Genitourinary Imaging
Original Research
Accuracy of Multiparametric MRI for Prostate Cancer Detection: A Meta-Analysis
Audio Available | 
Share
OBJECTIVE. The purpose of this diagnostic meta-analysis was to determine the diagnostic accuracy of multiparametric MRI for prostate cancer detection using anatomic T2-weighted imaging combined with two functional techniques: diffusion-weighted imaging (DWI) and dynamic contrast-enhanced MRI (DCE-MRI).
MATERIALS AND METHODS. We searched electronic databases, including MEDLINE, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) up to February 3, 2012. We included diagnostic accuracy studies using a combination of T2-weighted imaging, DWI, and DCE-MRI to detect prostate cancer with histopathologic data from prostatectomy or biopsy as the reference standard. The methodologic quality was assessed with version 2 of the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool by two independent reviewers. Sensitivity and specificity of all studies were calculated from 2 × 2 tables, and the results were plotted in a hierarchic summary receiver operating characteristic plot.
RESULTS. Seven studies that met the inclusion criteria (526 patients) could be analyzed. The pooled data showed a specificity of 0.88 (95% CI, 0.82–0.92) and sensitivity of 0.74 (95% CI, 0.66–0.81) for prostate cancer detection, with negative predictive values (NPVs) ranging from 0.65 to 0.94. Subgroup analyses showed no significant difference between the subgroups.
CONCLUSION. The high specificity with variable but high NPVs and sensitivities implies a potential role for multiparametric MRI in detecting prostate cancer.
Keywords: biopsy, diffusion-weighted imaging, meta-analysis, MRI, prostate neoplasms
Prostate cancer is the most common noncutaneous cancer in men [1]. Although most types of prostate cancer grow slowly and may need minimal or no treatment, other types are aggressive and can spread quickly. Prostate cancer that is detected early has a better chance of successful treatment. Therefore, detection of prostate cancer in an early stage is important but remains challenging.
The currently used diagnostic tools are digital rectal examination; serum prostate-specific antigen (PSA), a nonspecific blood test; and transrectal ultrasound (TRUS)–guided biopsy, a standardized but untargeted method [2]. Because of the limitations of these available diagnostic tools, much effort is being put into improving the accuracy of prostate cancer detection.
Advances in MRI techniques show potential for improving the diagnostic accuracy of MRI for prostate cancer detection. A recently developed multiparametric MRI approach that combines anatomic T2-weighted imaging with functional data appears to be one of the most promising techniques for prostate cancer detection [3–9]. The addition of functional MRI techniques can provide metabolic information, display altered cellularity, and aid in noninvasive characterization of tissue and tumor vascularity [10]. Although these techniques have not been implemented broadly in daily clinical practice yet, they are increasingly mentioned in prostate cancer guidelines [11]. The latest diagnostic consensus statement by the European Society of Urogenital Radiology (ESUR) recommends anatomic T2-weighted imaging combined with at least two functional techniques: diffusion-weighted imaging (DWI), dynamic contrast-enhanced MRI (DCE-MRI), and optionally MR spectroscopy [2]. The accuracy of this method has, however, not been studied systematically. We therefore performed a systematic review and meta-analysis to determine the diagnostic accuracy of the ESUR recommendation—that is, the recommendation of combining T2-weighted imaging with DWI and DCE-MRI for the detection of prostate cancer.
| Materials and Methods |   |
|---|
We systematically searched the electronic databases MEDLINE (U.S. National Library of Medicine), Embase (Elsevier), and Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Collaboration) to identify all relevant studies. The search strategy involved a filter combining imaging modality keywords with prostate cancer keywords in the title and abstract as follows:
(prostate OR Pca OR PSA OR prostatic) AND (MR OR NMR OR NMRI OR MRI OR magnetic resonance OR ADC OR DWI OR DCE OR diffusion weighted OR dynamic contrast OR multiparametric)
To retrieve additional publications, we manually searched reference lists from the included articles and relevant systematic and narrative reviews on the topic. No restrictions on language or date were used in this comprehensive search. The last search was performed on February 3, 2012. We imported all citations identified by the MEDLINE and Embase searches into a bibliographic database (End-Note, version X5, Thomson Reuters).
We screened all retrieved articles and included studies when they compared T2-weighted imaging and the functional MR techniques DWI and DCEMRI with histologic results from prostate biopsies or prostatectomy specimens as the reference standard in patients with suspected or previously diagnosed prostate cancer. One reviewer performed the first screening of titles and abstracts to select eligible studies. Subsequently, two reviewers independently assessed the eligibility by reading the articles.
To obtain 2 × 2 contingency tables from the included studies, we extracted or calculated true-negative (TN), false-negative (FN), true-positive (TP), and false-positive (FP) results of multiparametric MRI for the detection of prostate cancer. A standardized form was used to extract additional data on patient characteristics, imaging protocols, and methodologic characteristics. The authors of the studies that did not report all sufficient data were asked to provide additional information.
We extracted the following data: patient age, PSA level, Gleason score, previous prostate biopsies, cancer status (suspected or detected), MR imager model and manufacturer, magnetic field strength (in Teslas), use of endorectal coil, use of other coils, T2-weighted imaging sequences, DWI acquisition parameters, DCE-MRI acquisition parameters, use of additional techniques, year of publication, study population, reference standard (prostate biopsy or prostatectomy specimens), patient enrollment, study design, blinding, number of readers, region of interest, and scoring system of each modality and of the combination of modalities.
Quality assessment of the included studies was performed by two independent reviewers using the recently developed version 2 of the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool for diagnostic accuracy studies [12]. Any disagreements were discussed and resolved by consensus.
We constructed 2 × 2 contingency tables and calculated sensitivity and specificity with 95% CIs for each study individually. We drew forest plots to show variation and to explore heterogeneity for sensitivity and specificity and plotted their results on a receiver operating characteristic (ROC) plot. We used the Metadas tool [13] within the statistical software package SAS (version 9.2, SAS Institute) to carry out the meta-analyses. The analyses were imported to RevMan5 (The Nordic Cochrane Center) and used to fit the hierarchic summary ROC plot. Because of the substantial heterogeneity of the included studies, we analyzed subgroups of three clinically relevant covariates: reference standard (prostatectomy and biopsy), method of analysis (region-based and patient-based), and localization of the tumor (peripheral zone and whole gland).
| Results | |
|---|
The systematic literature search identified 10,166 records. Screening the titles and abstracts and removing duplicates yielded 367 potentially eligible studies using multiparametric MRI techniques. Another 241 studies were excluded because they did not determine diagnostic accuracy, leaving 126 studies for formal evaluation (Fig. 1). Seven studies used T2-weighted imaging, DWI, and DCEMRI for the detection of prostate cancer and were included in the meta-analysis [14–20]. Other studies were not included because they did not use T2-weighted imaging, DWI, and DCE-MRI as a combination of multiparametric MRI techniques (n = 103) or did not report a combined analysis of these techniques or available data were insufficient to construct a 2 × 2 contingency table (n = 16). Manual searching reference lists of narrative and systematic reviews, position papers, and editorials did not yield any additional results.
 View larger version (210K) | Fig. 1 —Flowchart shows process used to select studies for meta-analysis. We systematically searched MEDLINE (U.S. National Library of Medicine), Embase (Elsevier), and Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Collaboration) to identify all relevant studies. DCE-MRI = dynamic contrast-enhanced MRI, DWI = diffusion-weighted imaging. |
Patient characteristics, technical parameters, and study design of the seven included studies are presented in Tables 1–6. In total, 526 patients were evaluated, with a median age ranging from 63 to 69 years, a median serum PSA level varying from 7 to 11.1 ng/mL, and a Gleason score ranging from 4 to 10.
All studies were evaluated on their methodologic quality by two independent reviewers using the QUADAS-2 tool [12] (Figs. 2–4). The quality of the seven studies varied. Risk of bias regarding patient selection was low in three of the seven studies [16, 17, 20], whereas four studies had a high risk of bias for patient selection [14, 15, 18, 19]. The high risk was caused by the unavailability of data on patient enrollment and inappropriate exclusion. The risk of bias regarding the index test was low in five studies [14–18] and high in one study [19]; for the remaining study, the risk of bias was unclear because information about blinding was not sufficient [20]. The risk of bias regarding the reference standard was low in two studies [14, 19] and high in five studies [15–18, 20] because these latter studies used TRUS-guided biopsy or transperineal biopsy as the reference standard instead of a prostatectomy specimen. Furthermore, two of the five studies using TRUS-guided biopsy did not report sufficient information about the protocols and number of cores taken [15, 20]. The five studies using biopsy as the reference standard comprise studies with targeted biopsies, which have a lower risk of bias than random biopsies. This factor is taken into account in assessing concerns regarding applicability, where these studies are assigned as having a low risk of bias (see also Fig. 3). Risk of bias regarding flow and timing was low in four studies [16, 17, 19, 20] and high in three studies [14, 15, 18] because these latter studies did not include all patients in the final analysis.
 View larger version (36K) |
 View larger version (38K) |
 View larger version (136K) | Fig. 4 —Methodologic quality summary: review authors' judgment on each individual signaling question from version 2 of Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). |
In total, seven studies including 526 patients were considered in the final analysis. For each study, the number of TPs, FPs, FNs, and TNs are shown in Table 5. Pooled sensitivity and specificity values for all studies were 0.74 (95% CI, 0.66–0.81) and 0.88 (95% CI, 0.82–0.92), respectively. Negative predictive values (NPVs) were high, varying from 0.65 to 0.94, and positive predictive values (PPVs)—with larger variability—ranged from 0.31 to 0.95 (Table 6). Figure 5 shows the hierarchic summary ROC plot with 95% CI area and summary point.
 View larger version (32K) | Fig. 5 —Hierarchic summary receiver operating characteristic (ROC) (solid line) plot with summary point (•) with 95% CI area (circled area). Dashed line = no discrimination line (AUC of 0.5, meaning a worthless test), □ =data from individual studies included in meta-analysis (size of square indicates relative size of study population). |
Regarding the reference standard, there was no significant difference in pooled sensitivity and specificity between studies using prostatectomy and studies using biopsy. The two prostatectomy studies showed a pooled sensitivity and specificity of 0.69 (95% CI, 0.52–0.82) and 0.93 (95% CI, 0.81–0.97), respectively. The five studies using biopsy as the reference test (TRUS-guided biopsy or transperineal biopsy) showed a pooled sensitivity and specificity of 0.76 (95% CI, 0.66–0.84) and 0.86 (95% CI, 0.79–0.91). In addition, Table 7 shows forest plots of the pooled estimates overall and for different subgroups.
The four studies that assessed accuracy based on classification of region level of the prostate showed pooled sensitivity and specificity of 0.71 (95% CI, 0.63–0.78) and 0.89 (95% CI, 0.83–0.94), respectively. The studies reporting on a patient or tumor level could not be pooled because available data were not sufficient.
When subgroup comparisons based on the localization of the analyzed tumors were made, the pooled sensitivity values are comparable in the studies analyzing peripheral zone tumors and studies analyzing the whole prostate gland. The pooled sensitivity and specificity of the studies analyzing peripheral zone tumors were 0.81 (95% CI, 0.75–0.85) and 0.91 (95% CI, 0.67–0.98), respectively. The studies analyzing the whole prostate showed a pooled sensitivity of 0.78 (95% CI, 0.65–0.87) and a pooled specificity of 0.88 (95% CI, 0.80–0.94). We were not able to pool calculated estimates of the transition zone.
| Discussion | |
|---|
The results of this diagnostic meta-analysis on the accuracy of multiparametric MRI for prostate cancer detection using the combination of T2-weighted imaging, DWI, and DCE-MRI revealed a high overall sensitivity and specificity. The overall methodologic quality of the included studies was fair, but large heterogeneity was reported. Nevertheless, subgroup analyses did not show considerable differences among various subgroups.
To date, most studies have reported various sensitivity and specificity values for the accuracy of anatomic T2-weighted imaging with or without one or more additional functional techniques for the detection of prostate cancer. Recently, a systematic review and meta-analysis was published on the diagnostic accuracy of T2-weighted imaging combined with DWI compared with T2-weighted imaging alone [21]. The meta-analysis of the 10 included studies showed a higher diagnostic accuracy for T2-weighted imaging combined with DWI (sensitivity and specificity of 0.72 and 0.81, respectively) than for T2-weighted imaging alone (0.62 and 0.77).
The major strength of this diagnostic meta-analysis is that this study is the first meta-analysis to investigate the accuracy of the combination of anatomic T2-weighted imaging and two functional techniques, DWI and DCE-MRI, as recommended by the ESUR guidelines [2]. Furthermore, we are one of the first groups to undertake a meta-analysis of MRI using hierarchic summary ROC methods that are available using the Metadas macro for SAS [13].
This diagnostic meta-analysis showed that the accuracy of multiparametric MRI shows potential for the detection of prostate cancer. Although the FN rate of 26% still might be too high, TRUS-guided biopsy tends to miss tumors as well, with detection rates of 10–19% on repeat TRUS-guided biopsy [22, 23] and up to 59% on MRI-guided biopsy after two negative TRUS-guided biopsy sessions [24]. However, whether those FNs are clinically significant or insignificant tumors is still open to debate. A future randomized multi-center diagnostic trial comparing TRUS-guided biopsy with multiparametric MRI is needed to study potential benefits, harms, and cost-effectiveness in more detail.
The recommendation of the ESUR of using T2-weighted imaging, DWI, and DCE-MRI for prostate cancer detection is based on expert opinion, and the question remains about whether this strategy is the best multiparametric combination because prospective validation studies have not yet been performed. Of the seven included studies, four studies [16–19] recommend using both DWI and DCE-MRI as additional techniques and show significant differences in performance compared with the use of DWI or DCE-MRI alone. The other three studies [14, 15, 20] show no change in performance or worse results when comparing the combination of T2-weighted imaging, DWI, and DCE-MRI with T2-weighted imaging and DWI.
Some potential limitations should be mentioned. First, considerable heterogeneity was identified among the included studies, with differences in reference tests, prostate regions, patient characteristics, and methods of analyses. Four of seven studies used TRUS-guided biopsy as the reference standard for the primary diagnosis of prostate cancer. However, this technique mainly randomly samples the posterior part of the gland, tends to miss tumors on first systematic biopsy [22], and has been reported to underestimate Gleason grade in 43% of the cases [25]. The studies also analyzed different regions of the prostate and patients with other baseline characteristics, such as PSA level. Two studies included patients with mean PSA values of 20.51 and 19.4 ng/mL [17, 20]; these mean PSA values are considerably higher than levels observed in cases of early phase prostate cancer. In these studies a large tumor burden can be expected, with subsequently a possible higher accuracy of multiparametric MRI. The method of analysis varied among the different studies, with studies using a per-patient or region-based approach. For the studies that used a region-based approach, the prostate was subdivided in a varying number of regions, ranging from two to eight, which can artificially increase the specificity by generating more TN regions. Despite this heterogeneity, the pooled estimates of the different subgroups were comparable with the overall summary values.
Second, despite the small number and heterogeneity of the studies, we decided to pool the results because sensitivity and subgroup analyses showed no significant differences. We believe that there are many good examples of reviews, such as Cochrane reviews, performing a meta-analysis of only a few studies that have been proven useful. Furthermore, this review can be considered as a starting point and can be updated as soon as new evidence becomes available.
Third, the overall estimated sensitivity and specificity values were based on all included studies. Although some studies reported data on peripheral zone, transition zone, and whole gland separately and some reported the results of patient- and region-based analyses, these studies were not equally represented in this analysis.
Fourth, the technical parameters, such as the use of an endorectal coil, field strength, and b values, were not similar among the included studies. Furthermore, the analysis of DCE-MRI could be profoundly influenced by the difference in spatial resolution and temporal resolution and by the pharmacokinetic model that is used. The recently developed ESUR guidelines [2] can be used to fulfill the minimal technical and image acquisition parameter requirements for an MRI protocol to detect prostate cancer.
Finally, scoring systems for reporting the prostate images were not similar in all studies. Several studies used dichotomized scoring to distinguish between normal and abnormal appearance, whereas other studies used 3- or 5-point Likert scales for the overall or separate imaging techniques. ESUR experts developed a structured reporting system (the “Prostate Imaging Reporting and Data System,” or “PI-RADS”) with a standardized subscore for each sequence (T2-weighted imaging, DWI, and DCE-MRI); the sub-scores were summarized in a final score that could range between 1 and 5 [2], similar to the standardized score used for breast MRI (BI-RADS). Although this scoring system has not yet been validated, the ESUR classification, which was recently adopted by the ACR, is the best available guideline for using multiparametric MRI in the diagnosis of prostate cancer.
| References | |
|---|
