The added value of capsular enhancement sign (CES) in the extra-prostatic extension (EPE) grading system in predicting prostate cancer with extracapsular extension
The added value of capsular enhancement sign (CES) in the extra-prostatic extension (EPE) grading system in predicting prostate cancer with extracapsular extension
Federica Martini1, Matilde Mattiauda1, Maria Pigati1, Marta Ponzano2, Enrica Nicosia1, Andrea Cazzato1, Giulia Francese1, Veronica Giasotto3, Giuseppe Cittadini3, Jeries P. Zawaideh3 1. Department of Health Sciences (DISSAL), Radiology Section, University of Genova, 16126 Genova, Italy; 2. Department of Health Sciences (DISSAL), Section of Biostatistics, University of Genova, 16126 Genova, Italy; 3. Department of Radiology, IRCCS Ospedale Policlinico San Martino, 16132 Genova, Italy
The purpose of this study is to confirm the diagnostic accuracy of extra-prostatic extension (EPE) grading system and to validate the diagnostic value of the capsular enhancement sign (CES) with MRI in increasing the predictivity of extracapsular extension in the prostate cancer.
Patient sample and study design: This is a monocentric, retrospective observational study. The population of the study is composed by a series of consecutive patients who underwent radical prostatectomy at IRCSS San Martino Hospital from January 2021 until January 2023 included. All patients underwent an MRI study and were diagnosed with PCa proven by random or MRI-Ultrasound fusion biopsy. The exclusion criteria used to select patients summarized in Figure 1 are: a) patients who did not have MRI images in our institution, b) patients with MRI negative for suspected lesions (with tumor proven on random biopsy), c) presence of magnetic susceptibility artefacts which degrade the quality of MRI images, d) histopathological examination performed by different institutions. MRI acquisition: The MRI protocol was in accordance with the recommendations by PI-RADS Guidelines version 2.1. The imaging was performed using a 1,5 T MRI machine (Magnetom AERA, Siemens). T2-weighted images were obtained in the sagittal plane and successively in the coronal and axial plane, where coronal and axial plane were parallel and perpendicular to the long axis of the prostate, respectively. DWI images were acquired through single-shot echo-planar sequences: firstly, with a sequence with a high b-value (≥1500 s/mm2) and then with another one with three different b-values (0, 750 and 1000 s/mm2) which was used to calculate the apparent diffusion coefficient (ADC) map. Finally, during the intravenous administration of a gadolinium-based contrast agent with a 0,2 mL/kg dose at a flow rate of 3 mL/sec and 15 mL of saline solution, DCE acquisition was accomplished with a temporal resolution of 9 sec. Both DWI and DCE were acquired with the same plane and thickness as the T2-weighted axial images, to better compare the different sequences. MRI analysis: Image analysis was performed independently by three Radiology residents (F.M., M.M. and M.P.) respectively with 3, 3 and 4 years of experience in prostate MRI. The images with doubtful radiologic criteria for EPE were re-analyzed by a fourth radiologist with 7 years of experience in prostate imaging and 2000 MRI reports. Each radiologist had the MRI scan with T2w images on the three planes, the DWI and the DCE sequences and the report of the MRI and the histopathologic result available. For statistical analysis only one target lesion was considered for each patient and when a patient had two or more lesions, it was analyzed the one with the greatest size. For every lesion the localization was differentiated between origin from the peripheric or transition zone and anterior or posterior position. The T2w images were evaluated with the MRI criteria listed in the Grading System for the Assessment of the Risk of Extra-prostatic Extension of PCa at Multiparametric MRI which comprehend the length of capsular contact (LCC), the irregular bulging of the capsule, the asymmetry of the neuro-vascular bundle (NVB) in the setting of an unilateral Pca, the obliteration of the recto-prostatic angle (RPA) and the macroscopic EPE (tumor clearly visible outside the capsule or infiltrating rectum or bladder at MRI images). LCC was not prospectively reported on a consistent basis and, therefore, it was measured retrospectively in the axial plane as curvilinear LCC. In the DCE sequence, it was instead evaluated the presence of the CES. In the ADC map it was calculated the ADC-value positioning a ROI (region of interest) at the level of the target lesion. Reference standard definition Histopathological examination of the whole prostate after prostatectomy was considered the standard reference. Histopathological EPE as well as Gleason score and ISUP grade group were assigned among other parameters. Statistical analysis: Characteristics of the patients were described as mean value and standard deviation for continuous variables and as counts and percentage for categorical variables, overall and separately for patients with and without EPE. Characteristics were compared between the two groups using t-test or Mann-Whitney test for the continuous variables and chi square or Fisher’s exact test for the categorical variables. The ROC curve to assess the ability of LCC to discriminate between patients with and without EPE was plotted and the AUC was calculated. LCC best cut-off was calculated based on Liu, Youden and nearest to (0,1) criteria and patients with anteriorly and posteriorly positioned lesion were compared in terms of LCC, continuously (t-test) and as binary based on the best cut-off (Pearson Chi square test). Sensitivity, specificity and percentage of correctly classified patients were derived for CES, bulging, RPA obliteration, asymmetry of NVB, macroscopic extension and wide LCC (≥14 mm). Univariate logistic regression models were performed to study the association between the previous MRI features and EPE. Two separate multivariable models were fitted including variables with p<0.10 in the univariable models, one including the final score and the other one including the variables that are part of the score. Since CES was found to be a relevant predictor of EPE in both the multivariable models, we created a new score that incorporated both EPE score and CES. The new “enhanced” score was defined based on internal validation: 1) we estimated the penalized coefficients from a bootstrapped lasso model with the EPE score and CES as independent variables 2) we built a model based on the coefficients estimated at 1) and we used that model to estimate the probability of EPE as exp (β0+ β1*CES+ β2*EPE Score)/(1+exp β0+ β1*CES+ β2*EPE Score)). The EPE probability was thus derived for each class of the new score which consists in all the combinations of EPE score values and CES. Finally, AUC for the new “enhanced” score was estimated for comparison with the original EPE score.
Cohort of the study
Clinicopathological characteristics of the study cohort and MRI features among patients with and without EPE
LCC performance and distribution of LCC based on the position of the lesion. (* 4 patients were removed as large lesion invading both PZ and TZ)
Comparisons of patients with and without EPE: A total of 251 consecutive patients were enrolled; of those 152 patients were excluded with 99 patients finally included in our study (Figure 1). The mean age of participants included in the study was 66.6 years, with a standard deviation of 5.6 years. Among the 99 patients included in the study, 31 patients had EPE while the other 68 patients had negative margins for extra-prostatic extension. 86/99 (87%) of the analyzed lesions were located exclusively in the PZ and 10/99 (10%) in the TZ, respectively, with the remaining 3 lesions representing large-volume tumors spanning both zones. 18/99 (18%) were located in the anterior zone, while 77/99 (78%) in the posterior zone. However, according to our results, the position of the tumor was not significantly correlated with extracapsular invasion. Figure 2 ISUP category: We found that with an increasing ISUP grade group, the probability of EPE significantly raised (p=0.009) and EPE is suggested when ISUP grade group is ≥3 (p=0.005). The number and percentage of patients clustered by ISUP grade group can be found on Figure 2. Length of Capsular contact: Patients with EPE had a mean contact of 24 mm vs 12 of those without EPE. Considering EPE-score threshold of 15mm, 57 patients (58%) had extensive capsular contact while the others (42%) had a capsular contact of =14mm. At this cutpoint, sensitivity raises to=74.19%, specificity=67.65%, correctly classified=69.70%. No significant differences were noted between the cut-off in the anteriorly vs posteriorly located lesion. Figure 3 MRI detection performance of EPE: Between the MRI criteria used to determinate the EPE, the two with major sensitivity were shown to be bulging (sen 81%) and a large capsular contact (sen 68%). The most specific ones instead were macroscopic extension (spe 100%), asymmetry of the NVB (spe 97%) and obliteration of RPA (spe 96%). The most correctly classified was asymmetry of the NVB. Figure 4 With univariate analysis our study showed a significant correlation between EPE and respectively, LCC ≥14 mm (p<0.001), ISUP score ≥3 (p=0.005), CES (p<0.001), bulging (p=0.001), asymmetry of the NVB (p<0.001) and EPE score ≥2 (p<0.001). ADC value independently did not significantly correlate with an increasing risk of EPE. With multivariate analysis the results showed a significant correlation between CES (p=0.001), asymmetry of the NVB (p=0.04) and EPE score ≥2 (p=0.004). Figure 5 Even if neuro-vascular bundle asymmetry was associated with EPE, this MRI criteria was found to be rare and it has shown too much variability with a wide confidence interval (3.71; 88.78), thus this criterion was not included in our final score. We estimated the penalized coefficients from a bootstrapped lasso model with the original EPE score and CES as independent variables. Successively, we developed a new “enhanced” score considering EPE score together with CES and we subdivided the score in 8 groups (Figure 6) clustered in 4 grades accordingly to the probability of EPE detection. Subgroups performances are summarized in Figure 7. New score was defined based on the estimated coefficients. When we compared the original EPE score and the new score, the latter performed better (EPE score: AUC=0.76; 95%CI=(0.65; 0.88); new “enhanced” score: AUC=0.83; 95%CI=(0.74; 0.92)). Figure 8
MRI features performance in discriminating between patients with and without EPE.
Univariable and Multivariable Logistic Regression Model for Pathologic EPE Risk Prediction by Using Different Predictors. *Multivariable models include variables showing p<0.10 in the univariable models. Two separate models were performed: 1) including the EPE score and not its components; 2) including components of the EPE score instead of the score.
MRI feature of the new “enhanced” EPE Score. A-B: T2-weighted and DCE of a patient presenting a small lesion in the PZ demonstrating EPE score=0 and absent CES; C-D: T2-weighted and DCE of a male presenting a large lesion in the TZ demonstrating EPE score=1 (CCL=19 mm) and absent CES; E-F: T2-weighted and DCE of a patient presenting a small lesion in the PZ demonstrating EPE score=0 and present CES; G-H: T2-weighted and DCE of a patient presenting a lesion in the PZ demonstrating EPE score=2 (bulging and irregularity of the capsule) and absent CES; I-J: T2-weighted and DCE of a patient presenting a large lesion in the TZ demonstrating EPE score=1 and present CES; K-L: T2-weighted and DCE of a patient showing a lesion in the PZ demonstrating macroscopic EPE without CES; M-N: T2-weighted and DCE of a patient presenting a small lesion in the PZ demonstrating EPE score=2 (CCL ≥ 14 mm and bulging) and present CES; O-P: T2-weighted and DCE showing a giant lesion involving both PZ and TZ with macroscopic EPE and evident CES.
The preoperative evaluation of the presence of EPE in patients affected by PCa is decisive in the surgical management of this condition. In case of suspected EPE the gold standard surgical procedure includes the excision of the neurovascular bundles to decrease the likelihood of microscopic residual disease. MpMRI is the first diagnostic imaging for patients with elevated PSA or suspected PCa at DRE. However, the role of mpMRI is still limited for local staging of PCa because MRI criteria are relatively subjective and require a high degree of expertise by the radiologist. The diagnostic accuracy of the features suggesting EPE is operator-dependent with a high interobserver variability. Indeed, a recent meta-analysis shows that MRI yields a poor sensitivity in detecting EPE, even though the reported specificity is high. Several different MRI characteristics have been correlated to EPE as independent predictor such as LCC, bulging of the capsule, breach of the capsule with macroscopic extra-capsular extension and obliteration of the recto-prostatic angle. We found that LCC has a good accuracy for predicting EPE independently from anterior or posterior location of the lesion, with a sen of 74% (considering the threshold from our court of 14 mm). A recent meta-analysis of 23 studies with 3931 participants, concordant with our results, reports a sen of 79% for LCC, while cut-off values range from 6 mm to 20 mm, demonstrating that a shared and clear consensus on the correct threshold for LCC is still lacking. Matsumoto et al, reported that anterior tumors were less likely to invade extraprostatic tissues. This was also reported by Villers et al, which suggests the cause is due to the morphology of the prostate gland itself and for the presence of vessels penetrating the prostate starting from the neuro-vascular bundle, precisely located posterolateral respect to the prostatic peripheral zone. In a similar way, our study confirms a higher prevalence of tumors in the peripheral area (90%) however this ratio appears completely alike to that of tumors confined to the gland. Considering LCC, our study does not demonstrate a difference in the best dimensional cut-off in the different areas of the prostate, and appears consistent with the data of An et al. Another meta-analysis of 17 studies and 3062 participants, reports LCC and macroscopic EPE as the most accurate predictors for EPE, while bulging and irregular margins of the capsule yield the lowest diagnostic odd ratio. These findings are understandable since these latter signs are qualitative features, therefore more prone to operator-dependent evaluations. In our study, macroscopic EPE showed the highest spe (100%) but the lowest sen (16%) and it has a low prevalence. In our study, we showed that the asymmetry of the NVB is a highly specific sign for detecting pathologic EPE (spe 97%) but it has a low sensitivity (sen 35%). These results are concordant with Yu et al., who described that the asymmetry of NVB yields a specificity of 81-95% and a sensitivity of 21-38%. More recently, also Mehralivand et al. found similar outcomes, reporting a specificity of 99% and a sensitivity of 15% respectively. This feature, however, has appeared to be rare and unreliable (wide confidence interval) so we did not incorporate it in our final score. Moreover, CES resulted highly specific (87%) with a sen of 55%, similar to the results of Caglic et al., which found a spe of 100% and a sen of 16.5%. In their study, CES was also associated with lesions with higher Gleason grade and higher prevalence of lymphovascular invasion; the two appeared to be predictors of biochemical recurrence after surgery. There remains a debate about the value of using DCE sequence in prostate MRI regarding the detection of the tumor. However, it is important to note that some studies have reported benefits that go beyond detection, which also involve reader confidence. Our study similarly reports that MpMRI has an advantage in locoregional staging of the PCa and an additional benefit beyond detection. In our population, we found that ADC value independently did not significantly correspond with an increasing risk of EPE, thus in contrast with other precedent studies. It is important to noticed that Woo et al, reported an additional value of the ADC only when the T2 signal was equivocal and Bengtson et al., did not find any correlation between ADC value and ISUP-gg and tumor aggressiveness. In this light, further evaluations on ADC are suggested. Macroscopic tumor extension was the most specific MRI feature. This is consistent with the findings of Mehralivand et al. and Gatti et al. who reported respectively 96% and 100% of specificity, on the other hand this feature has low prevalence. Nevertheless, different studies highlighted that the combination of all these features can improve the diagnostic accuracy of predicting EPE, thus our final score, combining EPE score with CES, performed better than the single variables and EPE score alone (AUC=0.83; 95%CI=(0.74; 0.92)). Our results confirm the validity of the EPE scoring system, and the probabilities of EPE for the different classes overlap (class I of 26% vs 24.3%; class II 40% vs 38.2% and class III 68% vs 66.1%), we have further divided the classification into 8 subgroups which are then merged based on the probability of EPE into 4 classes. The authors acknowledge the complexity to use it in the clinical practice; on the other hand, its comprehensive nature help to consider the importance to look for all the reported features when investigating the presence or absence of EPE. Limits: Our study has some limitations and the results cannot be interpreted as conclusive. Firstly, the retrospective nature of this single-center study led to a little cohort of patients. However, readers were blinded to the histopathologic data mimicking a prospective reading. In this way, another limit is that all patients included underwent a total prostatectomy, therefore we cannot evaluate the false positives and we just obtain data from aggressive cancers. Particularly, the CES was sometimes noted in case negative for the presence of PCa and it could also represent an inflammatory condition. Moreover, our cohort lacks of transitional zone tumors with EPE. Eventually, every MR images was evaluated by a single reader and only the doubtful cases we re-analyzed. Therefore, an inter-observer agreement was not considered in our study. Conclusion: Preoperative MRI assessment of extraprostatic tumor extension remains a challenge for radiologists. The complexity of the EPE therefore requires a complex but detailed evaluation that combines morphological and contrastographic MRI features. In the era of the debate of mpMRI vs bi-parametric MRI, our classification combining the DCE together with the morphologic features “enhances” the clinical accuracy in reporting the locoregional staging.
Probability of EPE in the eight subgroup and four groups of the new SCORE based on the EPE original score and the CES presence. White: group 0; green: group 1; yellow: group 2; red: group 3.