Apparent diffusion coefficient in extraprostatic extension of prostate cancer: a systematic review and diagnostic meta-analysis

Abstract

Objective: To evaluate the diagnostic performance of apparent diffusion coefficient (ADC) for local staging of prostate cancer.

Methods: Databases of Web of Science, MEDLINE (Ovid and PubMed), Cochrane Library, EMBASE, and Google Scholar were searched up to May 31, 2018, with language restricted to English. All studies concerning multiparametric magnet resonance imaging (mpMRI) with ADC for detection of extracapsular extension (ECE, T3a) and/or extraprostatic extension (EPE, overall stage of T3) were identified by two reviewers independently, and quality of included studies was evaluated using Quality Assessment of Diagnostic Accuracy Studies-2 tool. True positive, false positive, false negative and true negative of each study were extracted to reconstruct the 2×2 tables for evaluating diagnostic accuracy. Summary estimates of sensitivity, specificity, and corresponding 95% CIs were calculated with bivariate model and hierarchical summary receiver operating characteristic model, then presented in forest plots. Multiple subgroup analyses and meta-regression were performed, and publication bias was evaluated with Deeks funnel.

Results: A total of 18 studies were included, with 6 involved ECE and 12 for EPE. Pooled sensitivity was 80.5% (95% CI 76.5–83.9%) with specificity of 69.1% (95% CI 62.3–75.2%). Multiple subgroup analyses showed that if ADC and length of capsular contact are regarded as independent predictors, pooled sensitivity was 85% (95% CI 77–90%) and 81.1% (95% CI 76.0–85.3%), with specificity of 70.8% (95% CI 56.3–82.0%) and 66.6% (95% CI 57.6–74.5%), respectively. Meta-regression demonstrated that there was no substantially significant difference in types of coil, magnet field strength (1.5T versus 3.0T), and analysis method (per-lesion versus per-patient).

Conclusion: By introducing ADC to MRI, we could obtain favorable sensitivity for diagnostic performance of EPE, but with a little decreased specificity.

Introduction

Prostate cancer (PCa) is the second leading cause of cancer-related mortality, accounting for the most new cases of cancer among Western countries.¹ Despite the fact that most types of PCa grow slowly and may need no intervention, the others are so aggressive that spread quickly. Therefore, early detection and location of PCa is instrumental in proper treatment and then has a better chance of successful treatment. Generally, the detection and local staging of PCa rely on the combination of several different diagnostic outcomes including serum prostate-specific antigen (PSA), tumor volume, Gleason score (GS), and digital rectal examination (DRE), and patients who are suspected of PCa would undergo transrectal ultrasound-guided (TRUS) biopsy.² Accurate preoperative knowledge is crucial in the determination of optimal therapy such as radical prostatectomy (RP), radiotherapy, and noncurative systemic therapies. However, it is estimated that pretreatment staging modalities that utilize clinical stage, PSA, and TRUS-guided random prostate biopsy inaccurately stratified more than one-third of total patients.^3,4 In recent years, multiparametric magnetic resonance imaging (mpMRI) has demonstrated its unique advantages for diagnosis, localizing, and staging of PCa and showed high accuracy. To standardize the evaluation and reporting for PCa with mpMRI, the European Society of Urogenital Radiology published the guidelines of Prostate Imaging Reporting and Data System (PI-RADS v1) in 2012,⁵ then released an updated edition (PI-RADS v2) in 2015,⁶ and demonstrated high sensitivity and specificity.^7,8 Nevertheless, local staging PCa of extracapsular extension (ECE) (T3a), seminal vesicle invasion (SVI, T3b), and extraprostatic extension (EPE) have worse prognosis while compared to organ-confined disease owing to higher risk of biochemical recurrence and metastatic disease after RP, radiotherapy, and noncurative systemic therapies.⁹ Thus, the clinical diagnosis of EPE remains challenging, and published literature suggested that the sensitivity for local staging of PCa ranged from 13% to 95%, with the specificity ranging from 49% to 97%.¹⁰ A prior meta-analysis revealed that on the whole, MRI has a poor sensitivity of 61% for detecting locally advanced PCa of EPE, with specificity of 88%.¹¹

In the past few years, researchers have developed various approaches to improve the diagnostic performance of EPE, and one of the most frequently reported approaches is by introducing ADC to MRI. However, it has not been assessed systematically, and we, therefore, aimed to evaluate the diagnostic performance of mpMRI for EPE, with RP specimens as the reference standard.

Methods

This systematic review and meta-analysis was performed and the reports comply with the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines.¹² Primary outcome of this study was the diagnostic performance (sensitivity and specificity) for EPE when ADC was introduced, and the secondary outcomes were the diagnostic accuracy of ADC and length of capsular contact (LCC) as independent predictors.

Search strategy and selection criteria

We performed a systematic search up to May 31, 2018, and covered databases as follows: Web of Science, MEDLINE (Ovid and PubMed), Cochrane Library, EMBASE, and Google Scholar. Search query combined synonyms for prostate cancer, mpMRI, EPE and ECE as follows: ([prostate] or [prostatic] and [cancer] or [carcinoma] or [tumor] or [PCa]) and ([functional] or [multi-parametric] or [mp] and [magnetic resonance imaging] or [MRI]) and ([ADC] or [apparent diffusion coefficient]) and ([extraprostatic] and [extension]) or ([extracapsular] and [extension]). References listed in qualified studies and review articles have also been screened to avoid omitting any eligible study, and publications were restricted to English. For studies in which data were not completely reported in the article, we contacted the authors for details to reconstruct the 2×2 contingency tables.

Including and excluding criteria

Studies were considered as eligible if they meet all of the following criteria: 1) using mpMRI as index test for local staging of ECE (T3a) or overall stage of EPE, 2) ADC was used to improve diagnosis accuracy, regardless of detailed values were reported or not, 3) reported the detailed data that could reconstruct the 2×2 tables, 4) standard reference was RP, and 5) studies must be original research. Studies were excluded if any of the following criteria were met: 1) using other diagnostic approaches such as ultrasound or positron emission tomography-computed tomography (PET-CT) rather than mpMRI, or just conventional MRI, 2) only involved SVI (T3b) or lymph node, 3) reference standard was not RP but TRUS biopsy or MRI-guided biopsy, 4) case report or sample size <10, and 5) letters, editorials, reviews, conference abstract, guidelines, or meta-analyses. Two reviewers (Bai and Sun) identified the eligible articles independently, and discrepancies were reconciled with discussion and if necessary, with the third reviewer (Li).

Data extraction and quality assessment

Following data were extracted by 2 reviewers independently and recorded in a predefined standardized form: 1) demographic and clinical characteristics, including patient age, prostate volume, PSA level, GS, and prevalence of ECE or EPE; 2) study characteristics which include authors, publication year, country and institution, study design and sample size, number of readers and experience, blinding to histopathological information or not, reference standard, interval between MRI and RP, scoring system, nomogram, or assessment along with corresponding cutoff values; 3) technical characteristics of mpMRI such as MRI manufacturer and model, magnetic field strength, sequences of T1-weighted imaging (T1WI), T2-weighted imaging (T2WI), DWI, and dynamic contrast-enhanced MRI (DCE-MRI), ADC cutoff value. For studies with overlapping patients, the most recently published would be included; however, if techniques and scoring systems were different, all of them would be incorporated. Methodological quality was evaluated using the tailored questionnaires and criteria based on Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2),¹³ in which four domains are scored: 1) patient selection domain describes the method for patient selection and the patients included, 2) index test domain describes the test being studied, along with how it was performed and interpreted, 3) reference standard domain describes the reference standard test used and how it was performed and interpreted, and 4) finally, flow and timing describe the flow of patient inclusion and exclusion and the interval between the index test and the reference standard. Each domain was assessed as “Yes”, “No” or “Unclear”, 2 reviewers (Bai and Sun) evaluated the quality of each study included independently, and the third reviewer (Li) was in charge of resolving the disagreements, through discussion and adjudication.

2×2 tables of true positive, false positive, true negative, and false negative values were reconstructed for each study to calculate the sensitivity and specificity. For studies that reported more than 1 result, the following strategies were adopted to extract related data: 1) for studies that reported different cutoff thresholds, the most clinically appropriate one was selected, 2) if accuracy was reported by 2 more different readers, we used the more experienced one, 3) in case of more than 2 scoring systems used, we chose the best or the quantitative one, and 4) If different techniques were used in a single study, we selected the most accurate. For studies that did not report sufficient data to reconstruct the 2×2 tables, we contacted the authors to obtain additional information.

Data synthesis and statistical analysis

Summary estimates of sensitivity and specificity along with corresponding 95% CIs were calculated utilizing the bivariate model and hierarchical summary receiver operating characteristic (HSROC) model.^14,15 Forest plots were presented to graphically show the variation of sensitivity and specificity to investigate the heterogeneity; in addition, a HSROC curve with a 95% confidence region and prediction region was plotted to depict graphic presentation of the results.¹⁵ The Deeks funnel plot was used to evaluate the publication bias, and the Deeks asymmetry test was applied to determine the statistical significance, and a P-value <0.1 for the slope coefficient was regarded as significant asymmetry.¹⁶

Multiple subgroup analyses were performed to evaluate various clinical settings as follows: 1) ADC value as independent predictor, 2) LCC as independent predictor, 3) assessment based on PI-RADS, 4) analysis only for ECE. We also performed meta-regression by adding predetermined covariates to the bivariate model: 1) use of any coil (any coil versus none), 2) magnet field strength of MRI (1.5 versus 3.0T), 3) analysis method (per-patient versus per-lesion). All of the analyses were performed with STATA 15.1 (Stata Corporation, College Station, TX, USA), with the modules of “midas” and “metandi”, and P-values <0.05 were considered as statistically significant.

Results

Literature search

Figure 1 demonstrates the concise overview of literature search and identification. The initial search yielded 853 results, of which 281 were excluded for duplication and 443 records were excluded for other reasons, then 129 studies were regarded as potential eligibility. Of them, 6 were excluded for overlapping patient with other studies, 18 were excluded for there was no sufficient data to obtain the 2×2 contingency tables, and 13 studies were ruled out for too small sample size. No additional study was identified through screening the reference list of eligible articles. Finally, a total of 18 original studies were included in this systematic review and meta-analysis.^{2,4,10,17–31}

Figure 1.

Study selection process for this systematic review and meta-analysis.

Characteristics of included studies

There were 6 studies that investigated the diagnosis performance on ECE,^{21,23–25,27,28} and the remaining 12 studies reported the diagnostic accuracy on EPE.^{4,10,17–20,22,26,29–31} Sample size of these studies ranged from 43 to 379 patients, and nearly all of them had a mean age ≥60 years. In most of studies, GS ≥5 except the study by Jäderling et al,²⁰ in which there were 9 patients had GS <6.²⁰ PSA level among these studies ranged from 0.02 to 123, with a mean value from 5.9 to 12.24 ng/mL. Only 4 studies reported the data of prostate volume, in spite of the fact that some studies suggested it is possible to play a part in the diagnosis.^17,18,21,32 Five studies performed the MRI with 1.5T scanners^{17,18,21,23,28} and the remaining 13 studies used 3.0T scanners.^{2,4,10,19,20,22,24,25,27,29–32} Except 5 studies,^{2,23,24,27,29,31} all of the others were used at least one type of endorectal, pelvic phased array (PPA), or body coil; moreover, 2 studies used two types of them (endorectal coil and PPA).^28,30 Besides T2WI, the most used functional imaging protocol was DWI; in fact, it was applied by all of the included studies. DCE-MRI also performed broadly, performed by 14 studies (2 studies not used it for analysis). Several studies performed T1WI protocol, but only for excluding hemorrhage. ADC values obtained from DWI protocol were 0/50/100/400/500/800/1,000/1,500 and 2,000 s/mm², but most studies only used 2–3 values of them. Sixteen studies reported the specific ADC values, whereas the remaining 2 study did not give further details.^23,32

For scoring system or assessment, there were several types existed across these studies. There were 3 studies that used PI-RADSv1 and 2 that used PI-RADSv2, and most of them adopted scored 3 or 4 as the cutoff value.^4,20,30,31 Four studies used the ADC values alone for assessment, with cutoff threshold ranging from 0.72 to 0.89.^17–19,27 Only 1 study used the prostate volume for evaluation.¹⁰ Finally, the most used independent quantitative predictor was LCC, adopted by 8 studies, and the best threshold reported according to them varied from 6 to 20 mm.^{2,21,22,24–26,28,29} Therefore, as a whole, nearly three-quarters of included studies used quantitative assessments, leading to better reproducibility. Demographic characteristics of included studies are summarized in Table 1, and clinical and technical characteristics are demonstrated in Table 2. Several studies used more than 2 analysis methods or techniques, and performed comparison; however, we only reported the best or quantitative one.

Table 1.

Demographic characteristics of included studies

First author	Year	Country	Period	Patients	Proved after RP	Age		PSA		GS (range)	Disease
Mean ± SD	Range	Mean ± SD	Range
Baco22	2015	USA	January 2010/September 2013.	111	40	64*	45–75	8.9*	2.5–44	6–9	ECE
Boesen26	2015	Denmark	September 2011/September 2013	87	31	66.5*†	47–74	12.0*†	4.6–45.0	6–10	ECE
Dominguez24	2018	Colombia	May 2011/December 2013	79	31	61.1±7.5	39–78	7.0±7.25*	0.02–31	6–10	ECE
Feng4	2015	USA	January 2010/December 2012	112	26	62.8±7.5	NR	8.2±7.2	NR	≥6	EPE
Giganti13	2016	Italy	January 2012/December 2014	70	23	63.97	58.9–70.5	6.82	5–9.88	≥6	EPE
Kido14	2016	Japan	April 2012/June 2014	43	18	69	54–84	7.57	1.2–30	6–9	EPE
Matsuoka29	2017	Japan	August 2007/March 2015	210	56	67	50–81	7.0	2.9–30.0	5–10	ECE
Woo28	2015	Korea	January 2012/December 2012	117	50	68.0±6.8	48–83	12.24	0.84–81.20	6–10	ECE
Woo25	2016	Korea	January 2012/December 2012	185	51	66.7±7.0	45–79	10.2±13.6	0.5–123	6–9	ECE
Schieda32	2015	Canada	April 2012/June 2014	65	42	62.8±6.0	NR	9.0±8.6	NR	6–10	EPE
Jäderling21	2017	Sweden	June 2014/January 2015	94	39	65.1*	51.0–76.5	5.9*	1.8–49.0	NR	EPE
Kongnyuy33	2016	USA	May 2007/December 2015	379	87	61.0*	47–75	8.8*	0.2–53.5	≥6	EPE
Bittencourt31	2015	Netherlands	January 2010/May 2013	133	60	61	47–72	8.6*	2.5–76.0	5–10	EPE
Rosenkrantz23	2015	USA	NR	180	46	64±8	NR	9.0±11.4	NR	6–9	EPE
Lim30	2016	Canada	January 2012/June 2014	113	76	61.9±5.4	NR	10.7±10.6	NR	6–9	EPE
Rud12	2018	Norway	December 2009/June 2012	183	103	65*	60–68	7.9*	5.8–11.5	6–9	EPE
Krishna2	2017	Canada	November 2012/May 2015	149	92	62.3±6.3	NR	6.33±3.29	NR	6–9	EPE
Lim10	2015	Canada	May 2012/May 2015	73	38	62.8±5.7	NR	10.7±10.6	NR	6–9	EPE

Notes: *Median, ^†only for proved.

Abbreviations: ECE, extracapsular extension; EPE, extraprostatic extension; GS, Gleason score; PSA, prostate-specific antigen; RP, radical prostatectomy; NR, not reported.

Table 2.

Clinical and technical characteristics of included studies

Study	Clinical						Assessment	Technical
Design	Reference	Analysis	Readers	Experience (Year)	Blinding	Field Strength	Coil	Protocols	Cutoff	b values (s/mm2)
Baco22	Retrospective	RP	Patient	1	≥800 cases	NR	LCC	1.5T	Body	T2/DWI	20	50/1000
Boesen26	Prospective	RP	Patient	2	≥2	Blind*	PI-RADS†	3T	PPA	T2/DWI/DCE	4	100/800
Dominguez24	Retrospective	RP	Patient	2	8/14	Blind*	LCC	1.5T	None	T2/DWI/DCE	12	NR
Feng4	Retrospective	RP	Patient	1	NR	NR	PI-RADS†	3T	PPA	T1/T2/DWI/DCE	3	400/800
Giganti13	Retrospective	RP	Patient	2	NR	Blind	ADC values	1.5T	ERC	T2/DWI/DCEa	0.84	0/800/1600
Kido14	Retrospective	RP	Patient	2	2/16	Blind*	ADC values	3T	PPA	T1/T2/DWI/DCE	0.72	0/2000
Matsuoka29	Retrospective	RP	Patient	2	5/10	Blind	LCC	1.5T	ERC+PPA	T2/DWI/DCE	10	0/1000/2000
Woo28	Retrospective	RP	Patient	2	9/21	Blind*	ADC values	3T	None	T1/T2/DWI/DCE	0.893	0/1000
Woo25	Retrospective	RP	Patient	1	22	Blind	LCC	3T	None	T1/T2/DWI/DCE	14	0/1000
Schieda32	Retrospective	RP	Patient	7	5/2–13	NR	PI-RADS	3T	None	T2/DWI/DCE	3	0/500/1000
Jäderling21	Retrospective	RP	Patient	2	≥20	Blind*	PI-RADS	3T	PPA	T1/T2/DWI	4	1500
Kongnyuy33	Prospective	RP	Patient	2	8/16	NR	LCC	3T	ERC	T1/T2/DWI	12.5	NR
Bittencourt31	Retrospective	RP	Lesion	1	≥20	Blind*	PI-RADS†	3T	ERC+PPA	T2/DWI/DCE	13	100/400/800
Rosenkrantz23	Retrospective	RP	Lesion	2	1/4	Blind	LCC	3T	PPA	T2/DWI/DCEa	6	50/1000
Lim30	Retrospective	RP	Lesion	2	11/15	Blind	LCC	3T	None	T2/DWI/DCE	15	0/500/1000
Rud12	Prospective	RP	Lesion	1	2	NR	ADC values	1.5T	Body	T2/DWI	0.89	50/1000
Krishna2	Retrospective	RP	Patient	2	11/15	Blind	LCC	3T	Pelvic	T2/DWI/DCE	11	0/500/1000
Lim10	Retrospective	RP	Patient	2	9/14	Blind	Volume	3T	Pelvic	T2/DWI/DCE	2.1	0/500/1000

Notes: *These studies were blinded to pathological findings but patients’ information was partly known. ^†PI-RADS version 1.

Abbreviations: RP, radical prostatectomy; ADC, apparent diffusion coefficient; DCE, dynamic contrast-enhanced magnetic resonance imaging; DWI, diffusion-weighted imaging; NR, not reported; LCC, length of capsular contact; PI-RADS, Prostate Imaging Reporting and Data System; PPA, pelvic phased array; ERC, endorectal coil.

Quality assessment

On the whole, evaluation for the 18 included studies was moderate (Figure 2). With regard to the patient selection domain, there was a high risk of bias, for almost all of the included studies were retrospective except 3 that were prospective in nature. Five studies were assigned high concern for applicability, for in these studies partly or even all patients had pathological diagnosis of PCa prior to MRI. There was a high risk of bias in 9 studies regarding index test domain. Five of 9 studies did not report whether the readers were blinded to the reference test results.^{4,17,21,31,32} Regarding the other 4 studies, readers were aware of the PSA level or/and GS of patients, though they were blinded to the histopathological findings.^20,23,25,30 As for reference domain, there were no concerns because all studies used the RP specimens as the reference standard. Regarding the flow and timing domain, it was considered as low risk of bias for all of the included studies.

Figure 2.

Risk of bias and applicability concerns: review of author judgments about each domain presented as percentages for the studies included using QUADAS-2.

Synthesis of included studies

Sensitivity for individual studies ranged from 60% to 92%, with pooled estimates of 80.5% (95% CI 76.5–83.9%), and Higgins I² statistics demonstrated moderate heterogeneity (I²=56%). As for specificity, it ranged from 46% to 96%, with pooled estimates of 69.1% (95% CI 62.3–75.2%), and I²=83% suggested the heterogeneity was more substantial compared to sensitivity. Coupled forest plots of the detailed sensitivity and specificity are presented in Figure 3, and HSROC curve plot with summary point along with 95% CI area is presented in Figure 4.

Figure 3.

Coupled forest plots of pooled sensitivity and specificity. Numbers are pooled estimates with 95% CI in parentheses. Corresponding heterogeneity statistics are provided at bottom right corners. Horizontal lines indicate 95% CIs.

Figure 4.

Hierarchic summary ROC plots with summary point and 95% CI areas for the overall stage of T3 detection.

Abbreviation: ROC, receiver operating characteristic.

Subgroup analysis

As there were several different techniques or methods that were exploited throughout included studies, we performed multiple subgroup analyses to evaluate various clinical settings. Regarding 8 studies that used LCC alone as objective criteria,^{2,21,22,24–26,28,29} pooled sensitivity for them was 81.1% (95% CI 76.0–85.3%), with specificity of 66.6% (95% CI 57.6–74.5%). For another quantitative assessment of ADC value,^17–19,27 pooled sensitivity was 85% (95% CI 77–90%) with specificity of 72% (95% CI 50–87%), higher than LCC both for sensitivity and for specificity. Summary estimates of sensitivity and specificity for ECE were 83.3% (95% CI 77.0–88.2%) and 70.6% (95% CI 60.2–79.2%), respectively. 1.5T scanners were used in 5 studies, and pooled sensitivity for them was 82.6% (95% CI 76.2–87.6%), with specificity of 68.0% (95% CI 56.4–77.8%). It is interesting that the 3.0T scanners performed slightly inferior to 1.5T scanners for sensitivity (79.8%, 95% CI 74.8–84.0%), but as for specificity, these two magnetic field strength performed comparable (69.6%, 95% CI 61.1–77.0%). Five studies did not use any coil, with the pooled sensitivity of 77.0% (95% CI 64.6–86.0%), and the corresponding specificity was 63.9% (95% CI 53.7–72.9%).^{23,24,27,29,31} While for studies that used at least one type of coil (endorectal, PPA or body), the diagnostic accuracy was higher, with 81.8% (95% CI 78.4–84.8%) for sensitivity and 71.2% (95% CI 62.6–78.5%) for specificity. Meta-regression manifested that there was no substantially significant difference regarding coil types, magnet field strength, and analysis method (Table S1). Deeks’ funnel plot with the P-value of 0.37 for the slope coefficient suggested that the likelihood of publication bias was low (Figure 5).

Table S1.

Subgroup analysis and meta-regression

Covariant		Sensitivity (95% CI)	P	Specificity (95% CI)	P
Field strength 5/9	1.5	82.62%/76.24–87.57%	0.64	68.02%/56.38–77.78%	0.89
3.0	80.76%/75.22–85.31%	68.99%/59.91–76.81%
Coil 4/10	None	80.39%/69.04–88.28%	0.78	62.69%/51.13–72.95%	0.23
Any	81.86%/77.82–85.30%	71.01%/62.32–78.39%
Analysis 10/4	Patient	81.9%/77.48–85.62%	0.74	68.92%/59.06–77.32%	0.95
Lesion	80.27%//71.61––86.78%	69.34%/60.99–76.58%
Disease 7/7	ECE	81.43%/74.5–86.81%	0.94	70.37%/62.05–77.54%	0.65
EPE	81.15%/75.73–85.58%	67.18%/55.26–77.24%
Assessment 6/8	ADC	81.41%/73.47–87.39%	0.95	70.69%/59.81–79.63%	0.54
LCC	81.13%/76.04–85.35%	66.59%/57.59–74.51%
Combination	-	81.43%/77.44–84.86%	-	68.68%/61.69–74.91%	-

Abbreviations: ADC, apparent diffusion coefficient; CI, confidence interval; LCC, length of capsular contact; ECE, extracapsular extension; EPE, extraprostatic extension.

Figure 5.

Deeks’ funnel plot, a P-value of 0.37 suggests that the likelihood of publication bias is low.

Discussion

In this meta-analysis, we evaluated the diagnostic performance of mpMRI for local PCa staging as introduced ADC. Compared to an earlier meta-analysis for local staging PCa, results from our showed higher sensitivity but lower specificity.¹¹ One possible explanation was that in the current meta-analysis, all of the data extracted from included studies were quantitative assessments, and it has established the fact that compared to subjective analysis, objective assessment was prone to acquire higher sensitivity but at the cost of decreased specificity.^2,22 At present, the diagnosis of EPE primarily depended on the experience and expertise of radiologists, and several studies have demonstrated that more experienced readers would obtain more accurate results. The PI-RADS scoring systems^5,6 are still considered as limited and have not been widely validated in clinical application for diagnosis of local advanced PCa.³³ In overall accuracy, it seemed that there was no difference between subjective analyses and quantitative assessment by introducing ADC. However, considering that quantitative evaluation is not depend on readers' expertise and experience, then lead to better reproducibility and inter-reader agreement, it therefore could be regarded as a promising approach for diagnosis of PCa.

The most used additional functional protocol in this meta-analysis was DWI; in fact, it was performed by all of the included studies, and the second most used protocol of DCE-MRI was reported by 12 studies. By comparison, in an earlier meta-analysis performed by de Rooij et al, only less than half of total included studies used additional functional protocols.¹¹ Specifically, there were 21 studies that used DCE-MRI and 16 studies that used DWI (some used both of them). In the PI-RADSv2 release in 2015, DCE-MRI was regarded as secondary to DWI for the peripheral zone and T2WI for transition zone. Although there still exists controversy over the role of DCE-MRI in PCa, it seems that DCE-MRI is not as important as before. It is considered that compared to 1.5T, 3.0T MRI scanners could obtain higher sensitivity. However, findings from our studies suggested a different situation, that studies performed on 1.5T scanners had higher sensitivity. Partly because 4 in 5 studies with 1.5T MRI used at least one type of coil, and according to reports, it is suggested that 1.5T MRI with ERC has comparable performance to 3.0T.¹¹ However, higher field strength using ERC will decrease the diagnostic accuracy, and our findings are in line with the earlier reports. It should be noted that some studies reported that endorectal coil likely outperforms pelvic phased array coil in detection of ECE, moreover, endorectal coil was considered significantly superior to body coil.²¹ As the evidence increasing that ADC may play an important role in improving the sensitivity for detection of local PCa, more and more studies regarded it as routine. It is considered that tumor ADC value is considered to correlate inversely with GS and could combine with other nomograms to predict EPE; moreover, ADC was proven to be an effective independent predictor of ECE, and have equivalent capacity compared to T2WI. For 4 studies that adopted ADC value as independent predictor, summary estimates showed that the diagnostic accuracy (85% for sensitivity and 72% for specificity) was higher than average level.^17-19,27 One study that was not included in this meta-analysis (for language problem) obtained a sensitivity of 83% with specificity of 61%, at the cutoff value of 0.87×10⁻³ mm²/s.³⁴ Different cutoff values were obtained probably on account of MRI manufacture, model, and selection of b-values, and for included studies, the b-values ranged from 0 to 2,000 s/mm², but the values are discrete, and the most used were 0, 50, 100, 400, 500, 800, and 1,000 s/mm². There were 2 studies that used high b-value up to 2,000 s/mm²; however, it seemed that the accuracy was not substantially superior to others.^19,28 Besides the ADC value, Giganti et al looked into the ADC ratio (was obtained by dividing pathological ADC by that of a symmetrical normal area in the same prostate zone), and then obtained a sensitivity of 83%, with a specificity of 72%, slightly better than ADC value (79% and 70%, respectively).¹⁸

For another most used quantitative assessment of LCC, it showed a promising prospect of sensitivity as well. Originated from ultrasound, LCC is defined as “the maximum length of the index lesion contact with the prostatic capsule among all axial section”, and most studies used T2WI sequence to measure this parameter, because of better spatial resolution.²⁴ For 8 studies that adopted LCC as the criteria for PCa, pooled sensitivity appeared inferior to ADC values. However, either Baco et al or Kongnyuy et al reported that with respect to microscopic (focal) ECE, LCC behaved better than subjective analysis and Partin tables, which is an improvement compared to de Rooij et al.^11,21,26 In spite of PI-RADSv2 suggesting that tumor contact length >10 mm is a significant independent predictor of EPE, studies reported that the the optimal threshold varied widely, ranging from 6 to 20 mm, with potential factors including magnet field strength, methodology of the studies regarding measurement of LCC (straight line versus curvilinear length), and separate analysis for focal and non-focal EPE. As there was only one study that regarded a <10 mm value (6 mm) as the best threshold, the other 7 studies regarded a >10 mm value as the best threshold. At present, lacking an appropriate cutoff value for determining EPE is the main obstacle for the wide application of LCC in clinical practice. For two studies investigated use the tumor volume as a predictor of EPE, Lim et al obtained the favorable result with sensitivity of 78.4% and specificity of 73.5% at the cutoff value of 2.1 mm³, by contrast, Rud et al gained the sensitivity of 81% in PZ and 86% in TZ when tumor volume at 0.9 mm³, corresponding specificity were 60% and 50%, respectively.^10,17

Excepting PI-RADS, there were several other nomograms or scoring systems that were developed to improve accuracy in detection of local advanced PCa. One of them is the Partin tables, which based on the combination of PSA level, biopsy GS, and clinical stage to diagnosis of PCa. By comparison another prostate cancer nomogram, the Memorial Sloan-Kettering (MSK) includes additional informations from the result of prostate biopsy. However, both of Partin table and MSK nomogram are under debate in clinical practice,³⁵ and several studies proved that mpMRI performed better than Partin tables.^4,36,37 Our analyses showed that for 5 studies using the PI-RADS guidelines to diagnosis of PCa, the diagnostic accuracy was lower than quantitative assessment. Moreover, sensitivity and specificity varied widely as whether PI-RADS score 3 lesions were regarded as positive or not, and many studies demonstrated that counted score 3 lesions as positive would result in higher sensitivity but lower specificity, and vice versa.^2,25,38 Another widely used scoring system was Likert scale, and a study conducted by Costa et al showed that it perhaps outperforms PI-RADS but the difference was not substantial; moreover, the absence of objective criteria raises the concerns about the reproducibility of it, and then a strategy to standardize the Likert scoring system would be desirable.³³ In our meta-analysis, per-lesion analysis and per-patient interpretation had comparable sensitivity; however, per-patient interpretation showed higher specificity. In view of this, there were only 4 per-lesion analysis studies included, which is not enough to draw a conclusion that which interpretation is superior.

There have been 3 meta-analyses published concerning MRI for local staging of PCa. Engelbrecht et al³⁹ included the studies up to 2,000 and Silva et al⁴⁰ restricted to 1.5T scanners. As for de Rooij et al,¹¹ the current meta-analysis has several aspects different from them. First, all studies included in this meta-analysis were mpMRI, whereas it accounted for less than half in their study. Second, almost all studies in the present meta-analysis used quantitative assessment, which obtained a high sensitivity and moderate specificity. This was opposite to the study of de Rooij et al,¹¹ because they acquired moderate sensitivity and high specificity. One possible explanation was that the main analytical approaches in their study were subjective, while all studies including ours adopted quantitative evaluation. Findings from this study revealed that as an independent predictor, ADC and LCC showed a favorable sensitivity even though accompanied with decreased specificity. At present, mpMRI is considered as the best means available for local staging PCa, and is significantly superior to PSA, DRE, and TRUS, or a combination of them. In addition, it has comparable overall accuracy with Likert scale, Partin tables, and MSK nomogram or even better.

Limitations

There were several limitations in our meta-analysis. First, most of the included studies were retrospective regarding study design, leading to a high risk of bias for patient selection, then possible inflation of the diagnostic sensitivity. Second, several studies did not report the details of blinding to clinical information or were not aware of the PSA levels and GS, which decreased the strength of evidence. Third, our study did not perform the analysis of SVI (T3b), due in part to insufficient data or because the sample size was. Finally, we did not make a comparison between PI-RADSv1 and PI-RADSv2 on diagnostic performance, and studies suggested that there was no substantial difference between them.

Conclusion

Compared to subjective analysis, quantitative assessment could obtain higher sensitivity, but at the cost of decreased specificity. Therefore, on the whole, accuracy was comparable to subjective analysis. However, ADC and LCC still should be viewed as promising prediction tools for PCa, especially considering that quantitative assessments have better reproducibility and inter-reader agreement, and they do not depend on radiologists’ experience and expertise.

Disclosure

The authors report no conflicts of interest in this work.