Since the early 1990s, the tumor marker prostate-specific antigen (PSA) has been widely adopted for prostate cancer (PCA) screening. However, the low sensitivity of PSA for detecting PCA (21% of pooled sensitivity using a cutoff of 4.0 ng/mL) [
1] resulted in a lead-time bias, so PCA was diagnosed before the appearance of clinically important PCA that could cause PCA-related morbidity or mortality [
2,
3]. The concept of insignificant PCA was on the rise in this background and was first proposed by Stamey et al. [
4] and followed by Epstein et al. [
5]. Currently, the definition of clinically significant PCA varies considerably among different studies [
6], and the recently revised Prostate Imaging Reporting and Data System version 2 (PI-RADSv2) states that clinically significant PCA is defined on pathology and histology as a Gleason score of 7 or greater (including 3 + 4 with prominent but not predominant Gleason 4 component) and a tumor volume of 0.5 cm
3 or greater with or without extra-prostatic extension (EPE) [
7,
8].
The conventional diagnostic pathway of PCA, consisting of PSA screening followed by random, systemic, transrectal ultrasound (TRUS)-guided biopsy, leads to overdiagnosis of clinically insignificant PCA and subsequent overtreatment [
6,
9]. The overtreatment of PCA can result in significant morbidity to patients, such as urinary incontinence or erectile dysfunction. Therefore, the American Urological Association (AUA) guideline recommends active surveillance (AS) for patients with very-low-risk or low-risk PCA as an alternative to radical treatment and mentions that even focal ablative therapy can be considered as a management option of lowrisk and intermediate-risk localized PCA in the context of a clinical trial [
10,
11]. However, TRUS-guided biopsy can also miss a considerable portion of significant cancer (30–40%) and can underestimate cancer volume or aggressiveness, mainly because of sampling error, especially in the anterior part of the prostate gland [
10,
12]. Consequently, including patients with underestimated PCA in an AS or focal therapy group can result in missing the opportunity for curative resection or other radical treatments.
Multiparametric MRI is expected to provide additional information for detecting significant PCA and may facilitate identification of clinically significant PCA, owing to its high soft-tissue contrast and function-based imaging techniques, such as DWI or dynamic contrast-enhanced MRI (DCE-MRI). Several studies have been conducted for evaluating the performance of multiparametric MRI for detecting clinically significant PCA. In a systemic review, Fütterer et al. [
6] reported that the detection rate of multiparametric MRI was 44–87% for clinically significant PCA. However, the definitions of clinically significant PCA (based on random biopsy, targeted biopsy, or prostatectomy specimen), the lesion allocation methods, and the method of reporting results (i.e., per patient or per lesion) were widely variable in the study. Furthermore, the PI-RADSv2, currently widely used as the standard for reporting MRI findings in patients with PCA, was not used as the image interpretation system.
Thus, the purpose of our study was to evaluate the performance of PI-RADSv2 for detecting all PCA lesions and clinically significant PCA lesions using a one-to-one correlation between pathologically proven lesions and MRI, using a per-lesion analysis, and whole prostatectomy pathology results.
Materials and Methods
This retrospective study was approved by the institutional review board with a waiver of informed consent.
Study Population
From December 2015 to August 2016, 62 consecutive patients underwent radical prostatectomy for resection of PCA at our institute. Patients who underwent prostate MRI at outside hospitals were excluded so that the study group would be composed of patients with MRI examinations with homogeneous sequences and parameters. Thus, 12 patients were excluded and 50 patients were initially included in our study. Among them, six patients were excluded for one of the following reasons: too many PCA lesions (> 6) in the prostate gland (n = 5 patients) or a temporal gap between MRI and surgery of 3 or more months (n = 1). The patients with more than six prostate lesions on pathologic examination were excluded because of potential inaccuracies of one-to-one correlation between the MRI examination and the prostatectomy specimen. In total, 93 PCA lesions from 44 patients were included in the final analysis.
MRI
MRI was performed on a 3-T MRI scanner (Achieva, Philips Healthcare). Before MRI acquisition, 20 mg of butyl scopolamine (Buscopan, Boeh-ringer-Ingelheim) was injected IV for suppression of bowel peristalsis. The details of the multiparametric prostate MRI protocol are as follows: multiplanar T2-weighted imaging in four planes (axial, coronal, sagittal, and oblique axial); oblique axial T1-weighted imaging; oblique axial fat-saturated single-shot echo-planar DWI (b values = 0 and 1000 s/mm
2) with generation of apparent diffusion coefficient (ADC) maps on a voxelwise basis; and DCE-MRI in the oblique axial plane using a 3D T1-weighted spoiled gradient-echo sequence. The DCE-MRI sequence was performed after IV injection of 0.1 mmol/kg of gadoterate meglumine (Dotarem, Guerbet) administered at a rate of 2 mL/s with an automatic injector (Spectris Solaris EP, Medrad). We used an axial oblique reference plane perpendicular to the rectal surface of the prostate because it is similar to the sectioning plane of pros-tatectomy specimens [
19]. The MRI parameters of each sequence are listed in
Table 1.
One-to-One Correlation Between MR Images and Prostatectomy Specimens
At our institute, radiology-pathology meetings for one-to-one correlations of PCA lesions between MR images and prostatectomy specimens have been held monthly since May 2016. The MRI and prostatectomy data during the study period were retrospectively analyzed in the first three meetings. Two radiologists (with 11 and 23 years of experience in radiology) whose subspecialty is genitourinary imaging identified and scored every visible lesion with a PI-RADSv2 score of 3, 4, or 5 in each patient's prostate MRI scan on the basis of PI-RADSv2 [
7,
8] (
Fig. 1). These image analyses were performed in consensus of both radiologists who were blinded to pathologic results. The interobserver agreement for PI-RADSv2 was revealed to be excellent (κ = 0.86–0.87) in a previous study by Muller et al. [
20].
For pathologic evaluation, the prostate specimen was coated with India ink and fixed in 4% buffered formalin. After the distal 5-mm portion of the apex was amputated and coned, the prostate was sliced from the base to the apex along the longitudinal axis at 4-mm intervals, followed by paraffin embedding. Subsequently, microslices were placed on glass slides and stained with H and E. All slices including cancer foci were transferred to a pathologic mapping sheet, and a genitourinary pathologist (with 20 years of experience) recorded the Gleason score (GS) and presence of EPE for every detectable PCA lesion on the mapping sheet independent of MRI findings. If there were six or more PCA lesions in one prostate specimen, the case was excluded from pathologic analysis, as mentioned earlier.
At the consensus meeting, the radiologists and pathologists performed one-to-one correlations of each MRI-identified and pathologically identified lesion, which were numbered from left to right, base to apex. MRI-invisible PCA lesions were also recorded and archived. The tumor volume of all pathologically confirmed PCA lesions was recorded using the following formula: length × width × height × 0.52 (cm3). The length and width were measured on the largest slice based on the mapping sheet. Height was regarded as the sum of intervals of all the slices. The volume of lesions that were too small (longest diameter ≤ 0.1 cm) was not measured; instead, these lesions were recorded as pinpoint lesions.
Definition of Clinically Significant Prostate Cancer
In this study, we followed the definition of clinically significant cancer as defined in PI-RADSv2.
Statistical Analysis
The detection rates of PI-RADSv2 for PCA and clinically significant PCA were calculated in two ways: defining MRI-visible PCA as PI-RADSv2 score 3, 4, or 5 or as only PI-RADSv2 score 4 or 5. The proportions of clinically significant PCAs among MRI-visible and MRI-invisible PCA lesions were also calculated.
ROC curve analyses for the diagnostic performance of PI-RADSv2 were performed, and the AUC values were reported under two different conditions: using a PI-RADSv2 score of 3 as the cutoff or using a score of 4 as the cutoff. The sensitivity and specificity as well as positive predictive value (PPV) and negative predictive value (NPV) at the apex of the ROC curve were also reported using the two different conditions. Pairwise comparisons of AUC values were conducted using DeLong tests [
21].
The differences in tumor volume, total GS, primary and secondary Gleason patterns between MRI-visible and MRI-invisible PCAs and between MRI-visible and MRI-invisible clinically significant PCAs were also evaluated using the t test. Pinpoint lesions were not included in tumor volume analyses. These analyses were performed using a PI-RADSv2 score of 3 as the cutoff for MRI-visible PCA.
MedCalc statistical software (version 17.2, MedCalc Software) was used for statistical analyses. A p < 0.05 was considered statistically significant.
Discussion
In this study, we report the detection rates of PCA and clinically significant PCA using PI-RADSv2 as a multiparametric MRI scoring system on a per-lesion basis. Our results showed that the use of multiparametric MRI with PI-RADSv2 can miss 53.12–57.81% of clinically significant PCA lesions using a cutoff value of PI-RADSv2 score of 3 or 4. The NPVs of PI-RADSv2 were 43.08–45.16% in our per-lesion analysis, which is significantly different from the results of previous studies [
6,
10,
15,
16]. The diagnostic performance, presented as an AUC value, was generally fair (0.69–0.72), slightly lower than those in previous studies by Park et al. [
14] (0.79–0.81) and Zhao et al. [
16] (0.87–0.91) and in a meta-analysis by Woo et al. [
13] (area under hierarchical summary ROC curve = 0.91). However, our results also showed that up to 96.77% of MRI-visible PCAs are clinically significant PCAs; thus, PI-RADSv2 showed a high PPV for detecting clinically significant PCAs.
The concept of incorporating multiparametric MRI into patient enrollment criteria in AS is based on the conclusions of previous studies that multiparametric MRI showed a high NPV for clinically significant PCA, meaning a negative MRI result is deemed to be a good predictor of suitability for AS [
10]. According to the systemic review conducted by Fütterer et al. [
6], NPVs in previous studies ranged from 63% to 98%. Even recently published studies of PI-RADSv2 by Baldisserotto et al. [
15] and Zhao et al. [
16] reported that the NPV of PI-RADSv2 was as high as 55–85%, and 86.5–91.0%, respectively. The possible reasons of our relatively low NPVs are as follows: First, the individual studies included in the systemic review by Fütterer et al. did not use PI-RADS and adopted various definitions of clinically significant PCA. Moreover, the studies were not based on per-lesion analyses but rather on per-patient or per-region analyses, which have the possible risk of mismatch between an actual lesion and an MRI-detected lesion. The studies using PI-RADSv2 as a multiparametric MRI reporting system [
15,
16] were also per-patient analyses, and the reference standard was not a prostatectomy specimen but rather a TRUS-guided biopsy specimen, causing a possible mismatch between a biopsy-proven lesion and a multiparametric MRI–detected lesion. Second, the existence of small (volume < 0.5 cm
3) but significant (GS ≥ 7) cancer foci could affect this relatively low NPV. A previous study by Vargas et al. [
22] showed that using multiparametric MRI with PI-RADSv2 correctly identified only 26% of small but significant tumors in the peripheral zone (PZ) and 20% in the transition zone (TZ). In our study, we found an approximately 29% (7/24) detection rate for small but significant PCA, similar to previous studies.
The possible causes of relatively low detection rates and diagnostic performance (in terms of AUC value) may be explained by speculations similar to those mentioned earlier. Considering the slightly high AUC value, we can conclude that a PI-RADSv2 score of 3 should be regarded as an MRI-visible lesion that needs further clinical evaluation, although this difference in performance was not statistically significant. A previous study with PI-RADSv2 also showed a higher accuracy for detecting significant PCA when considering a PI-RADSv2 score of 3 as positive for PCA [
15]. Further studies of lesions with a PI-RADS score of 3 are required.
Our result of a high PPV (up to 96.77%) when using multiparametric MRI with PI-RADSv2 for detecting significant PCA shows that multiparametric MRI is useful for follow-up of patients undergoing AS. Previous studies of MRI not based on PI-RADS reported moderate PPVs (38–68%) [
6], and recent PI-RADSv2-based per-patient studies showed improved PPVs (77.5–89.3%) [
15,
16]. It seems that PI-RADSv2, as a simplified reporting system, provides greater PPV than previous systems, for both per-patient and per-lesion analyses. Although patient enrollment in studies of AS is on a per-patient basis, follow-up results of enrolled patients could be reported on a per-lesion basis because a new lesion or a lesion showing interval change can lead to initiation of active treatment or the patient can remain in the AS program. Thus, according to our per-lesion analysis, a new MRI-visible lesion that has a high PI-RADS score could be a clinically significant PCA with high probability and may lead to suspending the AS program and the recommendation for further evaluations such as image-guided targeted biopsy or even a change in the treatment strategy to active treatment.
Furthermore, a high PPV of multiparametric MRI with PI-RADSv2 also suggests that this test can replace the annual prostate biopsy during AS follow-up, because the rates of misses of clinically significant PCA for multiparametric MRI with PI-RADSv2 and TRUS-guided biopsy are comparable (30–40% and 53.1–57.8%, respectively) [
11], whereas the patient morbidity and quality of life are much better with multiparametric MRI than with biopsy. The role of multiparametric MRI in AS follow-up has been investigated from various angles: Felker et al. [
23] reported that the addition of serial multiparametric MRI on AS follow-up significantly improved the detection of disease progression. Chang et al. [
24] found that tracking biopsy (both systematic and MRI-targeted) could detect upgrading more often than nontracking biopsy. More evidence, including our result of high per-lesion PPV, is needed through larger prospective studies to ensure that MRI-visible lesions with a high PI-RADSv2 scores on AS follow-up justify a change from AS to active treatment without biopsy confirmation.
This model of looking at per-lesion detection can add incremental information to plan focal therapy. Focal therapy of PCA can be used for three main purposes: to cure low-risk localized PCA that can be completely eradicated, to control disease and prolong the surveillance period during AS by eliminating the index lesion, or to be used as part of multimodal treatment of a high-risk patient who cannot undergo radical prostatectomy because of other medical problems [
25]. In our study, only one of 29 clinically insignificant PCA lesions was MRI-visible; therefore, using multiparametric MRI for exploring low-risk (low-volume, GS = 6) PCA that can be curable through focal therapy is perhaps not appropriate. However, considering the high PPV of our per-lesion study result, multiparametric MRI can be a useful tool for assessing the target of focal therapy and deciding the ablation area by providing additional information about location, size, and clinical significance of targeted PCA when planning focal therapy for disease control during AS or as an adjunct or adjuvant option in the treatment of high-risk PCA.
We also showed that MRI-visible clinically significant PCAs had a larger mean volume than MRI-invisible clinically significant PCAs (3.19 and 1.11 cm
3, respectively) and, interestingly, a higher secondary Gleason grade. It has been reported that tumor size and a GS of 7 or more were associated with identification of the PCA lesion with both T2-weighted imaging and DWI [
26]. The associations of GS with ADC values have also been reported [
27]. Thus, the difference of tumor volume between MRI-invisible and MRI-visible significant PCAs is well supported by previous studies. However, the result that only secondary Gleason pattern rather than total GS or primary Gleason pattern was different between MRI-invisible and MRI-visible significant PCA is unique to our study. The primary Gleason grade was rather uniform (3 or 4, lack of pattern 5), whereas the secondary Gleason grade was more variable (3–5) in our subjects. This wider variability of secondary Gleason grades could affect the difference of MRI visibility among significant PCA lesions. A previous study also reported that a solid tumor growth focus, which is abundant in Gleason pattern 4 or 5, was significantly associated with visibility on DWI [
26]. Our MRI-visible PCAs were more likely located in the PZ (
n = 20) than the TZ (
n = 11); thus, more PCA lesions were scored with DWI or ADC, the dominant sequence of PZ cancer. Therefore, the presence of a solid tumor growth focus could have an influence on overall MRI visibility and subsequently could explain the higher mean values of the secondary Gleason pattern in MRI-visible significant PCA. Further studies for evaluating the causal relationship of MRI visibility of significant PCA and secondary Gleason grade are needed in larger prospective settings.
Our study has several limitations. First, the number of overall PCA lesions was relatively small. Second, this study was a single-institution retrospective study. Further multicenter prospective studies are needed to validate our results.
Regarding the b value on DWI, we used b values of 0 and 1000 s/mm
2 and did not adhere to the PI-RADSv2 recommendation of a b value of at least 1400 s/mm
2, because we obtained the best DWI quality from our machine at a b value of 1000 s/mm
2. Agarwal et al. [
28] reported that the optimal b value for acquired DWI in differentiating intermediate- and high-risk PCAs from low-risk PCAs in the PZ is 1600 s/mm
2, but they also reported that there was no significant different diagnostic performance of DWI with b values ranging from 1067 to 2000 s/mm
2. Thus, using a b value of 1000 s/mm
2 could be an acceptably high b value because the goal of our study is measuring detection rather than suspicion level. However, not using a higher b value could have partially affected the detection rate in our study.
Furthermore, we included only patients undergoing radical prostatectomy, which could have caused a selection bias. In other words, this population not only does not represent the AS population, but also may be different from the population for whom AS fails. This could result in several problems: First, patients with low-risk tumors who were on AS were not included in our study, possibly resulting in an underestimation of the proportion of clinically insignificant PCAs. Per-lesion analyses like ours are less affected by this problem because presurgical insignificant lesions are automatically included; despite this, the problem of underestimation is still not fully overcome. Second, false-positive lesions (MRI-visible non-PCA lesions) were not included in our study, because the denominator of our analyses was pathologically proven PCA with or without clinical significance. This would influence the PPV, possibly making the PPV greater than expected. However, because the overall PPV using PI-RADSv2 was sufficiently high in previous studies [
15,
16], the possible high PPV in our study could not have affected our results. Last, the likelihood of clinically significant PCA regarding MRI visibility was not evaluated because the actual probability of clinically significant PCA was not accurately represented by using only surgical cases.
In conclusion, multiparametric MRI with PI-RADSv2 missed a considerable number of clinically significant PCA lesions in this per-lesion analysis. We found a relatively low NPV and diagnostic performance compared with the per-patient results reported in previous studies. However, because of its high PPV, multiparametric MRI with PI-RADSv2 could be useful for follow-up of AS patients. Secondary Gleason grade as well as tumor volume can influence the MRI visibility of clinically significant PCAs.