Biopsy-Integrated 3D Magnetic Resonance Imaging Modeling of Prostate Cancer and Its Application for Gleason Grade and Tumor Laterality Assessment

This retrospective study was approved by the Institutional Review Board of Asan Medical Center (Seoul, Republic of Korea) (2020-1145). We searched for consecutive patients with a suspicious prostate lesion at mpMRI from May 2019 to May 2020 who subsequently underwent MRI-TRUS–targeted biopsy with a 12-core systematic biopsy followed by RP. Among the 87 patients initially included, 4 patients who underwent neoadjuvant androgen deprivation therapy were excluded. Hence, 83 patients were included in this study.

The mpMRI was acquired by using a 3.0T MRI system (Skyra, Siemens Healthcare, Erlangen, Germany and Ingenia, Philips Healthcare, Best, the Netherlands) with pelvic phased-array coils. The details for image acquisition parameters are as previously described.⁴  The degree of suspicion was determined by Prostate Imaging Reporting and Data System (PI-RADS) version 2.1 scoring system. PI-RADS categories 3 to 5 were regarded as mpMRI-suspicious lesions.⁵  The biopsy-proven mpMRI-suspicious lesion was designated as MRI-visible PCa, whereas the biopsy-proven tumor at the nonsuspicious site, as MRI-nonvisible PCa. The longest diameter of suspicious lesion was measured from 3 sequences of T2-weighted image, diffusion-weighted image, and dynamic contrast-enhanced image. The index tumor on mpMRI was defined as that with the highest PI-RADS score. In multiple lesions with the same PI-RADS score, the largest suspicious lesion with the highest volume was considered the index tumor. The tumor volume (mL) was calculated as follows: anteroposterior dimension (height in mm) × the transverse dimension (width in mm) on the axial scan showing the maximum tumor area × the craniocaudal dimension (length; multiplying [number of MRI slice containing tumor + 1] × thickness or MRI slice [3 mm]) × the coefficient for sphere volume (π/6 ≈ 0.524). Noncontiguous tumors on axial, sagittal, and coronal T2-weighted images, dynamic contrast-enhanced (DCE) images, and/or diffusion-weighted imaging (DWI) were regarded as separate lesions. For radiologic evaluation of extraprostatic extension, MRI-extraprostatic extension grade was used as follows: grade 0 for no suspicious finding; grade 1 for either curvilinear contact length of 1.5 cm or capsular bulge and irregularity; grade 2 for both grade 1 findings; and grade 3 for frank capsular breach or adjacent structure invasion.⁶ 

The targeted biopsy was performed by using a fusion MRI-TRUS biopsy platform (Artemis; Eigen, Grass Valley, California) and harvested 2 to 3 cores per a single suspicious lesion by a single genitourinary (GU) radiologist, followed by a 12-core systematic biopsy. During biopsy procedures, the biopsy paths of targeted and systematic cores were recorded in the biopsy platform, allowing retrospective analysis of all biopsy paths on a 3D basis.

The prostate core(s) of each biopsy site was delivered in a separate container and diagnosed separately. The handling of RP specimens, including the total embedding of the prostate and tumor mapping, was performed as previously described.⁷  The prostate needle biopsy and RP specimens were reassessed by 2 GU pathologists according to the 2019 International Society of Urological Pathology (ISUP) consensus guideline.¹  The clinically significant PCa (csPCa) was defined as an organ-confined GG2 tumor with a volume of at least 0.5 mL or any tumor with a GG of 3 or higher, and the index tumor was defined as the largest tumor in the RP specimen.⁸ 

A GU radiologist reviewed mpMRI findings of all 83 cases and provided their topographic information of the index tumor and satellite tumors and biopsy paths to a GU pathologist. From the topographic relationships between the index tumor and all biopsy paths on a 3D basis—which had been automatically generated in the Artemis MRI-TRUS biopsy platform during biopsy—the radiologist marked the biopsy cores, which had been harvested from the mpMRI-suspicious index tumors (ie, S1–6 and T1–2; Figure 1, A). Then, the information was provided to the GU pathologists.

When necessary, pathologists could access the Artemis MRI-TRUS biopsy platform and review the spatial relationship between the biopsy path and mass in 3D. To integrate mpMRI and biopsy findings of the index lesion, the GU pathologist identified the targeted (ie, T1–2) and systematic cores (ie, S1–6) obtained from the index tumor, based on the 3D information, and divided the PCa-positive cores into those obtained within (ie, S1–6 and T1–2) and outside (ie, S7) the index tumor (Figure 1, B). PCa-positive cores obtained from satellite tumors were also incorporated into the 3D mpMRI model in addition to those from the index tumor (Figure 1, A). Next, the mpMRI-integrated GG of the index tumor was generated by comprehensively summarizing tumor grades of all positive cores of targeted and systematic biopsies within the index lesion. When there were multiple mpMRI-suspicious lesions with biopsy detection, mpMRI-integrated GGs were given to each lesion, and the highest mpMRI-integrated GG on a patient basis was used for comparison. The integration of GG was also possible even with fragmented biopsy cores because each biopsy core was submitted to the pathology department in a separate container, and the cores were inked in different colors, enabling us to distinguish them from each other. The image of the biopsy-integrated 3D mpMRI model was produced by a semiautomatic segmentation of the prostate, neurovascular bundle, seminal vesicle, and mpMRI-defined PCa with an indication of the cancer-positive paths (Figure 1, A). The highest GG was recorded for other biopsy interpretation methods including systematic biopsy, targeted biopsy, and a combination of both in each case. The mpMRI-integrated GG and highest GGs of the combined biopsy (targeted and systematic biopsy), targeted biopsy, and systematic biopsy specimens were compared with the GG of the RP specimen (Figure 1, C through F). In addition, the assessment of tumor laterality (ie, unilateral or bilateral) according to the biopsy-integrated 3D mpMRI model and the presence or absence of undetected csPCa in the model were made through comparison with the RP specimen (Figure 1, C through F). Upon slide review of the biopsy cores for this study, 2 tumor foci separated by benign intervening stroma of 1 mm or less were regarded as a single focus, whereas those separated by more than 1 mm were considered as separate foci. The 3D mpMRI model also included the longest lengths of the index and satellite lesions where the tumor length in each prostate core was measured in 2 methods for comparison: collapsed (continuous) and discontinuous fashions. In the collapsed method, which had been used in routine practice in our institution, the tumor length was measured as the sum of the length of separate tumor foci, excluding the benign intervening stroma.^9–11  Upon slide review for this study, we also quantified the tumor length in a discontinuous fashion, measuring the span of tumor foci from end to end, including the benign intervening stroma.^9–11 

Continuous variables were compared by using the Mann-Whitney U test and the categorical variables, by using the χ² test or Fisher exact test. Differences in the concordance rate or any upgrading or downgrading rate between biopsy interpretation methods were compared by using the McNemar test with 95% CIs, calculated by using the Wald interval. The concordance of the GG estimation by each biopsy interpretation method compared with the RP specimen was represented as the weighted Cohen κ with 95% CIs. Bland-Altman plots were used with 95% CIs as the limit of the agreement to assess the agreement of tumor size between the largest tumor diameter of RP specimens and mpMRI- or biopsy-defined tumor lengths.¹²  Spearman correlation was used for correlation analysis. The receiver operating characteristic (ROC) curves and area under the ROC curve (AUC) were used to evaluate the predictive performances of extraprostatic extension. All statistical analyses were performed with Statistical Package for the Social Sciences (version 21; IBM Corp, Armonk, New York), and a 2-tailed P value of less than .05 was considered statistically significant.

Most cases had 1 suspicious lesion on mpMRI (76 of 83 cases, 91.6%) and PI-RADS categories 4 and 5 (63 of 83 cases, 75.9%) (Table 1). The mpMRI-suspicious lesions were biopsy-proven in 76 of 83 cases (91.6%) by targeted biopsy only (5 of 76 cases, 6.6%), systematic biopsy only (9 of 76 cases, 11.8%), and both types of biopsies (62 of 76 cases, 81.6%), with a median of 2 systematic biopsies harvesting the index tumor (range, 0–11). In the remaining 7 of the 83 cases (8.4%), the mpMRI-suspicious lesion was nonneoplastic, and PCa was diagnosed at the nonsuspicious site by using systematic biopsy. MRI-extraprostatic extension grades 0, 1, 2, and 3 were observed in 40 (48.2%), 24 (28.9%), 7 (8.4%), and 12 (14.5%) cases, respectively, of the 83 cases.

By integrating mpMRI findings and pathology results of the prostate biopsy, a biopsy-integrated 3D mpMRI model was developed (Figure 1). It demonstrated both MRI-visible and MRI-nonvisible PCa. For mpMRI-visible PCa, the longest tumor length among biopsy cores and the MRI-measured tumor size was illustrated. For mpMRI-nonvisible PCa, the longest tumor length among biopsy cores was illustrated. It explained the spatial relation of the biopsy-proven PCa to adjacent structures, such as neurovascular bundle and seminal vesicles. Among the 76 cases with MRI-visible PCa, the mpMRI-integrated GG2 was the most common (36 of 76 cases, 47.4%).

In RP specimens, most tumors were multifocal (68 of 83 cases, 81.9%), confined to the prostate (50 of 83 cases, 60.2%), and GG2 (38 of 83 cases, 45.8%). Most of the index tumors, defined as the largest tumor in the RP specimens, were concordant with mpMRI (72 of 76 cases, 94.7%). Notably, none of the smaller csPCa undetected by the biopsy-integrated mpMRI model showed a higher GG than the index tumor in RP specimens (data not shown in tables).

The GG using mpMRI-integrated biopsy revealed the highest concordance rate (57.9%, 44 of 76) and agreement (weighted κ, 0.545; 95% CI, 0.415–0.675) with the GG of RP specimen. The concordance rate was significantly higher than those of targeted (47.8%, 32 of 67; P = .008) and systematic (39.4%, 28 of 71; P = .01) biopsies (Table 2). Although no statistical difference in the concordance rate was observed between mpMRI-integrated GG and the highest GG of 2 types of biopsies (combined biopsy) (57.9% [44 of 76] versus 50.0% [38 of 76]; P = .16), the mpMRI-integrated biopsy demonstrated significantly lower downgrading rates (14.5% [11 of 76 cases] versus 28.9% [22 of 76 cases]; P = .001) without significant differences in upgrading rates (27.6% [21 of 76 cases] versus 21.1% [16 of 76 cases]; P = .06). Moreover, mpMRI-integrated biopsy showed a significantly lower upgrading rate than for targeted biopsy (27.6% [21 of 76 cases] versus 37.3% [25 of 67 cases]; P = .03) and a significantly lower downgrading rate than for systematic biopsy (14.5% [11 of 76 cases] versus 26.8% [19 of 71 cases], P = .01).

The upgrading rate of mpMRI-integrated biopsy to GG4 and more (9.2%, 7 of 76 cases) in RP specimens was unremarkable compared with the others (5.3%–11.9%); however, it showed the lowest downgrading rate (7.9%, 6 of 76 cases) to GG3 or less (9.0%–13.2%). The cross-tabulations of GG according to biopsy interpretation methods (mpMRI-integrated, combined, targeted, and systematic) compared with the RP are illustrated in Supplemental Figure 1, A through D (see supplemental digital content at https://meridian.allenpress.com/aplm in the February 2023 table of contents).

According to the 3D mpMRI model, 32 of the 83 cases (38.6%) were unilateral diseases, and 51 of the 83 cases (61.4%) were bilateral diseases when both MRI-visible and MRI-nonvisible PCa were counted (Table 3). Among the 32 unilateral diseases, 27 (84.4%) were bilateral in RP specimens, and most (70.3%, 19 of 27 cases) were non-csPCa. Therefore, 8 of the total 83 cases (9.6%) had undetected csPCa in the contralateral side of the index tumor according to the 3D mpMRI model (Table 3).

Bilateral disease on the 3D mpMRI model was associated with higher preoperative serum levels of prostate-specific antigen (P = .009), higher diameters of the index tumor on mpMRI (P = .02), lower PI-RADS v2.1 categories (P < .001), higher MRI-extraprostatic extension grades (P = .04), higher volumes of the index tumor in RP (P = .02), and higher pT stages than unilateral disease (Supplemental Table 1).

The largest tumor diameter at RP was compared with the largest index tumor diameter on mpMRI and the highest positive core lengths in biopsies. The largest tumor diameter measured on mpMRI best correlated with that at RP (r = 0.592, P < .001) as compared to any biopsy-measured maximum tumor lengths (Supplemental Table 2). Among the various biopsy interpretation methods, the maximum tumor length of the combined biopsy measured by the discontinuous method best correlated with that at RP (r = 0.590, P < .001). The Bland-Altman plot revealed +3.1 mm (95% CI, −8.8 to +15.0 mm) of the mean difference between RP and mpMRI-defined tumor (Figure 2, A) and +13.8 mm (95% CI, +1.3 to +26.4 mm) between RP and the highest tumor length of combined biopsy measured by the collapsed method (Figure 2, B) and +13.0 mm (95% CI, +1.5 to +24.5 mm) measured by the discontinuous method (Figure 2, C).

ROC curves were used to evaluate the predictive performances of extraprostatic extension in RP where 4 biopsy GG interpretation methods, 2 maximum tumor length interpretation methods, and MRI-extraprostatic extension grade were examined (Supplemental Figure 2, A through E). The best performance for predicting extraprostatic extension was observed in the MRI-extraprostatic extension grade (AUC = 0.814) (Supplemental Figure 2, E). Among the biopsy GG interpretation methods, GGs by mpMRI-integrated biopsy showed the highest predictive performance (AUC = 0.671) (Supplemental Figure 2, A). Among the biopsy maximum tumor length interpretation methods, the maximum tumor length of the combined biopsy in the discontinuous method had the best predictive performance (AUC = 0.745) (Supplemental Table 3).

With remarkable advances in therapeutic technology and early detection of low-volume PCa, focal therapy is receiving growing attention as an initial treatment option for localized PCa that is organ sparing and a minimally invasive procedure.¹³  Although patient selection criteria have not been established, unilateral PCa with clinical stage T2 or less, serum prostate-specific antigen level of 20 ng/mL or less, and GS 7 (4 + 3) (GG3) or less have been suggested as eligibility criteria for focal therapy.^14,15  In this regard, integrated biopsy reporting would help provide a more accurate assessment of GG and the extent of PCa.

In response to clinical benefit from MRI-directed targeted biopsy from high-quality evidence and technical developments of the MRI-TRUS fusion biopsy platform, an MRI-directed biopsy has been incorporated into the clinical care guidelines.^16,17  Subsequently, optimal methods for GG reporting and for integrating multiple biopsy cores from MRI-defined single lesion have recently been discussed among pathologists.^1–3,18,19  Gordetsky et al¹⁸  compared the global (aggregate) GS versus the individual GS for concordance with the RP GS, but they could not reach definite conclusions owing to the limited number of cases. Deng et al³  revealed that better agreement with the RP GS was shown by the global (aggregate) GS than by the individual GS for either the highest or largest GS. In the 2019 ISUP meeting, consensus based on 78% agreement of the pathologists was reached to report a global GS for each MRI-defined lesion.¹  Accordingly, the most recent guidelines from the ISUP and Genitourinary Pathology Society (GUPS) recommend reporting a single global (aggregate) GS for a single mpMRI-visible lesion with multiple cancer-positive targeted cores.^1,2 

However, how to assign GS when a single mpMRI-visible PCa is detected in multiple positive cores of systematic and targeted biopsies has remained an unsolved issue. Having multiple positive cores in systematic and targeted biopsies with dissimilarities in GG is common even though they are harvested from a single lesion. Given the risk of missing csPCa in a nonnegligible minority case (4.2%–12.5%), it is advised to perform a systematic biopsy in conjunction with targeted biopsy where at least 2 cores are obtained from each MRI-suspicious lesion.^{17,18,20–26}  Moreover, a median of 2 systematic cores were overlapped with the MRI-visible lesion, as shown in this study.²⁷  In cases with multiple positive cores and dissimilar GS, most clinicians prefer to use the highest GSs and number of positive cores to estimate the tumor grade and extent, respectively, and do not distinguish between systematic and targeted cores.¹⁹  Although using the highest GS allowed a better estimation of the tumor grade with a low rate of GS upgrading in RP,¹⁷  it is likely to overestimate GS with a high rate of downgrading in RP, which might lead to inappropriate exclusion for focal therapy. Moreover, in cases with multiple positive cores and dissimilar GS, some other clinicians prefer to use global GS and GS in the most involved core.¹⁹  Although the 2005 ISUP consensus left case-level GS optional, allowing clinicians to interpret GS as per their practice for patient management, such a nonuniform approach by clinicians creates a need to adopt standardized methods of pathology reporting and patient management in the era of MRI-TRUS fusion biopsy.^1,2,28  Accordingly, there is active discussion among pathologists on case-level reporting of prostate biopsy.^1,2,19  In the 2019 ISUP meeting, 41% of pathologists agreed to provide a global (aggregate) GS when the systematic and targeted GS are not equal, although they failed to reach an agreement.¹  In a 2019 GUPS white paper, providing a global score factoring both systematic and targeted biopsy cores was recommended, but the authors could not determine the method of global scoring.²  However, the MRI-guided spatial tracking of biopsy cores can predict the distribution of MRI-visible and MRI-nonvisible lesions, so we were able to check which systematic biopsy cores overlap with the targeted biopsy, leading to the generation of mpMRI-integrated GG. Our novel integrative grading system (mpMRI-integrated GS), encompassing not only targeted but also systematic biopsies of a single mpMRI-visible lesion, would be the most appropriate method for determination of a case-level global GS. Furthermore, our 3D mpMRI model would differentiate multifocal tumors and enable estimating percentages of patterns 4 and 5 within each tumor in a biopsy specimen, as anticipated in the 2019 ISUP meeting.¹  Our research also merits attention not only for urologists or radiologists but also for pathologists, as the role of pathologists to incorporate information from biopsy procedures and biopsy cores is important for a multidisciplinary approach in the context of providing the best patient care. Given that clinicians have limited knowledge of pathologic examinations and heavily rely on pathology reports for treatment decisions,¹⁹  pathologists need to take the initiative to provide mpMRI-integrated GG as a case-level GS to guide clinicians' treatment decisions.

In attempts to find the best biopsy interpretation method concordant with the final GG in the RP specimen, the GG of the MRI-TRUS fusion biopsy had been previously compared between systematic biopsy, targeted biopsy, and a combination of both with RP specimens.^17,29  The highest GG had been used when multiple positive cores include different GG.^17,29  They showed that combined biopsy predicted the final GG better, with the highest concordance.^17,29  In this study, the concordance rate of mpMRI-integrated biopsy (57.9%, 44 of 76 cases) was higher than that of the combined biopsy (50.0%, 38 of 76 cases). Furthermore, the mpMRI-integrated GG revealed the lowest downgrading rate without significant upgrading, suggesting that it is the most suitable grading system for the therapeutic planning of PCa.

Laterality assessment of PCa is important because hemiablation might be the most feasible strategy of focal therapy. However, the standard extended TRUS biopsy appears to be limited for laterality assessment. Most unilateral diseases on the extended biopsy were proven to be bilateral, whereas ∼20% remained unilateral in RP.^30,31  A transperineal mapping biopsy might be appropriate in selecting patients for focal therapy. However, it is relatively invasive and associated with morbidity due to its high sampling density. As an alternative, the combination of MRI with transrectal biopsy has been suggested to overcome the limitation of the TRUS biopsy. Similar to other studies, among the 32 cases with unilateral diseases in the 3D mpMRI model, 27 (84.4%) were bilateral diseases in RP specimens; however, most undetected tumors (70.4%, 19 of 27 cases) had non-csPCa on the contralateral side. In the entire cohort, a minor proportion (9.6%, 8 of 83 cases) missed contralateral csPCa according to the 3D mpMRI model.

Accurate estimation of histopathologic tumor volume is important for focal therapy planning.³²  It has been reported that MRI underestimates histopathologic tumor volume, as shown in this study.^32–34  From this point of view, a 9-mm treatment margin around an MRI-visible tumor was suggested to ensure complete tumor control during focal therapy.³⁴  However, instead of applying a fixed safety margin to all patients, individualized estimation of the tumor extent may optimize treatment outcomes and reduce undesirable complication rates. Therefore, the integration of MRI and biopsy core lengths may predict the histopathologic tumor volume better because the biopsy core length, especially measured with the discontinuous method, allows a more accurate estimation of the anteroposterior diameter of the index tumor.³⁵  It is expected that the 3D MRI model estimates the tumor boundary better, incorporating information from MRI and biopsy specimens. Further investigation incorporating information of 3D MRI and biopsy specimens will be helpful to estimate histologic tumor extent.

The biopsy tumor lengths measured by using the discontinuous method were better correlated with the tumor length in RP, regardless of the biopsy interpretation methods (ie, targeted, systematic, or combined), as shown by Schultz et al¹¹  and Deng et al.³  Additionally, the extraprostatic extension was best predicted by the maximum tumor length of the combined biopsy in the discontinuous method similar to a report by Karram et al.¹⁰  Therefore, the best biopsy interpretation method incorporating information from targeted and systematic biopsies warrants further discussion.

The present study has several limitations. The retrospective design performed in a single center might result in patient selection bias. To our knowledge, our study is the first to incorporate information from biopsy cores obtained by systematic and targeted biopsy and from MRI-directed targeting procedures, and this novel method has not been validated yet. Given that several MRI-TRUS biopsy platforms provide a 3D model containing MRI-driven tumor segmentation and biopsy paths from all cores, our biopsy-integrated 3D MRI model can also be generated at different institutions through communications between practitioners conducting MRI-directed biopsy and pathologists. In this regard, the biopsy-integrated 3D model requires validation and may need refinement to implement in clinical practice. Furthermore, the usefulness of the biopsy-integrated 3D mpMRI model on proper therapeutic planning and improved patient outcomes should be confirmed in subsequent prospective multicenter studies with a long follow-up period. The image of the biopsy-integrated 3D mpMRI model was constructed by semiautomatic segmentation of the prostate and adjacent structures by an experienced GU radiologist. Automated segmentation of the mpMRI findings, using an artificial intelligence algorithm, might be essential for the biopsy-integrated 3D mpMRI model to be easily and widely used for PCa management.

The biopsy-integrated 3D mpMRI model was developed through the collaboration of uroradiologists and uropathologists, on the basis of automatically recorded biopsy tracking information from biopsy procedures. The model allowed us to easily recognize the location and extent of the MRI-visible and MRI-nonvisible PCa within the prostate and their spatial relations to adjacent structures. It can also provide an integrated GG of the MRI-visible PCa, which was highly concordant with the final GG in the RP. Therefore, the biopsy-integrated 3D MRI model might be useful for risk stratification and proper therapeutic planning of PCa, but this model still requires validation.