Diagnostic accuracy of multiparametric magnetic resonance imaging in detecting extracapsular extension in intermediate and high - risk prostate cancer

ABSTRACT Objectives: To evaluate the diagnostic performance of preoperative multiparametric magnetic resonance imaging (mp-MRI) as a predictor of extracapsular extension (ECE) and unfavorable Gleason score (GS) in patients with intermediate and high-risk prostate cancer (PCa). Materials and Methods: Patients with clinically localized PCa who underwent radical prostatectomy (RP) and had preoperative mp-MRI between May-2011 and December-2013. Mp-MRI was evaluated according to the European Society of Urogenital Radiology MRI prostate guidelines by two different readers. Histopathological RP results were the standard reference. Results: 79 patients were included; mean age was 61 and median preoperative prostate-specific antigen (PSA) 7.0. On MRI, 28% patients had ECE evidenced in the mp-MRI, 5% seminal vesicle invasion (SVI) and 4% lymph node involvement (LNI). At RP, 39.2% had ECE, 26.6% SVI and 12.8% LNI. Sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) of mp-MRI for ECE were 54.9%, 90.9%, 76%, 81% and 74.1% respectively; for SVI values were 19.1%, 100%, 77.3%, 100% and 76.1% respectively and for LNI 20%, 98.4%, 86.7%, 66.7% and 88.7%. Conclusions: Major surgical decisions are made with digital rectal exam (DRE) and ultrasound studies before the use of Mp-MRI. This imaging study contributes to rule out gross extraprostatic extension (ECE, SVI, LNI) without competing with pathological studies. The specificity and NPV are reasonable to decide surgical approach. A highly experienced radiology team is needed to provide accurate estimations of tumor extension and aggressiveness.

transrectal ultrasound-guided (TRUS) prostate biopsy (1). Recently, additional parameters such as MRI and the percentage of positive biopsy cores have been added into risk classification methods, both with promising results (2).
Traditionally, RP and radiotherapy have been considered the reference standard treatment for patients with localized PCa (3) both achieving long-term disease control. High rates of erectile dysfunction and incontinence are related to surgical techniques and difficult preservation of the neurovascular bundles. Therefore, accurate preoperative knowledge of tumor stage and possible ECE is crucial in achieving the best surgical, oncological and functional results (4).
Recent findings support the use of mp-MRI of the prostate, combining T2-weighted imaging with diffusion-weighted imaging (DWI) and perfusion imaging, as the most sensitive and specific imaging tool for different clinical scenarios in patients with PCa such as detection, staging, and follow-up (5)(6)(7)(8)(9)(10)(11)(12)(13). Nonetheless, its routine use is still a topic of debate given the high variability among studies regarding the diagnostic accuracy of mp-MRI in staging and prediction of ECE (14).
The aim of this study was to evaluate diagnostic accuracy of a 1.5 tesla mp-MRI in detecting ECE, SVI, LNI and unfavorable GS in patients with intermediate and high-risk PCa.

MATERIALS AND METHODS
After approval from our hospital review board and ethics committee, clinical records were reviewed retrospectively. Patients with clinically localized PCa, who underwent RP and extended lymph node dissection between May 1st 2011 and December 31st 2013 at our institution, and had preoperative mp-MRI, were identified. Inclusion criteria comprised intermediate or high-risk cancer patients as defined by D´Amico classification (intermediate risk: clinical stage T2b or PSA levels between 10.1 and 20ng/mL or GS 7; high risk: clinical stage ≥T2c or PSA levels >20ng/mL or GS 8-10). Finally, the diagnostic accuracy of mp-MRI in detecting ECE, SVI, LNI and unfavorable GS (equal or greater than 8) was analyzed.

Multiparametric MRI and Image analysis
Preoperative mp-MRI was performed with a 1.5 tesla Siemens system. Patient should wait 6 weeks after biopsy for MRI study; no rectal preparation nor endorectal coil were used in any patient. Studies included T2-weighted imaging, dynamic contrast-enhanced imaging and diffusion-weighted imaging. Mp-MRI results were assessed and reported according to the European Society of Urogenital Radiology (ESUR) MRI prostate guidelines from 2012 by two different radiologists with different years of experience (14 and 8 years of experience). Readers knew diagnosis and initial PSA level but were blinded to details of histopathological report. They adopted one of the following image signs to subjectively classify ECE: 1) bulging of prostatic contour; 2) irregularity of prostatic contour; 3) neurovascular bundle thickening; 4) loss or discontinuity of prostatic contour line; 5) measurable extra-capsular disease. The sum of three of the previously mentioned signs were considered positive for ECE. Tumor contact length (TCL) to the prostate contour was also measured and a 12mm-threshold was used as an additional sign of ECE and independently analyzed as a different variable. For SVI, the subjective image signs adopted were: 1) expansion; 2) low T2 signal; 3) filling in of angle; 4) enhancement and restricted diffusion. The sum of two signs were considered positive for SVI. For LNI, size criteria were used and a short axis >10mm was considered positive (15). Histopathological evaluation RP specimens were macroscopically marked with ink and fixed in formalin. The specimen was serially sectioned at 3mm thickness. Perpendicular cuts were done for the first portion of the apex and base, the remaining prostate was then cut in the sagittal plane and, posteriorly, in quadrants from apex to base. Slices were further cut into microscopic sections of 3-5μm and stained with Haematoxylin and Eosin. ECE was defined as tumor cell growth into the extraprostatic tissue and subclassified as: 1) focal ECE when involving only a few glands or a tumor involving less than one high power (40X) field in one or two sections or 2) stablished ECE when a more extensive spread was seen beyond the prostatic edge (16). SVI was defined as tumor infiltration of the muscular wall of the seminal vesicles. Lymph nodes were processed as a whole with posterior differentiation of nodules from fat. The pathological T-stage (pT) was defined according to the TNM classification.

Statistical analysis
Socio-demographic characteristics were described. A descriptive analysis was conducted and the measures of central trend and variability were reported. Median or mean for continuous variables and absolute or relative frequencies for categorical variables were calculated. Contingency tables (using most experienced radiologist data) were used to calculate overall diagnostic accuracy, sensitivity, specificity, positive predictive value and negative predictive value of mp-MRI in predicting ECE, SVI, LNI and GS greater or equal to 8. Finally, a receiver operating characteristic curve (ROC curve) with an area under the curve (AUC) values were generated to analyze predictive accuracy of ADC in detecting unfavorable GS, and Spearman correlation was used. Inter-reader reliability was calculated using kappa statistics (0-0.2 none, 0.21-0.39 minimal, 0.4-0.59 weak, 0.6-0.79 moderate, 0.8-0.9 strong, above 0.9 almost perfect). A p-value below 0.05 was considered significant. The analysis was performed using STATA 13.1 software.
When focal and established ECE were considered separately, accuracy, sensitivity, specificity, PPV and NPV were 73%, 53.9% (95% CI 42. 6 (15), therefore, it underscores the importance of a correct staging and identification of intermediate and high-risk patient preoperatively, which may lead to improved decision-making regarding treatment and surgical approach.
Operative characteristics of mp-MRI for the detection of ECE, SVI and LNI, differ among published studies and are highly influenced by the use of endorectal coil (er), MRI parameters and clinical interpretation (16). Recent reports are focused on improving er-MRI imaging technology and functional imaging techniques (dynamic contrast enhancement and diffusion weighted imaging) to optimize accuracy in staging suspicious lesions (17). These advances lead to the use of combined mp-MRI with anatomic T2-weighted imaging, with promising results for PCa detection and extension. One meta-analysis (18) determined diagnostic accuracy of mp-MRI for PCa detection using anatomic T2-weighted imaging combined with two functional techniques (DWI and DCE-MRI). Pooled data of seven studies (526 patients) showed sensitivity of 0.74 (95% CI, 0.66-0.81), Our results are consistent with those reported in the literature. For ECE detection, mp-MRI has a reported sensitivity between 35-62.5% and specificity of 89-92% (Table-2). Our study showed sensitivity of 55% and specificity of 91%, both within the range of the literature. Feng et al. (15) sought to evaluate mp-MRI ability to detect focal or established ECE reported by pathology. When considered separately, mp-MRI had a low performance in identifying focal ECE, with a sensitivity of 14.3% and a PPV of 5.6%. However, mp-MRI was accurate in predicting established ECE, with a sensitivity of 73% and a PPV of 57%. Our study reported a sensitivity of 53% and a PPV of 48% for established ECE, and for focal ECE performance was poor. Mp-MRI is unable to identify and localize focal ECE, which accounts for a large majority of the cases (18). Nerve-sparing RP requires of mp-MRI to provide a high NPV in detecting ECE. Our results for SVI and LNI are also comparable to literature (Table-2). SVI detection has been reported with PPVs of 62% to 95% and NPVs of 73% to 83%. Sensitivity and specificity for LNI is between 14-33% and 91-97%, respectively (19).
Measurement of diagnostic agreement between radiologists in our study for ECE, SVI and LNI demonstrated minimal and weak. It has been demonstrated in literature that inter-reader reproducibility tend to be higher for relatively experienced readers. This can be explained due to study complexity, different reading scales used such as the PI-RADS of Likert scale and interpretation sensibility may be affected by lesion location, volume and tumor aggressiveness (20,21).
Furthermore, TCL parameter has also shown correlation with pathologically confirmed microscopic ECE. Baco et al. (22) documented accuracy of 82%, 79% sensitivity, 85% specificity, 76% PPV and 88% NPV when using a 20mm TCL threshold. Our study evaluated the differences in predictive values by using a 12mm threshold when comparing established and focal microscopic ECE, both with acceptable NPVs when evaluating these parameters. Other studies have concluded that a higher length of contact between tumor margin and prostatic capsule is associated with an increased risk of ECE.
In general, literature concludes that mp-MRI has a fair performance capacity for predicting unfavorable GS. Hegde et al. (12) conducted a retrospective study with 118 patients, demonstrating that ECE or SVI findings on mp-MRI are associated with a higher risk of identifying previously undetected GS 8-10 disease. In our study, mp-MRI evidenced low power either to confirm or rule out unfavorable GS. Other studies have evidenced that a high percentage of patients have GS ≥7 at non-index tumors and mp-MRI has low power for the detection of these lesions (23). Also, the diffusion coefficient ADC allows for quantitative and qualitative assessment of tumor aggressiveness and has positive correlation with GS (24). In this study, ADC as binary variable when compared to GS >=8 had high specificity and NPV, low AUC and its values are inversely correspondent to GS as reported with Spearman rho, although it shows a low negative correlation. Mp-MRI has shown to be useful in identifying PCa extension and especially in ruling in locoregional extension.
The limitation of this study is its retrospective nature and small sample size leading to large confidence intervals. Also, operative characteristics depend on how the presence of ECE, SVI and LNI are defined. For our study, the presence of locally advanced disease consisted of binary variables (yes or no), and the data would be improved by using ordinal scales such as prostate imaging reporting and a standardized data system (PIRADS) for extension evaluation. According to the magnetic field strength, the 3T magnet increases signal to noise ratio therefore increases resolution. However, both 1.5T and 3.0T can provide accurate and reliable diagnostic exam when technical parameters are properly used (23). Although members of the PIRADS steering committee prefer the use of 3T, 1.5T magnets are more widely available and our study reflects the practice of most of the devices currently installed worldwide. Finally, both radiologists had a 10year experience difference which makes data heterogeneous and consequently the low interrater reliability.

cONcLUSIONS
Major surgical decisions are made with DRE and ultrasound studies before the use of Mp-MRI. This imaging study contributes to rule out gross extraprostatic extension (ECE, SVI, LNI) without competing with pathological studies. The specificity and NPV are reasonable to decide surgical approach. A highly experienced radiology team is needed to provide accurate estimations of tumor extension and aggressiveness. Moreover, it is necessary to carry out prospective and multicenter studies in order to achieve higher consensus regarding MRI use in PCa assessment in developing countries.