Accuracy of multiparametric magnetic resonance imaging for detecting extracapsular extension in prostate cancer: a systematic review and meta-analysis

Abstract

Objective:

To evaluate the diagnostic accuracy of multiparametric MRI (mpMRI) for detecting extracapsular extension (ECE) in patients with prostate cancer (PCa).

Methods and materials:

We searched MEDLINE, PubMed, Embase and the Cochrane library up to December 2018. We included studies that used mpMRI to differentiate ECE from organ-confined PCa with a combination of T₂ weighted imaging (T₂WI), diffusion-weighted imaging, and dynamic contrast-enhanced MRI. All studies included had pathological diagnosis with radical prostatectomy. Two reviewers independently assessed the methodological quality of included studies by using Quality Assessment of Diagnostic Accuracy Studies 2 tool. We calculated pooled sensitivity, specificity, positive and negative predictive values, diagnostic odds ratios and receiver operating characteristic curve for mpMRI from 2 × 2 tables.

Results:

A total of 17 studies that comprised 3374 participants were included. The pooled data showed a sensitivity of 0.55 (95% confidence interval 0.43, 0.66]) and specificity of 0.87 (95% confidence interval 0.82, 0.91) for extracapsular extension detection in PCa.

Conclusion:

First, our meta-analysis shows moderate sensitivity and high specificity for mpMRI to differentiate ECE from organ-confined prostate cancer before surgery. Second, our meta-analysis shows that mpMRI had no significant differences in performance compared with the former meta-analysis with use of T₂WI alone or with additional functional MRI.

Advances in knowledge:

It is the first meta-analysis to evaluate the accuracy of mpMRI in combination of TWI, diffusion-weightedimaging and dynamiccontrast-enhanced-MRI for extracapsular extension detection.

Introduction

Preoperative accurate diagnosis of extracapsular extension (ECE) is conducive to disease control. Compared with organ-confined disease, extracapsular expansion of prostate cancer is related to a reduction in overall and cancer-specific survival after radical prostatectomy (RP).^1,2 Patients receiving neurovascular bundle (NVB)‐sparing proc edures can partly reduce the rates of erectile dysfunction and incontinence, however, NVB‐sparing procedures may result in biochemical recurrence and treatment failure in patients with ECE, due to the higher risk of positive surgical margins.³ Therefore, knowledge of tumor stage and possible ECE may be helpful for patients consultation and selecting patients suitable for NVB‐sparing procedures.

Many nomograms have been constructed to improve pre-operative risk assessment of ECE, such as Partin tables, the Memorial Sloan-Kettering (MSK) prostate cancer nomogram. Partin tables included prostate-specific antigen (PSA), biopsy Gleason score, and clinical stage.⁴ MSK prostatecancer (PCa) nomogram included additional percent positive cores and the percent cancer in the biopsy specimen.⁵ Those methods of cancer detection are limited in their specificity, sensitivity, or both.

In recent year, multiparametric MRI (mpMRI), using T₂ weighted imaging (T₂WI) combined with diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) imaging,⁶ has improved the sensitivity and specificity of determining tumor staging and extracapsular extension.⁷ The additional functional MRI techniques can provide high-contrast resolution morphologic imaging, metabolic information and direct depiction of tumor vascularity, with the potential to improve the diagnostic accuracy of MRI in PCa.⁸ However, the reported accuracy of mpMRI is highly variable and has not been studied systematically. The aim of this systematic review and meta-analysis was to determine the diagnostic performance of mpMRI for the detection of ECE in prostate cancer.

Methods And materials

Literature search

We performed a systematic literature search in MEDLINE, PubMed, Embase and the Cochrane library to identify studies that assess the diagnostic accuracy of mpMRI in differentiating ECE from organ-confined disease. The literature search was restricted to English-language publications and human subjects. The following search terms and keywords were used: “multiparametric”, “magnetic resonance imaging”, “mpMRI”, “extracapsular extension”, “extracapsular invasion”, “ECE”, “prostate cancer”, “PCa”. All relevant and related articles on the database before December 2018 were searched.

Study selection

The inclusion criteria for articles were as follows: (1) patients had histologically proven PCa; (2) mpMRI included T₂WI combining with DWI and DCE imaging or at least two functional MR techniques; (3) the reference standard was pathological diagnosis; and (4) studies were required to allow construction of 2 × 2 tables. The exclusion criteria were review, editorials, conference abstracts, commentaries and case reports. This was undertaken as a two-stage process, excluding on title and abstract first and then on review of full papers. Two reviewers (ZF and LC) evaluated the eligibility of the studies included independently. Any disagreements were discussed and resolved by consensus.

Data extraction

To acquire 2 × 2 table from the included studies, the true-negative (TN), false-negative (FN), true-positive (TP), and false-positive (FP) results of mpMRI for the detection of ECE were extracted. Data on study characteristics and imaging characteristics were extracted in our article. Study characteristics included patient age, number of patients, study design, reference standard and number of readers; imaging characteristics included field strength, type of coil, MR sequences and features. The review protocol was registered and available from the authors on request.⁹ One reviewer (ZF) extracted data from the included studies and developed a standardized data table. Another reviewer (LC) helped resolve unclear issues in consensus.

Quality assessment

Two reviewers independently assessed the included study quality by using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool¹⁰ for diagnostic accuracy studies. The tool contains four domains: (a) patient selection, which describes how to select the patients; (b) index test, which describes how the test was conducted and how to interpret the results; (c) reference standard, which describes how the reference standard was conducted and how to interpret the results; (d) flow and timing, which describes the flow of patient inclusion and exclusion and the interval between the index test and reference standard.

Statistical analysis

We summarized 2 × 2 tables of TP, FP, FN, and TN to calculate sensitivity and specificity values for differentiating ECE from organ-confined disease. When different cut-off thresholds were reported in a study, we followed the most clinically suitable one; when different readers were reported to present results, the most experienced or first reader was selected.

All analyses were conducted by using software (Stata 14; Stata, College Station, TX and Review Manager 5.3, Cochrane Collaboration, Oxford, UK and MetaDisc 1.4; Ramón y Cajal Hospital, Madrid, Spain). We used a bivariate random-effects model to calculate pooled sensitivity and specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratios and their corresponding 95% confidence intervals (CIs). PLR was identified as the likelihood that the MRI in combination of additional techniques result positive for differentiating ECE from organ-confined disease, and NLR was identified as the MRI in combination of additional techniques result negative for differentiating ECE from organ-confined disease. The diagnostic odds ratio was identified as the odds of having a positive MRI in patients with ECE compared with a positive MRI in patients without ECE.

Forest plots were drawn to explore variation and to calculate heterogeneity for sensitivity and specificity, and results were shown on a receiver operating characteristic (ROC) curve. To explore the heterogeneity in test, we used the inconsistency index (I² value) and Cochran’s Q test (p-value) for each forest plot. p-value less than 0.05 indicated that the difference was considered significant and I² values greater than 50% indicated heterogeneity in studies. To research heterogeneity, sensitivity analysis of four clinically relevant factors were performed: magnetic field strength (1.5 or 3.0 T); study design (retrospective or prospective); use of the coil (a pelvic phased‐array coil or an endorectal coil); QUADAS-2 applicability risk (high risk or no high risk). And publication bias was assessed by a Deeks funnel plot asymmetry test.

Results

Literature search

Figure 1 provides an overview of the literature search and study selection. The initial search yield 436 records, and 69 studies were identified to be eligible after reviewing the titles and abstracts. The full text of these eligible studies was assessed. Of the 69 eligible studies, 52 studies were excluded for the following reasons: not English-language publications (n = 4), or not a combined accuracy of mpMRI techniques was reported in study(n = 34), or no available data were reported to construct a contingency 2 × 2 table (n = 14). In total, 17 studies used T₂WI combining with DWI and DCE-MRI for the detection of ECE and were included in the meta-analysis.

Figure 1.

Flowchart of study selection.

Data extraction and quality assessment

Table 1 provides a summary of study characteristics and MRI characteristics. Figure 2 shows that the methodologic quality (risk of bias and concerns regarding applicability) of all included studies was evaluated by the QUADAS-2 tool. We found that most included studies had a low risk of bias and low concerns regarding applicability. In the patient selection domain, six studies had a high risk of bias because of inconsecutive patients enrolled, case–control design or inappropriate exclusions.^{13,14,21,23,25,26} Only one study had a high concern regarding applicability because only patients with high-grade tumor were included.¹⁴ In the index test domain, seven studies had unclear risk of bias, because the information that MRI interpretation was blind to the reference standard was not provided.^{12,14–18,20} In the reference standard domain, one study had an unclear risk of bias and an unclear concern regarding applicability, because the reference standard were obtained without blinding to the index test.²⁷ In the flow and timing domain, four studies had a high risk of bias, because few patients were not included in the 2 × 2 table.^11,13,23,25

Table 1.

Principal characteristics of the included studies

Study Characteristics								Imaging characteristics
No.	Study	Year	Age	No. of patients	Design	Reference standard	No. of readers	Field strength (T)	Type of coil	MR sequences	Features(s/mm2)
1	Boesen11	2015	65	87	P	RP	2	3	PA	T2WI,DWI,DCEI	100, 800, 1400
2	Cerantola12	2013	67	60	R	RP	2	3	ER	T2WI,DWI,DCEI	NR
3	Davis13	2016	61	133	P	RP	NR	3	PA, ER	T2WI, DWI, DCEI	NR
4	Dominguez14	2018	61	79	R	RP	1	1.5	ER	T2WI, DWI, DCEI	NR
5	Feng15	2015	62.8	112	R	RP	1	3	PA	T2WI, DWI, DCEI	400, 800
6	Gaunay16	2016	NR	74	R	RP	2	3	PA, ER	T2WI, DWI, DCEI	NR
7	Hegde17	2012	NR	118	R	RP	1	3	PA, ER	T2WI, DWI, DCEI	500, 1400
8	H Lee18	2017	65	1145	R	RP	2	1.5 or 3	PA	T2WI, DWI, DCEI	NR
9	KayatBittencourt19	2015	NR	133	R	RP	1	3	PA, ER	T2WI, DWI, DCEI	50, 400, 800
10	Lee20	2017	60.9	48	R	RP	1	3	PA, ER	T2WI, DWI, DCEI	NR
11	Lista21	2014	63.7	85	P	RP	2	1.5	ER	T2WI,DWI,DCEI	50, 550, 1000
12	Martini22	2018	NR	589	R	RP	NR	3	PA	T2WI, DWI, DCEI	50, 2000
13	Raskolnikov23	2015	61	169	P	RP	NR	3	ER	T2WI,DWI,DCEI	NR
14	Somford24	2013	NR	183	P	RP	2	3	ER	T2WI, DWI, DCEI	0, 50, 500, 800
15	Tanaka25	2013	NR	67	P	RP	2	3	PA	T2WI,DWI,DCEI	NR
16	Toner26	2016	63.2	152	R	RP	2	1.5 or 3	NR	T2WI, DWI, DCEI	NR
17	W Kim27	2016	64	292	R	RP	2	3	PA	T2WI, DWI, DCEI	0, 100, 1000

DCE, dynamic contrast-enhanced; DWI, diffusion-weighted imaging; ER, endorectal coil; NR, not reported; P, prospective; PA, pelvic phased‐array coil; R, retrospective; RP, radical prostatectomy; T₂WI, T₂ weighted imaging.

Figure 2.

Stacked bar charts show the assessment of risk of bias and applicability concerns of included studies. QUADAS-2 scores of methodologic study quality are expressed as a percentage of studies that met each criterion. For each quality domain, the proportion of included studies that were suggested to have low, high, or unclear risk of bias and/or concerns regarding applicability is displayed in green, yellow, and red, respectively. QUADAS-2, Quality Assessment of Diagnostic AccuracyStudies 2.

Figure 3 shows the slope coefficients for the Deeks funnel plot for the detection of ECE (p = 0.45), which illustrates no asymmetry in the data.

Figure 3.

Deeks funnel plots shows the presence of publication bias. Numbers in the circles correspond to included studies numbers. ESS = effective sample size.

Data analysis

17 studies with 3374 participants were included in the meta-analysis. The primary analysis showed that the pooled sensitivity, specificity, PLR, NLR, and diagnostic odds ratios for diagnostic accuracy of mpMRI in detecting ECE in prostate cancer were 0.55 (95% CI 0.43, 0.66), 0.87 (95% CI 0.82, 0.91), 4.2 (95% CI 3.1, 5.8), 0.52 (95% CI 0.40, 0.67) and 8 (95% CI 5, 13), respectively (Figure 4). The area under the ROC curve was 0.83 (95% CI 0.79, 0.86) (Figure 5). And the forest plots suggest that heterogeneity was high, because I² values for sensitivity and specificity exceeded 50%.

Figure 4.

Forest plots show the sensitivity and specificity of included studies. The Cochran’s Q test (p-value) and inconsistency index (I² value) are measurements of heterogeneity.

Figure 5.

Hierarchical SROC plots of mpMRI for differentiating ECE from OC prostate cancer. Numbers in the circles correspond to included studies numbers. ♦=estimate of sensitivity (SENS) and specificity (SPEC). Values in brackets are 95% CIs. CI,confidence interval; ECE, extracapsular extension; OC, organ-confined; SROC, summaryreceiver operating characteristic.

Subgroup analysis

Table 2 shows the results of subgroups to investigate the influence of study design, technical details on pooled sensitivity, specificity, PLR, NLR, DOR. 13 studies with a 3.0 T MRI had equivalent sensitivity 0.57 (95% CI 0.43, 0.71), lower specificity 0.85 (95% CI 0.80, 0.91), compared with two studies with a 1.5 T MRI had sensitivity 0.57 (95% CI 0.44, 0.70), specificity 0.94 (95% CI 0.88, 0.98). The statistics may have a bias, because few studies with a 1.5 T device were included.

Table 2.

Sensitivity analyses performed for subgroups of studies

Analysis	No.of studies	Sensitivity (%)	Specificity (%)	PLR	NLR	Diagnostic odds ratio
Overall	17	0.55 (95% CI: 0.43, 0.66)	0.87 (95% CI 0.82, 0.91)	4.2 (95% CI 3.1, 5.8)	0.52 (95% CI 0.40, 0.67)	8 (95% CI 5, 13)
Magnetic field
1.5 T	2	0.57 (95% CI 0.44, 0.70)	0.94 (95% CI 0.88, 0.98)	10 (95% CI 1.7, 68.3)	0.46 (95% CI 0.34, 0.61)	23 (95% CI 3, 164)
3.0 T	13	0.57 (95% CI 0.43, 0.71)	0.85 (95% CI 0.80, 0.91)	3.9 (95% CI 2.7, 5.6)	0.51 (95% CI 0.36, 0.71)	8 (95% CI 4, 14)
Design
Retrospective	11	0.56 (95% CI 0.41, 0.70)	0.86 (95% CI 0.79, 0.91)	4.1 (95% CI 2.7, 6.2)	0.51 (95% CI 0.37, 0.70)	8 (95% CI 4, 15)
Prospective	6	0.53 (95% CI 0.34, 0.71)	0.87 (95% CI 0.80, 0.93)	4.4 (95% CI 2.6, 7.6)	0.54 (95% CI 0.36, 0.79)	8 (95% CI 4, 19)
Coil
PA	6	0.71 (95% CI 0.52, 0.85)	0.84 (95% CI 0.78, 0.88)	4.4 (95% CI3.4, 5.7)	0.34 (95% CI 0.20, 0.59)	13 (95% CI 7, 25)
ER	5	0.53 (95% CI 0.46, 0.60)	0.89 (95% CI 0.79, 0.95)	4.8 (95% CI 2.3, 10.2)	0.53 (95% CI 0.43, 0.65)	9 (95% CI 4, 223)
QUADAS
No high risk	11	0.61 (95% CI 0.46, 0.74)	0.85 (95% CI 0.78, 0.90)	4.1 (95% CI 2.8, 5.9)	0.46 (95% CI 0.32, 0.65)	9 (95% CI 5, 16)
High risk	6	0.42 (95% CI 0.28, 0.58)	0.90 (95% CI 0.83, 0.95)	4.4 (95% CI 2.4, 8.2)	0.64 (95% CI 0.49, 0.83)	7 (95% CI: 3, 16)

CI, confidence interval.

11 retrospective studies had higher sensitivity 0.56 (95% CI 0.41, 0.70), equivalent specificity 0.86 (95% CI 0.79, 0.91), compared with 6 prospective studies with sensitivity 0.53 (95% CI 0.34, 0.71), specificity 0.87 (95% CI 0.80, 0.93).

Six studies with a pelvic phased‐array coil had higher sensitivity 0.71 (95% CI 0.52, 0.85) and lower specificity 0.84 (95% CI 0.78, 0.88), compared with five studies with an endorectal coil had sensitivity 0.53 (95% CI 0.46, 0.60) and specificity 0.89 (95% CI 0.79, 0.95).

Six studies with high risk in QUADAS score had lower sensitivity 0.42 (95% CI 0.28, 0.58) and higher specificity 0.90 (95% CI 0.83, 0.95), compared with eleven studies with no high risk had sensitivity 0.61 (95% CI 0.46, 0.74) and specificity 0.85 (95% CI 0.78, 0.90).

Discussion

Our meta-analysis showed that mpMRI had pooled sensitivity and specificity of 55 and 87%, respectively, for differentiating ECE from organ-confined disease in combination of T₂WI, DWI and DCE-MRI. The methodologic quality used in the included studies was fair, but large heterogeneity existed. Meanwhile, considerable difference in subgroup analyses was not seen among various subgroups.

The quantitative parameters derived from mpMRI have been introduced into the assessment of pretreatment PCa, because they can show histologic characteristics such as tumor aggressiveness,^28,29 angiogenesis,²⁹ and tissue composition.³⁰

In the detection of ECE, T₂WI allows the best description of prostate region, and provides cancer detection, location and staging.^31,32 In T₂WI sequences, prostate cancer appears weak contours of low signal intensity. Cases suspected with ECE show neurovascular thickening, capsular irregularities, bulge or loss of capsule and enhancement of the prostatic capsule. DWI has played a positive role in ECE detection and seminal involvement assessment.³¹ Quantitative and qualitative assessment of tumor aggressiveness and Gleason score were provided by the apparent diffusion coefficient (ADC).^31,33,34 The diagnostic criteria on DWI were presence of tumors showing hyperintensity with high b-value and hypointensity in ADC map.³⁵ The lowest values are the typical features of tumors and data suggest that values in the range of 0.49–1.43 × 10³ mm²/s are in relation with tumorigenesis.³⁰ The diagnostic criteria on DCE were a focal bulging of asymmetrically bright color beyond the capsule seen on color‐coded pharmacokinetic images overlaid on T₂WI. Due to the lack of conclusive data, the studies using spectroscopy MR are not included in our meta-analysis according to European Society of Urogenital Radiology (ESUR) guidelines.³⁵

Though, it is known that combination of T₂WI, DWI and DCE sequences shows better sensitivity and specificity than T₂WI, the diagnostic accuracy of mpMRI remain controversial in cancer detection. Several studies^11–27 have a wide range of partially conflicting results for mpMRI in detecting ECE with reported sensitivity ranging from 13 to 92% and specificity ranging from 56 to 98%.

Recently, a meta-analysis³⁶ including 45 studies was reported on diagnostic accuracy of T₂WI with or without one or more additional functional techniques and showed sensitivity and specificity of 57 and 91% for the detection of ECE. However, this meta-analysis did not report diagnostic accuracy of mpMRI for the detection of ECE due to the lack of enough studies published on mpMRI for ECE detection. Four studies included^12,17,24,25 overlapped between two meta-analysis.

Until now, many studies^11–27 have showed various sensitivity and specificity for accuracy of mpMRI for differentiating ECE from organ-confined disease in combination of T₂WI, DWI and DCE-MRI. The purpose of our meta-analysis is to investigate diagnostic accuracy of mpMRI including a combination of T₂ weighted images and two additional functional MRI techniques, DWI and DCE-MRI, as recommended by ESUR.³⁵

Our meta-analysis showed that mpMRI had moderate sensitivity and high specificity for the detection of ECE.

First, our findings showed no significant differences in performance compared to using T₂WI alone or with additional functional MRI published by previous meta-analysis.³⁶ So, whether this strategy using T₂WI, DWI and DCE-MRI for the detection of prostate cancer and ECE recommended by ESUR guideline is the best multiparametric combination or better than T₂WI alone remains in doubt.

Second, our findings suggest that the imaging techniques should not be used as first-line diagnostic test to detect locally advanced prostate cancer. In fact, a high sensitivity is the basic requirement to choose a test for screening. However, high specificity of mpMRI results can help urologists minimize excluding patients from curative surgery. To compensate the low sensitivity of mpMRI, a specific nomogram based on prostate-speciffic antigen, DRE and biopsy findings can be introduced to detect ECE, which shows a high sensitivity at 88%.⁵ Considering the results and lack of accurate pre-operative assessment, an appropriate approach to determine TPs would be to combine this sensitive nomogram with the specific value of mpMRI.

We recognize that our meta-analysis had some limitations. First, a limited number of studies included in the meta-analysis due to the strict inclusion criteria. This result in a smaller small sample size and an underestimation of the diagnostic accuracy potentially. Second, 3 studies^14,21,25 of the 17 studies that only comprised intermediate or high-risk cancer patients: clinical stage T2 or T3, may influence accuracy of diagnosis. Third, MR spectroscopy as additional technique was not included in our study, this could influence our overall judgment on the diagnostic accuracy of mpMRI. It has been reported that addition of volumetric data from MR spectroscopic imaging to mpMRI may improve local staging and reduce interobserver variability.³⁷ Fourth, due to the lack of enough studies with DWI or DCE alone in studies included, we fail to compare the diagnostic value between mpMRI and DWI or DCE. Introduction of previous meta-analysis³⁶ may bring a bias for the overlap of four studies between the two meta-analysis. Fifth, the subgroup analysis shows that retrospective studies had a higher sensitivity than the prospective studies, this may lead to methodological heterogeneity rather than clinical heterogeneity. A final limitation of this study is the possibility of publication bias because we could not seek unpublished studies and conclusions of published studies on mpMRI for the detection of ECE may be overly optimistic, as studies with favorable results are more likely to be submitted and published.

Conclusions

First, our meta-analysis shows moderate sensitivity and high specificity for mpMRI to differentiate ECE from organ-confined prostate cancer before surgery. Second, our meta-analysis shows that mpMRI had no significant differences in performance compared with the former meta-analysis with the use of T₂WI alone or with additional functional MRI.