MRI Features Associated with Histology of Benign Prostatic Hyperplasia Nodules: Generation of a Predictive Model

Abstract

Background:

Histologic phenotypic variation of benign prostatic hyperplasia (BPH) has been hypothesized to underlie response to medical therapy. We evaluate preoperative MRI of robot-assisted simple prostatectomy (RASP) specimens and determine imaging features associated with histologic phenotype.

Materials and Methods:

All patients undergoing RASP from November 2015 to November 2019 with a multiparametric MRI ≤1 year before RASP were included. Patients without identifiable BPH nodules on histologic specimens were excluded. Histology slides were obtained from whole mount adenoma specimens and corresponding MRI were reviewed and graded independently by a blinded expert in BPH histopathology (D.W.S.) and an experienced radiologist specializing in prostate imaging (D.N.C.), respectively. Each nodule was assigned a phenotypic score on a 5-point Likert scale (1 = predominantly glandular; 5 = predominantly stromal) by each reviewer. Scores were compared using the sign test and univariate analysis. Signal intensity relative to background transition zone and nodule texture were noted on T2, diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) imaging sequences. Univariate and multivariate stepwise linear regression analysis were conducted to identify MRI features associated with histology score. All analyses were performed using Statistical Analysis System (version 9.4).

Results:

A total of 99 prostate nodules in 29 patients were included. Median phenotypic scores by histology and MRI were comparable (2, interquartile range [IQR] 2–3 vs 2, IQR 2–4, respectively; p = 0.63). Histology scores were positively correlated with MRI scores (Pearson's correlation 0.84, p < 0.0001). Multivariate stepwise linear regression analysis showed that low apparent diffusion coefficient (ADC) signal intensity (p < 0.001) and DCE wash-in (p = 0.03) were positively associated with more stromal histology, whereas ADC standard deviation (p = 0.03), DCE wash-out (p = 0.001), and heterogeneous T2 texture (p = 0.003) were associated with more glandular histology.

Conclusion:

There is a strong correlation between MRI features and the histologic phenotype of BPH nodules. MRI may provide a noninvasive method to determine underlying BPH nodule histology.

Introduction

Benign prostatic hyperplasia (BPH) is a pathologic process involving proliferation of both stromal and epithelial elements of the prostate over time.¹ This results in progressive lower urinary tract symptoms (LUTS), which demonstrate growing prevalence with age.^2,3 Effective medical management of LUTS is thought to be predicated in part on the histologic phenotype of BPH nodules. Previous study has shown that patient response to alpha-adrenergic antagonists is greater among those with primarily stromal hyperplasia.⁴ Similarly, 5-alpha reductase inhibitors significantly reduce prostate volume and thus LUTS progression by inducing apoptosis of luminal epithelium, suggesting that this may be most effective in BPH patients with more glandular hyperplasia.⁵

Despite this, clinical treatment of BPH with pharmacologic agents typically proceeds without knowledge of the primary underlying histologic phenotype of the prostate. It has been hypothesized that more accurate characterization of the underlying prostate histologic phenotype may facilitate more tailored selection of medical therapy for BPH patients with LUTS, thus allowing for more personalized treatment of the individual patient.⁶

The idea of using MRI to identify BPH histology is not new. Unique MRI features of glandular and fibromuscular stromal elements of the prostate gland have been previously described, with stromal nodules having lower signal intensity and glandular nodules having higher signal intensity on T2-weighted imaging sequences.⁷ Furthermore, MRI has been proposed as a modality to evaluate underlying prostate morphology for BPH patients considering medical therapy for LUTS and represents a noninvasive alternative to transrectal prostate biopsy for tissue evaluation.^7–10 Although neither we nor previous studies advocate that prostate MRI be used to solely to identify BPH histology, we have taken advantage of the many patients presenting with prostate-specific antigen (PSA) elevation who have undergone multiparametric MRI for prostate cancer detection. Because of its increasing use, there is an ever-growing cohort of men in whom cancer is not identified but who are found to only have BPH, often with a substantial gland size driving the PSA elevation. This in combination with the increased use of robot-assisted simple prostatectomy (RASP), results in a substantial population of men with intact whole gland adenoma and a preoperative multiparametric MRI.

In this study, we sought to identify multiparametric prostate MRI features of patients undergoing RASP that define either predominantly glandular or stromal histologic phenotypes, without attempting to correlate to clinical characteristics. We then sought to develop a predictive model of nodular histology based on preoperative imaging features.

Materials and Methods

Patient population

All patients who underwent RASP from November 2015 to November 2019 at our institution were retrospectively reviewed. Preoperative MRI was typically obtained for these patients to further work up an elevated PSA or prior history of low-risk prostate cancer. In some instances, the MRI provided additional information regarding prostate size and configuration that may have influenced the decision to undergo RASP. Those having a multiparametric MRI with no obvious identifiable nodules on histopathologic specimens were excluded. Demographical and clinical data were obtained for the cohort, including preoperative prostate volume, PSA level, international prostate symptom score, maximal urinary flow rate, postvoid residual, and prior BPH medication use history (alpha adrenergic antagonist or 5-alpha reductase inhibitor). Histopathologic slides obtained from whole mount adenoma specimens and corresponding multiparametric MRI images were reviewed for each patient (Fig. 1). This study received Institutional Review Board approval.

FIG. 1.

Whole-mount simple prostatectomy histology specimen from a 70-year-old African American man with BPH and preoperative IPSS of 19, demonstrating multiple nodules on gross specimen (A). Within the same prostate, glandular predominant (nodule #3) and stromal predominant (nodule #4) nodules exist (B). Corresponding multiparametric MRI shows these two nodules on apparent diffusion coefficient map (C), axial T2 (D), and DWI (E). The predominantly glandular nodule #3 is mildly heterogeneous and hyperintense on T2-weighted images and hyperintense on ADC with no wash-in or wash-out on DWI. The predominantly stromal nodule #4 is homogeneous and hypointense on T2, iso-intense on ADC, with wash-in but no wash-out on DWI. ADC = apparent diffusion coefficient; BPH = benign prostatic hyperplasia; DWI = diffusion-weighted imaging; IPSS = international prostate symptom score. Color images are available online.

Histopathologic analysis

After RASP, fresh prostate adenoma specimens were collected from the pathology department. These were fixed in 10% formalin for 1 week and then embedded in paraffin. Standard hematoxylin + eosin staining was performed and whole mount specimens were sectioned. Representative histology specimens were then scanned in high resolution into an electronic tissue biorepository housed at UT Southwestern Medical Center.

Image scans were independently reviewed by an expert in prostate histopathology (D.W.S.) blinded to MRI. Each observed nodule was assigned a phenotypic score on a Likert scale from 1 to 5 (1 = predominantly glandular, 2 = ∼75% glandular, 3 = 50% glandular +50% stromal, 4 = ∼75% stromal, and 5 = predominantly stromal).

MRI and analysis

All multiparametric MRI images were performed using a 3 Tesla MRI scanner (Philips, Best, The Netherlands) using an endorectal and phased-array surface coil. Routine clinical imaging multiparametric MRI protocol at our institution includes multiplanar T2-weighted diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE) imaging sequences. Intravenous (IV) contrast was administered using gadobutrol and pharmacokinetic analysis was performed for perfusion evaluation.

Multiparametric MRI images were retrospectively reviewed for each patient by a fellowship-trained abdominal radiologist with 15 years of experience in prostate imaging (D.N.C.), blinded to histopathologic findings. For each nodule, T2-weighted signal intensity relative to transition zone background as well as texture (homogenous, relatively heterogeneous, and heterogeneous), signal intensity on DWI, signal intensity (mean and standard deviations [SDs]) on apparent diffusion coefficient (ADC) map, and wash-in and wash-out on DCE sequences were recorded. Similar phenotypic scores on a 1 to 5 scale were also assigned for each nodule in a blinded manner to the histopathologic grading.

Statistical analysis

Assigned phenotype scores by histology and multiparametric MRI features were compared using the sign test and univariate analysis. Univariate and multiple stepwise linear regression analysis were then conducted to identify MRI features significantly associated with histology score. Multiple nodules from the same patient were assessed independently, as both glandular-predominant and stromal-predominant nodules can be found within the same prostate. All analyses were performed using Statistical Analysis System statistical software (version 9.4).

Results

Fifty-four patients who underwent RASP during the study period had both a preoperative MRI and available whole-mount surgical pathology. Of these, 25 patients were excluded because of technical issues (e.g., tissue degradation during processing of the whole mount specimen), a lack of obvious nodule on MRI, or an inability to visually correlate specific nodules on MRI to nodules visible on the available whole-mount sections. A total of 99 well-defined discrete nodules with corresponding MRI were identified in 29 patients with whole-mount sections. Preoperative patient demographic and clinical characteristics are listed in Table 1. 86.2% (n = 25) preoperative MRIs were obtained for work-up of elevated PSA. Of these patients, 64% (n = 16) had a prior negative biopsy. The remaining 13.8% (n = 4) of MRIs were obtained because of prior history of prostate cancer; all of these patients had Grade Group 1 (Gleason 6) disease on prior biopsy. Prostate Imaging and Reporting Data System v.2 lesions >3 underwent fusion biopsy preoperatively, and all preoperative biopsies were ultimately negative. Final surgical pathology analysis demonstrated incidental T1a Grade Group 1 disease in three patients (10.3%) and T1b Grade Group 1 disease in one patient (3.4%). The remaining 25 patients (86.2%) had benign histopathology.

Table 1.

Demographic and Clinical Characteristics of Patients Undergoing Robot-Assisted Simple Prostatectomy

Mean age (years)	69.4 ± 5.2
Race/ethnicity, % (n)
Caucasian	79.3 (23)
Hispanic	3.4 (1)
African American	6.9 (2)
Asian	10.3 (3)
Mean BMI	29.2 ± 5.8
Mean prostate volume (cc)	133.9 ± 47.2
Mean PSA (ng/dL)	10.0 ± 7.7
Median no. identifiable nodules per patient (IQR)	4 (3–5)
Median IPSS score (n = 24) (IQR)	19.5 (11–22.5)
Median Qmax (cc/sec) (IQR)	8.3 (7.3–9.3)
Median PVR (cc) (IQR)	111 (48–174)
Alpha-adrenergic antagonist use, % (n)
Yes	51.7 (15)
No	48.3 (14)
5-alpha reductase inhibitor use, % (n)
Yes	20.7 (6)
No	79.3 (23)

BMI = body mass index; IPSS = international prostate symptom score; IQR = interquartile range; PSA = prostate specific antigen; PVR = postvoid residual; Qmax = maximal urinary flow rate.

Median phenotypic scores based on histology and MRI features were comparable (2, interquartile range [IQR] 2–3 vs 2, IQR 2–4, respectively; p = 0.63). Histology scores were positively correlated with MRI scores (Pearson's correlation r = 0.84, p < 0.0001). On univariate analysis, low signal intensity on T2-weighted images (p < 0.001) and on ADC map images (p < 0.001) were positively associated with stromal histology score. Heterogeneous T2 texture (p < 0.001), high SD on ADC map images (p = 0.005), and wash-out on DCE (p = 0.02) were negatively associated with stromal histology score.

Multiple stepwise linear regression analysis identified several MRI features that were independently associated with histologic phenotype (Supplementary Table S1). The associations between multiparametric MRI features and glandular or stromal histology are summarized in Table 2. Greater ADC signal intensity, or hyperintensity (p < 0.001), greater ADC SD (p = 0.03), increasing T2 texture heterogeneity (p = 0.003), and DCE wash-out (p = 0.001) were associated with more glandular histology. In contrast, stromal histology was associated with DCE wash-in (p = 0.03), as well as lower ADC signal intensity (hypointense), lower ADC signal SD, and lower T2 texture heterogeneity (or more homogeneous T2 signal). Signal intensity on T2-weighted imaging was not independently associated with either glandular or stromal histology score. The R² of the multivariable model was 0.42.

Table 2.

Association between Multiparametric MRI Features and Benign Prostatic Hyperplasia Nodule Histologic Phenotype

MRI parameter	Correlation with histologic phenotype		p
	Stromal	Glandular
Increasing ADC SD	−	+	0.03
DCE wash-in	+	−	0.03
DCE wash-out	− −	++	0.001
Increasing T2 texture heterogeneity	− −	++	0.003
High ADC signal (hyperintense)	− −	++	<0.001

Number of + or − correspond with relative strength of association.

ADC = apparent diffusion coefficient; DCE = dynamic contrast-enhanced; SD = standard deviation.

Discussion

In this study, we identify several unique features on multiparametric MRI that are significantly associated with specific histologic phenotypes using multiparametric MRI. Independent predictors of a stromal phenotype include ADC signal hypointensity, lower ADC SD, T2 signal homogeneity, and DCE wash-in. Independent predictors of a glandular phenotype were ADC signal hyperintensity, greater ADC SD, increasing T2 heterogeneity, and DCE wash-out. Moreover, there was good correlation between phenotypic scores assigned based on histologic and radiologic analysis, suggesting that stromal or glandular nodules may be uniquely identifiable on multiparametric MRI.

Prostate morphology on MRI has been previously investigated, and a prostate lobar classification based on anatomic features has been proposed by Wasserman and colleagues to help define morphologic phenotypes and guide treatment.¹¹ This validated classification system has been correlated with LUTS and is thought to facilitate determination of more targeted treatment options.^12,13 However, such a classification system relies on gross morphologic features of specific prostatic lobar zonal enlargement, and does not distinguish between more granular histologic phenotypes of a given prostate, which are hypothesized to underlie, at least in part, patient response to pharmacologic therapy for LUTS secondary to BPH.^4,6,10 Although this classification system may have utility to help determine the optimal procedural or surgical intervention for a given prostate morphology, it does not define clinically useful phenotypes for selection of pharmacologic treatment. This remains important as prior studies have demonstrated that response to medications is dependent on tissue histology.^4,5,14

To define clinically relevant histologic phenotypes, prostate needle biopsy was traditionally required to obtain required a histologic sample.¹ However, prostate biopsy is painful and comes with complications, including bleeding, infection, and potentially sepsis requiring hospitalization.¹⁵ In this setting, there has been growing interest in MRI as a noninvasive modality to evaluate BPH histology, but there is mixed evidence in the literature regarding the efficacy of MRI for this purpose.^8–10 Our study provides supporting evidence for the use of multiparametric MRI in this context and identifies specific imaging features that are associated with more glandular or stromal phenotypes.

Early studies by Ishida and colleagues⁸ and Weijers and colleagues⁹ correlating MRI and histopathologic findings relied primarily on representative samples from prostate biopsy or surgical specimens from transurethral resection of the prostate; this significantly limits the ability to accurately correlate histopathology of specific regions of the prostate with observed MRI features. Others have characterized histomorphology of the entire gland based on MRI features, but none have defined the unique MRI characteristics of individual glandular or stromal nodules.^7,10 Thus, a significant strength of our study lies in evaluation of intact transitional zone specimens, the most anatomically relevant portion of the prostate for BPH patients with bothersome LUTS, and in specific characterization of individual nodules, which provide a more granular understanding of imaging features associated with glandular or stromal phenotypes.

In addition, previous studies evaluating MRI as a noninvasive indicator of histologic phenotype have been limited by heterogeneous definitions of prostate morphology. Based on histopathologic correlations, Schiebler and colleagues⁷ determined glandular areas within the prostate to be those that had the highest signal intensity on MRI and areas of fibromuscular stroma to be those with the lowest signal intensity. Isen and colleagues¹⁰ defined nonstromal (i.e., glandular) BPH as those prostates with heterogeneous enhancement on T1-weighted imaging, an obvious surgical capsule, or an inner gland to total gland volume ratio of >0.75; stromal tissue was defined by the absence of any of these features. Similar to Weijers and colleagues,⁹ we objectively define imaging features that are independently associated with histologic phenotype, based on analysis of intact pathologic specimens, and include additional characteristics on ADC map, DWI, and DCE sequences beyond just T1- and T2-weighted sequences.

Based on these findings, we generated a predictive model to determine histologic phenotype from these imaging features. The R² value of this model was 0.42, which is comparable with that of similar predictive imaging models.⁹ Further prospective studies to validate the predictive nature of this model and inter-rater reliability of MRI evaluation of BPH features are warranted. Moreover, the correlation of these imaging findings and histologic phenotypes to clinical response to pharmacotherapy remains to be evaluated.

Our study has several limitations, including its retrospective nature and a relatively small sample size. A large proportion of men who ultimately underwent RASP did not have preoperative MRI, and when obtained, MRI was performed because of concerns of prostate cancer rather than for the evaluation of BPH. In addition, only patients with one or more well-defined prostate nodules on MRI with a corresponding nodule on histopathology were included, as the initial goal of this study was to identify objective measures on MRI that correlate to a specific pathology analysis, which required a well-defined sample. This may have introduced selection bias into our cohort. Furthermore, the Likert scale used to determine relative glandular and stromal components of each nodule has not yet been externally validated, thus limiting the generalizability of the model currently. Clinical variables were also not included in this model, which may add additional value; however, based on a preliminary sensitivity analysis, including clinical variables such as age, ethnicity, nodule size, and prior alpha-antagonist or 5-alpha reductase inhibitor use, the addition of such variables did not appear to significantly improve the R² value of the model (none were significantly associated with histologic phenotype scores on univariate analysis). Nonetheless, this initial study defines specific imaging features on multiparametric MRI that are independently predictive of histologic phenotype of BPH nodules and provides a predictive model that can determine the relative glandular and stromal composition of BPH nodules based on a noninvasive study.

Conclusions

There is a strong correlation between the imaging features on multiparametric MRI and the histologic phenotype of BPH nodules. In this initial study, we find that specific multiparametric MRI characteristics, including signal intensity on ADC map, ADC SD, T2 texture heterogeneity, and DCE wash-in and wash-out, independently correlate with the glandular and stromal composition. This study provides a foundation for future study identifying whole gland phenotypes with the goal of predicting response to pharmacologic intervention.

Abbreviations Used

ADC

apparent diffusion coefficient

BMI

body mass index

BPH

benign prostatic hyperplasia

DCE

dynamic contrast-enhanced

DWI

diffusion-weighted imaging

IPSS

international prostate symptom score

IQR

interquartile range

LUTS

lower urinary tract symptoms

MRI

magnetic resonance imaging

PSA

prostate-specific antigen

PVR

postvoid residual

Qmax

maximal urinary flow rate

RASP

robot-assisted simple prostatectomy

SD

standard deviation

Author Disclosure Statement

None of the authors have any competing financial interests or disclosures in relation to this article.

Funding Information

National Institute of Diabetes and Digestive and Kidney Disease, Grant number R01DK115477 (D.W.S)

Supplementary Material

Supplementary Table S1