indicated little significant difference between these two groups [bpMRI, 0.78 (95% CI, 0.66–0.87); mpMRI, 0.81 (95% CI, 0.66–0.90); P=0.135] (). It means those tumors ignored by bpMRI are mostly clinical insignificant and may also be ignored by mpMRI. Moreover, these tumors are more likely to remain latent in long-term follow-up and active surveillance.

Figure 12

Coupled forest plots show pooled estimates of sensitivity and specificity of multiparametric MRI for clinically significant cancer.

Barth et al. (20) suggested that for the diagnose of csPCa, there is no significant difference between the diagnostic performance of a bpMRI and mpMRI protocol, which met our results. Boesen et al. (8) demonstrated the high NPV of bpMRI in ruling out csPCa in biopsy-naive men, a simple, rapid bpMRI method could be used as a triage test to improve risk stratification and to exclude aggressive disease and avoid unnecessary biopsies. On the other hand, Greer et al. (9) indicated that adding DCE-MRI to DWI scores in the peripheral zone yielded meaningful progress for detecting csPCa. Although the application of bpMRI prior to biopsy could decrease the risk of over-biopsy, reduce rates of over-detection, future work must be finished for bpMRI towards maintaining the same high diagnostic yield of mpMRI without compromising oncologic outcomes and cancer detection.

Based on our current results, for the detection of cancer located at transitional zone, both the sensitivity and specificity did not demonstrate a significant difference between these two groups [sen: bpMRI, 0.80 (95% CI, 0.73–0.85); mpMRI, 0.75 (95% CI, 0.45–0.91); P=0.0845,spe: bpMRI, 0.80 (95% CI, 0.70–0.87); mpMRI, 0.86 (95% CI, 0.74–0.93); P=0.0982] DWI alone is enough for cancer located in transitional zone which met the results of PI-RADSv2. While for the cancer located in peripheral zone, the pooled specificity demonstrated significant difference between bpMRI and mpMRI [bpMRI, 0.81 (95% CI, 0.73–0.87); mpMRI, 0.96 (95% CI, 0.92–0.98); P<0.05]. The sensitivity, however, indicated little significant difference between these two groups [bpMRI, 0.75 (95% CI, 0.67–0.82); mpMRI, 0.74 (95% CI, 0.66–0.80); P=0.943].

From our analysis, the application of DCE contributes to unignorable improvements in specificity for peripheral PCa. Multiple studies have demonstrated that DCE-MRI can successfully detect PCa with a high sensitivity and specificity and help in tumor staging in peripheral zone (64-66). However, Delongchamps et al. (23) suggested DCE-MRI may decrease the accuracy of T2WI and DWI for the cancer located at the central gland without significant improvement in peripheral zone. These debatable reports might be explained by different references to evaluate DCE-MRI in a quantitative way. After the PI-RADS score was updated in 2016 by ESUR and American College of Radiology (3), the question whether DCE-MRI could lead to an added value and better performance in the interpretation of mpMRI might be answered in the future.

The b-value is one of the significant factors that lead to the heterogeneity based on our subgroup analysis, it reflects the timings and strength of magnetic field gradients of DWI applied to the patient, and the collection of multiple b-values permits the calculation of ADC map. Currently, based on the PI-RADSV2, the recommended b-values is at least 1,400 s/mm2, or if possible, up to 2,000 s/mm2 (3). Our subgroup analysis demonstrated that high b values ≥1,400 s/mm2 lead to significantly higher sensitivity and specificity for detecting PCa, Therefore, forest plots were also accomplished in present study to make a comparison between mpMRI and bpMRI with high b values ≥1,400 s/mm2 (). As shown in our results, there is no significant difference in both sensitivity [bpMRI with high b values 0.83 (95% CI, 0.72–0.90); mpMRI 0.84 (95% CI, 0.78–0.89), P=0.431] and specificity [bpMRI with high b values 0.78 (95% CI, 0.63–0.88); mpMRI 0.82 (95% CI, 0.72–0.88) P=0.621] (). The AUC is 0.88 which is similar to that of mpMRI (AUC =0.90) (). Maas et al. (67) indicated that the application of high-b-value computed could avoid artefacts and improve lesion-to-background contrast ratios for the detection of PCa. Syer et al. (68) suggested that diagnostic accuracy of combined DWI and T2WI is trustable with high b-values improving sensitivity while maintaining specificity. Further large-scale studies specifically exploring the comparison between high b-value bpMRI and mpMRI should be made to acquire an exact result.

Figure 13

Coupled forest plots show pooled estimates of sensitivity and specificity of biparametric MRI combined with high b value MRI.

Figure 14

Summary ROC (SROC) curves with prediction and confidence contours of biparametric MRI combined with high b value MRI

There are several potential limitations in our review. First, the included studies were heterogeneous in their methods, which affected the general applicability of the summary estimates. To explore the heterogeneity of our data, we performed meta-regression and multiple subgroup analysis so that the diagnostic accuracy of bpMRI could be improved in the future. Second, until recently the definition of clinically relevant PCa varied considerably between each studies, which might have resulted in unreliable conclusions in our study. Third, studies with negative results are less likely to be published, which may lead to exaggeration of the beneficial effects in meta-analysis. Fourth, the different versions of PI-RADS score the included studies used may have an impact on our results. Finally, our meta-analysis focused on newly diagnosed or clinically suspected PCa. The results of our meta-analysis do not apply to detection or staging of recurrent PCa.

Conclusions

A head-to-head comparison showed that the performance of bpMRI was similar to that of mpMRI for the diagnosis of PCa though the sensitivity was significantly lower. With the combination of high b value MRI, the sensitivity and specificity could improve to 0.83 and 0.78 respectively. The result of multiple subgroup analysis showed consistency with overall pooled estimates.

The article’s supplementary files as