Multiparametric MRI (mpMRI) of the prostate is currently the most accurate imaging method for detection, localization, and staging of PCa [3
], which was confirmed by a meta-analysis [4
]. An mpMRI examination includes anatomic imaging sequences (T2-weighted imaging and T1-weighted imaging) and functional imaging sequences (DWI and dynamic contrast-enhanced MRI [DCE-MRI]) and has become a recognized standard method for evaluating clinically significant PCa [5
]. With combined morphologic and functional MRI, mpMRI could help to improve diagnostic sensitivity and specificity correlated to biopsy Gleason grades [6
]. According to the Prostate Imaging Reporting and Data System version 2 (PI-RADSv2) scoring system, adopting mpMRI is recommended as a noninvasive and standard diagnostic tool compared with ultrasound-guided biopsy [7
A recent study showed that the additional value of DCE-MRI to T2-weighted imaging and DWI was minimal: PCa detection was aided by the addition of DCE-MRI in only four of 125 peripheral zone (PZ) tumors ≥ 0.5 mL and 0 of 27 PZ tumors < 0.5 mL [8
]. However, in another study, the authors found that the addition of DCE-MRI to DWI scores in the PZ yielded substantial improvements in the probability of PCa detection [9
]. In yet another study, investigators reported that comparable accuracy could be achieved simply by a combination of bipara-metric MRI (bpMRI) and lesion volume [10
]. In addition, clinical utilization of bpMRI helps to reduce costs and scanning time [10
]. Three previous meta-analyses showed that the addition of DCE-MRI did not improve PCa detection compared with T2-weighted imaging and DWI alone [11
]. In the current study, we conducted a meta-analysis to compare abbreviated bpMRI versus standard mpMRI in diagnosing PCa.
Materials and Methods
This systematic review and meta-analysis were carried out according to the recommendations of Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines. A comprehensive literature search of the PubMed (U.S. National Library of Medicine), Web of Science (Clarivate Analytics), and Cochrane Library databases was performed by two researchers independently to identify studies that evaluated the diagnostic value of bpMRI versus mpMRI for PCa with and using transrectal ultrasound (TRUS)-guided biopsy, MRI-TRUS fusion targeted biopsy, or pros-tatectomy and clinical follow-up data as the reference standard. We limited our search to studies published on or before April 14, 2018. The search strategy was as follows: diagnosis AND (multiparametric MRI OR multiparametric magnetic resonance imaging OR multiparametric MRI prostate OR multiparametric MRI prostate cancer) AND (bpMRI OR ((DWI OR diffusion-weighted imaging OR diffusion weighted MRI) AND (T2-weighted MRI OR T2WI)) OR bpMRI OR biparametric MRI OR biparametric prostate OR biparametric prostate MRI) AND prostate cancer. Medical subject heading terms and variations of each term were used. Language was not confined.
The reference lists of the relevant articles were manually screened to identify other potentially eligible articles. Titles and abstracts were initially screened independently; eligible abstracts were selected by reading the full text and finally were included or excluded. Any disagreement was discussed and resolved by consensus.
Types of studies—Retrospective and prospective studies comparing bpMRI with mpMRI in diagnosing PCa were considered. Studies in which only bpMRI outcomes were investigated were excluded, because these studies disregard essential information about comparison. Studies with insufficient data to create 2 × 2 tables to assess diagnostic accuracy were excluded.
Patients—The following patients were considered as participants: patients with clinical suspicion of PCa based on elevated prostate-specific antigen (PSA) value or an abnormal digital rectal examination (DRE) finding and patients with a previous negative prostate biopsy with continually increasing PSA values or an abnormal finding at DRE performed an interval before MRI examination. For this review, we focused on primary PCa diagnosis. Patients who already had a diagnosis of low-grade PCa and had been on active surveillance were not eligible, and patients with recurrent PCa were also excluded, as this generated multiple biases. The main indicator of observation in this study was detection of PCa regardless of Gleason score and location of the lesion.
Quality Assessment and Data Extraction
Each included study was assessed by two observers (with 8 and 5 years' experience of systematic reviews) using the revised instrument for the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). The observers used tailored QUADAS-2 quality checklists [14
]. Each item was scored as Yes or No or as Unclear when there was insufficient information to make a precise judgment. Discrepancies were resolved by consensus. We used RevMan software (version 5.3, The Cochrane Collaboration) [16
] to display the QUADAS-2 results graphically.
The following data were extracted from each study: the first author's name, year of publication, country where study was performed, sample size, description of study population (age), study design (prospective, retrospective, or unknown), PSA value, location of lesions, characteristics of lesions analyzed, magnetic field strength, b values for DWI techniques, descriptions of interpretations of the diagnostic tests, and description of the reference standard. True-positive, false-positive, false-negative, true-negative, sensitivity, and specificity results for the detection of lesions were extracted, and 2 × 2 contingency tables were constructed.
Heterogeneity was assessed using the Cochran Q test and the Higgins I2
heterogeneity index [17
], where p
≤ 0.5 indicated significant heterogeneity. When I2
< 50%, a fixed-effect model was used for meta-analysis. When I2
> 50%, the random-effects model of DerSimonian and Laird was applied [18
]. Source of heterogeneity was explored through meta-regression analysis, which was performed using Meta-DiSc software (version 1.4, Ramón y Cajal Hospital). If there was obvious heterogeneity, sensitivity analysis was performed. Our primary analysis included all the studies, but the sensitivity analysis was performed after removal of the studies indicating heterogeneity.
The statistical computations of sensitivity, specificity, diagnostic odds ratio (DOR), positive likelihood ratio (LR), and negative LR with 95% CIs were performed using Stata software (version 14, StataCorp). Summary ROC (SROC) curves and area under the SROC curve (AUC) were constructed to assess the diagnostic accuracy capabilities.
The findings of this meta-analysis showed that bpMRI had a similar diagnosis performance compared with mpMRI with almost the same diagnosis sensitivity, specificity, and DOR. These results indicate that the role of DCE-MRI in mpMRI was minimal in diagnosing PCa and that an abbreviated bpMRI protocol consisting of T2-weighted imaging and DWI has the potential to replace standard mpMRI.
The sensitivity and specificity for mpMRI varied in the individual studies, with sensitivities ranging from 0.50 to 0.91 and specificity ranging from 0.80 to 1.00 except in the study by Thestrup et al. [27
]. In the latter study, sensitivity was more than 90% while the specificity was 10.3%, which indicates an obvious threshold effect. The sensitivity of bpMRI ranged from 0.50 to 0.92, and specificity varied from 0.80 to 0.98, showing a similar diagnosis performance with mpMRI. In addition, SROC curve analysis showed that AUCs for mpMRI and bpMRI were very close, 0.88 and 0.89, respectively, and the result of sensitivity analysis was consistent, with an AUC of 0.89 for both. Moreover, posttest analysis showed that the positive LR did not change significantly and the negative LR decreased, indicating a good diagnostic performance.
In clinical practice, the majority of studies included in our meta-analysis used two or more radiologists to interpret the MR images, which was inevitable to avoid interpretation bias, but interobserver agreement was generally excellent or good in these studies; thus, the mean sensitivity and specificity were adopted in statistical analysis. PI-RADS or PI-RADSv2 or confidence scale or 5-point scale was adopted to assign the possibilities of PCa in these studies. Although the threshold for PCa diagnosis was not the same, the majority of the included studies used a threshold of PI-RADS ≥ 3. However, three studies [19
] did not report the grading system used for PCa diagnosis, and one study reported that no grading was used [27
]. In the latter study [27
], images were categorized only according to harboring significant PCa or not, which may lead to interpretation bias.
We found a marked heterogeneous effect among different studies, with I2
far greater than 50%, even up to more than 90%. The fact that patient selection was not clearly reported as consecutive or random and that there may have been inappropriate exclusions may have introduced selection bias. A threshold not reported or prespecified may have led to interpretation bias. The study by Thestrup et al. [27
] showed an obvious threshold effect, which may also have introduced interpretation bias. The fact that the same reference standard was not used for all studies may have led to flow and timing bias. Furthermore, we used meta-regression analysis to explore the sources of heterogeneity, and our results showed that research design was a source of heterogeneity. In this case, we performed sensitivity analysis, according to the study design in retrospective or prospective manner, and the results showed that sensitivity and specificity were not affected, which indicates that the results are robust.
Most of the included studies used 3-T MRI scans, but a study conducted in 2011 using 1.5-T MRI showed no difference in performance [24
]. At present, the use of 3-T MRI of the prostate is widespread because of its higher image quality; however, low-field-strength MRI, which conforms to European Society of Urogenital Radiology (ESUR) guidelines, is also acceptable.
According to ESUR guidelines [29
], a PI-RADS score of 3 means that findings are suspicious for clinically significant cancers and a PI-RADS score of 4 means that clinically significant cancer is likely to exist. Two included studies used a cutoff value for PCa diagnosis of PI-RADS 3, whereas two studies [24
] used the cutoff value of a confidential score of 3 [22
]; however, meta-regression analysis did not recognize that the cutoff value systems for the studies in the meta-analysis were a source of heterogeneity. Although PI-RADSv2 has been shown to detect PCa with a higher sensitivity than PI-RADS version 1 (PI-RADSv1) with no significant difference in specificity [30
], we did not find any difference between PI-RADSv2 and the other scoring systems, such as the confidence scale that also uses five categories of scoring to indicate the likelihood of PCa. This may be due to the intrinsic similarity of the prostate scoring system [8
] and the good consistency between the observers. A previous study has shown a need to further determine whether a lesion with PI-RADS 3 needs to be biopsied [31
]; however, according to the results of this study—that is, the high sensitivity and high specificity of the meta-analysis, biopsy of PI-RADS 3 lesions should be recommended to avoid missing a PCa diagnosis. These results are consistent with the results reported by Schoots [32
Approximately 70–75% of PCa lesions occur in the PZ [33
]. Of the 10 studies included in this meta-analysis, five studies did not report specific localization of the lesions, three studies reported the lesions were located at the PZ or TZ, one study reported lesions in the TZ, and one study reported lesions in the TZ and anterior fibrous stroma. Therefore, PZ PCa and TZ PCa were not separated because of the limited studies in this primary meta-analysis. Future studies could further investigate the differences between PZ PCa and TZ PCa. The difference between PI-RADSv2 and PI-RADSv1 is that the imaging sequences are weighted according to the lesion location in the PZ or TZ to improve the performance of diagnosis. In PI-RADSv2, the role of DCE-MRI is secondary to T2-weighted imaging and DWI and DCE-MRI is expressed as positive or negative only [34
]. DWI is defined as being dominant for evaluation of PZ lesions, and T2-weighted imaging is dominant for evaluation of TZ lesions. Further sequences will contribute to only the overall PI-RADS score if an intermediate finding—more precisely, a lesion of PI-RADS score 3—in the dominant sequence is found. In this case, a second sequence (DCE-MRI of PI-RADS 4 or 5 lesion in the PZ and DWI of PI-RADS 5 lesion in the TZ) can upgrade the lesion score to an overall PI-RADS 4. Therefore, DCE-MRI does not contribute to the TZ PI-RADS score, and T2-weighted imaging does not have relevance in scoring PZ lesions [35
]. A recent study has confirmed the superiority of DWI and T2-weighted imaging compared with DCE-MRI and T2-weighted imaging [12
]. In view of the time and cost of DCE-MRI, more studies have suggested that DCE-MRI should be omitted [36
]. Moreover, performance of DWI has been found to be similar to that of DCE-MRI for PCa diagnosis [11
]; bpMRI becomes an attractive proposition as a quick screening examination in biopsy-naive patients [36
]. Our results showed that bpMRI and mpMRI were not statistically different both in sensitivity and specificity and in DOR values; thus, these results suggest that bpMRI has the potential to replace the tedious and time-consuming mpMRI as a simple solution.
The strengths of this systematic review include adequate studies in the literature available for meta-analysis, meta-regression analysis to find possible causes of heterogeneity, and credible and robust sensitivity analysis. However, it also has limitations. The sources of heterogeneity between studies were hard to identify even after excluding heterogeneity of study design. Four studies did not report a PCa diagnosis cutoff value, which may have caused bias. PZ PCa and TZ PCa were not divided into subgroups because of the limited number of studies in this initial metaanalysis. Future studies may investigate the differences between PZ PCa and TZ PCa.
In conclusion, abbreviated bpMRI was verified to be similar in diagnostic efficacy to mpMRI; therefore, abbreviated bpMRI may be of potential value in clinical practice.