Zonal anatomy of the prostate using magnetic resonance imaging, morphometrics, and radiomic features: impact of age-related changes

Abstract

Objective:

To evaluate the impact of age on the zonal anatomy of the prostate by MRI using morphometric and textural analysis.

Methods:

A total of 154 men (mean age: 63 years) who underwent MRI due to a high prostate-specific antigen (PSA) level were included retrospectively. At each MRI examination the following variables were measured: overall dimensions of the prostate (whole gland (WG), transitional zone (TZ), and peripheral zone (PZ)), and thickness of the anterior fibromuscular stroma (AFMS) and the periprostatic venous plexus (PPVP) on T ₂ weighted images. Identical regions of interest (ROIs) were delineated on the apparent diffusion coefficient (ADC) map on the anterior (horn) and posterior part of the PZ. Textural (TexRAD^®) parameter differences between TZ and PZ ROIs on T ₂ weighted images were analyzed by linear regression. Results were correlated with age (distributed into five decades from 22 to 89 years).

Results:

Age was positively correlated with PSA level and glandular volumes (WG, TZ, and TZ/WG ratio; p < 0.0001) and was negatively correlated with AFSM and PPVP thickness (p < 0.0001). There was a positive correlation between ADC values of the PZ and age (p = 0.003) and between entropy of the TZ and PZ and age (p < 0.001).

Conclusion:

Gradual variations in morphologic and textural features of the prostate were observed with age, mainly due to the increase in TZ volume while PZ volume tended to decrease. These modifications resulted in textural changes mainly at the expense of entropy.

Advances in knowledge:

Entropy could be relevant for studying the process of aging of the prostate.

Introduction

MRI is the reference for detecting and localizing prostate cancer ^1,18–20 Prostate MRI interpretation depends on zonal anatomy; whether lesion are located in the peripheral or in the transitional zone. ^2–4 Given the importance of zonal anatomy for prostate MRI interpretation, knowledge of age-related morphologic changes is essential. Currently, the most commonly used nomenclature to describe prostatic structures and morphologic and pathologic changes related to the age of the human prostate is McNeal’s zonal anatomy classification. ⁵ McNeal divided the prostate into four histologically distinct areas: a non-glandular anterior fibromuscular stroma (AFMS), two glandular regions termed peripheral (PZ) and central zones (CZ), and an additional glandular region that surrounded the prostatic urethra, referred to as the transition zone (TZ). Some data have been published on the changes seen by MRI during prostatic ageing, including an increase in prostate volume, morphologic distortion of the prostatic edges, and signal intensity modifications related to age. ^6–8 However, none of these studies have evaluated the relationship between morphological changes and textural analysis of prostate parenchyma.

Textural analysis is a novel imaging analysis technique that can quantify image heterogeneity resulting from changes not appreciated by the human eye. We hypothesized that complex morphologic changes in the prostate gland during normal development may be associated with textural changes, and that texture analysis tools could refine our comprehension of morphologic changes related to age.

The aim of this study was to evaluate the impact of age on the zonal anatomy of the prostate on MRI, combining morphometric and textural analysis.

Study population

Our institutional radiology database was reviewed retrospectively to identify patients who underwent prostate MRI at our institution between August 2017 and December 2019 for clinical suspicion of prostate cancer. Patients were randomly selected from our database in order to have a homogeneous distribution by age group. Patients with proven prostate cancer were excluded from the study as were patients with PI-RADS 4 and 5 on MRI. A total of 154 patients was included and grouped by age in decades: Group 1, < 50-years-old (n = 38); Group 2, 50‒59-years-old (n = 31); Group 3, 60‒69-years-old (n = 30); Group 4, 70‒79-years-old (n = 29); Group 5, > 79-years-old (n = 26). This study was approved by the local ethics committee.

MRI protocol

All images were acquired with a 3 Tesla MR imaging system (Siemens Healthcare, Erlangen, Germany) using a 32-channel phased-array torso coil according to a standardized protocol in which: (i) the patients were all advised to perform bowel preparation before the exam and to empty their bladder; (ii) 1 mg glucagon was administered intravenously to reduce peristaltic motion.

The standardized protocol (Table 1) performed in all examinations included a three-dimensional T ₂ weighted image, axial diffusion-weighted image (DWI) of the prostate using b-values of 100, 500, 1000, and 2000 s/mm² with inline reconstruction of calculated DWI images (3000 s/mm²) and apparent diffusion coefficient (ADC) map.

Table 1.

ESUR recommendation-compliant description of multiparametric MRI protocol

		3-Tesla MRI
	T 2 weighted three-dimensional	Diffusion-weighted imaging	VIBE dynamic contrast enhanced imaging
Section thickness/gap (mm)	0.85/0	4/1.2	3/0.6
Phase-encoding direction	Right-left	Anteroposterior	Anteroposterior
Repetition time (ms)	1550	6100	5,70
Echo time (ms)	173	64	2,66
Field of view	230 × 184	250 × 240	200 × 138
Acquisition matrix	320 × 288	108 × 108	256 × 179
b values (s/mm2)		50, 500, 1000
No. of repetitions	1,4	6, 9, 23	1
Turbo factor	24
Acquisition duration	5 min 35 s	4 min 3 s	1 min 23 s
Flip angle (degrees)	115	90	10

All patients received 1 mg glucagon intravenously. No endorectal coil was used. VIBE: volumetric interpolated breath-hold examination.

Dynamic contrast-enhanced imaging of the prostate was always obtained using a 3D fat-suppressed T ₁ weighted gradient echo sequence with a temporal resolution of 10 sec after an intravenous bolus injection of 0.2 mL/kg of gadoterate meglumine (Dotarem, Guerbet). Delayed post-contrast fat-suppressed T ₁ weighted imaging of the pelvis was also performed for nodal work-up.

Image analysis

All measurements were performed retrospectively by one observer on images displayed on our standard review console. T ₂ weighted images were evaluated with respect to prostatic zonal volumes: whole-gland (WG), TZ, and PZ volumes. On each examination, the prostate zonal volumes were calculated with the elliptical formula, length * width * height* 0.52. Length was defined as the greatest longitudinal distance on a mid-sagittal image, and width and height as the largest transverse and antero posterior distance on axial images. The reader was blinded from the PSA level and patient age. Measurements of the WG and TZ were made on the same images to calculate the TZ/WG ratio. Maximum measurable thickness at any level of the AFMS and periprostatic venous plexus (PPVP) was recorded as shown on Figure 1a. Regions of interest (ROIs) were drawn on the ADC map first on the anterior part (horn) and secondly in the posterior part of the PZ in order to keep the same size for both.

Figure 1.

a: Measurement method: maximal thickness of AFMS (blue line) and PPVP (red line) on axial T2-weighted image. b: bis. Changes in clinical and morphologic data with ranges of age.

MRI textural analysis: quantitative analysis

Axial T ₂ weighted images were analyzed using dedicated software for textural analysis (TexRAD Ltd, Feedback Plc, Cambridge, UK). Two identical ROIs were drawn by one observer in the PZ and TZ. The ROIs were placed in an area of normal appearing PZ and TZ signal. A suspicious lesion was defined as a focal increase in signal intensity within the PZ on DWI images or decreased voxel values on the ADC map, or a lenticular, homogeneous, moderately hypointense focal lesion on T ₂W in the TZ.

MRI textural analysis comprised image histogram analysis to quantify first-order statistics of mean, standard deviation (SD), entropy, mean of positive pixels, skewness, kurtosis, and sigma of the PZ and TZ ROIs. These parameters reflect, to varying extents, the number, intensity, and variability of areas of high and low signal intensity within the PZ and TZ.

Statistical analysis

Frequentist inference was performed to describe the distribution of variables. Associations between each MRI variable and age as continuous variable were assessed by Pearson’s correlation. MRI variables distribution was also determined after discretization by decade (<50-years-old, 50‒59 years; 60‒69 years; 79‒80 years; >80 years).

A supervised machine learning algorithm using Bayesian inference (Markov Blanket which is particularly helpful when there is a large number of variables in a data set) was also used to analyze the probability distribution and importance (mutual information). To evaluate overfit evidence (G-test) was performed between MRI variables which are discretized automatically by computing with the Tree algorithm (the main reason for discretizing continuous variables is the advanced capability of the model to capture more complex non-linear relationships between the variables) and age distribution classified manually by decades (<50-years-old, 50‒59 years; 60‒69 years; 79‒80 years; >80 years). To visualize how MRI patterns change with age, direct effects analysis were performed from supervised Bayesian network and are presented in Figures 2 and 3. Direct effects measure the sensitivity of one variable to changes in another, analyzing both linear and non-linear dependencies. It also provides a ranking of input variables based on their relative contributions mutual dependence (Mutual information) of the evaluated results. It is typically used to rank the significant factors contributing to risk. Direct effects are calculated from the percentage change (d) in age (mean value) divided by the percentage change in each inputted variable according to formula: Direct effect D_ex = dy/dx. So, in -axis represents the age and in x-axis, computed MRI patterns normalized variables to scale 0–100 to make them comparable. Mean values for MRI patterns (variables) are standardized between 0 and 100. The analysis was performed by using “soft evidence” feature, which gives virtual decimal values between actual values. The Mean m is computed using the numerical values (c) of the distribution states (according to discretization) and the marginal probability distribution (p) of the States: m=∑p*c.

Figure 2.

Relationship between age (<50-years-old, 50‒59-years; 60‒69-years; 79‒80-years; >80-years). and MRI variables (ADC and AFMS) normalized on probability distribution means

Figure 3.

Relationship between age (<50-years-old, 50‒59-years; 60‒69-years; 79‒80-years; >80-years). and MRI variables (entropy and sigma) normalized on probability distribution means

XLSTAT-Biomed (Addinsoft, Paris, France) and BayesiaLab 9.1 (Bayesia S.A.S, Change, France) software were used for the statistical analyses.

Results

Patients and demographic data

The demographic, biologic and morphologic data for the patients are summarized in Table 2. Mean patient age was 63 years (median: 62.5, range: 22–88 years), mean BMI was 25 (median: 25, range: 18.6‒38.6), and mean PSA level was 10 ng ml⁻¹ (median: 7.8, range: 0.7‒48).

Table 2.

Demographic, biologic, and morphologic data for the study population (N = 154)

Characteristic
Age Mean ± SD Range (min–max) Median [IQR: first–third] Variance	63.0 ± 14.4  (22.0–88.0)  62.6 [53.2–74.0]  208.9
BMI Mean ± SD Range (min–max) Median [IQR: first–third] Variance	25.2 ± 3.9  (18.6–38.6)  24.9 [23.2–26.4]  15.4
PSA (ng/ml) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	10.3 ± 8.6  (0.7–48.0)  7.8 [5.2–12.0]  74.5
TZ/WG ratio Mean ± SD Range (min–max) Median [IQR: first–third] Variance	0.6 ± 0.1  (0.2–0.9)  0.6 [0.5–0.7]  0
TZ volume (cc) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	35.5 ± 27.8  (2.2–221.0)  30.0 [17.9–43.5]  775.3
WG volume (cc) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	55.5 ± 34.3  (9.0–229.0)  49.0 [32.0–69.9]  1177.4
AFMS (mm) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	2.6 ± 2.0  (0–9.0)  2.4 [1.0–3.8]  4.1
PPVP (mm) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	4.0 ± 1.5  (1.0–11.0)  4.0 [3.0–5.0]  2.4
PZ ADC (anterior part) (mm2/s) Mean ± SD Range (min–max) Median [IQR: first–third] Variance	1652.3 ± 385.9  (875.0–3254.0)  1640.0 [1352.0–1930.0]  148 985.4

BMI: body mass index; PSA: prostate-specific antigen; TZ: transitional zone; WG: whole gland; PZ: peripheral zone; AFMS: anterior fibro-muscular stroma; PPVP: peri-prostatic venous plexus; ADC: apparent diffusion coefficient; ant: anterior; post: posterior.

Mean WG, TZ, and PZ (=WG TZ) volumes were 55.5 ml (median: 49, range 9‒229), 35.5 ml (median: 30, range 3‒221 ml) and 20 ml (median: 16.9, range, 4‒75 ml), respectively.

Mean WG volume was 29 ml for males aged <50 years (n = 37), 57 ml for males aged 50‒59 years (n = 57), 66 ml for males aged 60‒69 years (n = 30), 55 ml for males aged 70‒79 years, and 79 ml for males aged >80 years (n = 29).

Mean TZ volume was 16 ml for males aged <50 years, 36 ml for males aged 50‒59 years, 41 ml for males aged 60‒69 years, 33 ml for males aged 70‒79 years, and 58 ml for males aged >80 years.

Mean size of the AFMS and PPVP was 2.6 mm (median: 2.45, range: 1‒9), and 4 mm (median: 4, range: 1‒11), respectively.

Correlation between clinical and morphologic parameters and age

Age was positively correlated with serum PSA level, WG, and TZ volumes (p < 0.0001) (Figure 1b). In contrast, age was negatively correlated with AFMS thickness Figure 1b) and PPVP diameter (p < 0.0001 and p < 0.002)) (Table 3).

Table 3.

Correlation coefficients for clinical and morphological data with age

Variable	Age	BMI	PSA	TZ/WG	TZ	WG	AFMS	PPVP	ADC-ant	ADC-post
Age	1	0.033	0.369	0.470	0.414	0.406	−0.047	−0.245	0.235	0.076
BMI	0.033	1	0.092	0.031	0.045	0.028	0.066	0.006	0.015	−0.014
PSA	0.369	0.092	1	0.256	0.254	0.221	−0.153	−0.107	−0.003	−0.052
TZ/WG	0.470	0.031	0.256	1	0.630	0.452	−0.428	−0.049	0.134	0.081
TZ	0.414	0.045	0.254	0.630	1	0.951	−0.517	0.223	0.223	0.087
WG	0.406	0.028	0.221	0.452	0.951	1	−0.522	0.184	0.284	0.117
AFMS	0.477	0.066	0.153	−0.428	−0.517	−0.522	1	0.057	−0.273	−0.105
PPVP	0.245	0.006	0.107	−0.049	0.223	0.184	0.057	1	−0.052	−0.095
ADC-ant	0.235	0.015	0.003	0.134	0.223	0.284	−0.273	−0.052	1	0.547
ADC-post	0.076	0.014	0.052	0.081	0.087	0.117	−0.105	−0.095	0.547	1

BMI: body mass index (kg/m²); PSA: prostate-specific antigen (ng/ml); TZ: transitional zone (ml); WG: whole gland (ml); AFMS: anterior fibro-muscular stroma (mm); PPVP: peri-prostatic venous plexus (mm); ADC: apparent diffusion coefficient (mm²/s); ant: anterior; Bold data indicate correlations that are significant (p value < 0.05)

Figure 4.

AFMS thickness decreased significantly with age (p<0.0001); MRIs from groups 1, 2, and 3, respectively

ADC values and textural metrics

Mean ADC values of PZ and TZ ROIs were 1666 mm²/s (min–max: 813‒4306) and 1652 mm²/s (min–max: 875‒3254), respectively. There was a positive correlation between ADC value of PZ and age (p < 0.0001) (Table 3, Figure 3).

There were significant positive correlations between entropy in the TZ and age at all filter values (SSF: 0, 2, 3, 4, 5, 6; p < 0.001) (Table 4, Figures 3 and 5), between age and sigma in the PZ and TZ (p < 0.001) (SSF: 2, 3, 4, 5, 6, and 2, 3, 4, 5, 6), and between entropy of the PZ and TZ/WG ratio (SSF 2, 4, 5). Entropy over time showed a peak in the fifth and sixth decades at all filter values (Figure 5, Figure 6).

Table 4.

Summary of significant associations between texture features and MRI according to age and prostate volume

Filter size (mm)	MRI features	Textural parameter	p value
2	PZ ROI and age	Sigma	0.00001
2	PZ ROI and age	Entropy	0.00014
2	PZ ROI and TZ/WG	Entropy	0.034355
2	TZ ROI and TZ/WG	Entropy	0.000293
3	PZ ROI and age	Sigma	0.000010
3	TZ ROI and age	Entropy	0.000010
3	TZ ROI and TZ/WG	Entropy	0.000008
4	PZ ROI and age	Sigma	0.000010
4	TZ ROI and age	Entropy	0.000020
4	PZ ROI and TZ/WG	Entropy	0.031047
4	TZ ROI and TZ/WG	Entropy	0.000043
5	PZ ROI and age	Sigma	0.000010
5	TZ ROI and TZ/WG	Entropy	0.044130
5	TZ ROI and age	Entropy	0.002328
5	PZ ROI and TZ/WG	Entropy	0.04413
5	TZ ROI and TZ/WG	Entropy	0.006224
6	PZ ROI and age	Sigma	0.000010
6	TZ ROI and age	Entropy	0.001963

PZ: peripheral zone; TZ: transitional zone; WG: whole gland; ROI: region of interest.

Figure 5.

Evolution of transitional zone entropy with age (<50-years-old, 50‒59-years; 60‒69-years; 79‒80-years; >80-years).

Figure 6.

Mean and median values (and interquartile range 25 %-75%) for entropy at all filter according to age.

Bayesian inference was used to explore conditional dependence relationships between age and MRI variables.

The relationship between age and MRI variable normalized probability distribution means are presented in Figure 5.

The supervised learning algorithm applied was the Markov Blanket learning algorithm. This algorithm restricts the selection of nodes to nodes belonging to the target node’s fathers, sons, and spouses. ⁹

Discussion

We found, in this study, a gradual variations in morphologic and textural features with age.

Textural feature analysis depicted variations in entropy with respect to age and prostate gland volume. A strong correlation was found between entropy of the TZ and age at all filters (p < 0.00001), and between entropy of the PZ and TZ/WG ratio. Entropy is a measure of image “irregularity”. An increase in entropy value indicates non-uniformity of the PZ and the TZ increasing with age and prostate volume under normal conditions. Different growth characteristics in each prostate zone may contribute to differences in the overall growth rate with age. ¹⁰ These changes that occur over time in the prostate gland have been described by Mc Neal et al. ⁵ In the TZ, two distinct stages have been described: first the development and increase in number of nodules in the glandular tissue from the fourth decade into the 60 s, and then an abrupt increase in the mass of individual nodules. ¹⁰ Inversely, PZ is compressed by the increase in size of the TZ, with entropy in this area highly correlated with the TZ/WG ratio. All these structural changes may explain the positive correlation between entropy and age with a peak obtained during the fifth and the sixth decade.

We also observed a strong correlation between increasing PZ ADC values and age. De-Visshere et al ¹¹ noticed that cystic atrophy of large glands occurred predominantly in older males. Because the water content of these large glands is higher than in pure normal glands it could explain the increase in ADC values of the PZ with age.

Most of available studies about textural analysis in prostate MRI concern prostate cancer detection and risk stratification. Wibmer et al ¹² found in their study that Haralick texture features derived from T ₂ weighted images and ADC maps have the potential to differentiate between prostate cancer and non-cancerous prostate tissue and that ADC map textural features correlated significantly with gleason score (GS). Nketiah et al ¹³ found that T ₂W MRI derived textural features correlate significantly with GS. In this study, entropy was found to correlate significantly with GS, possibly explained by the increase complexity of the tissue secondarily to glandular structure deformation.

Few studies have evaluated the changes in textural features in non-pathologic states and none in prostate imaging. In their study, Kalpana et al ¹⁴ correlated MRI features of brain white matter in normal subjects and those of HIV+ patients using a computational approach. They found that in non-pathologic states changes (either decrease or increase) in textural features (Haralick’s parameter) may occur between 9 and 50 years of age and could be explained by changes in the architecture of the white mater.

Unlike many other organs that exhibit atrophy with age, prostate volume increases with age,. ¹⁰ In this study, values of 29 ml were being found for males aged <50 years to 79 ml for males > 80 years. This increase in size was highly correlated with the increase in TZ volume and TZ/PZ ratio. Knowledge of these morphologic changes on MRI will give us a better understanding of the meaning of variations in signal essential to improve the quality of reporting. In a previous study of 500 patients, Turkbey et al found that the WG volume peak was in the sixth and seventh decades of life and was mainly driven by changes in the TZ volume. ⁷ Our results are similar with a peak in Group 3 (sixth and seventh decades). However, we found that the maximum volume was obtained in the last decade (Group 5, > 80 years). This difference may be due to a difference in sampling in groups; median age in the study of Turkbey et al was 60 years (range: 38‒83 years), whereas it was 62.5 years in our study (range: 22‒88 years). In addition, PZ volume decreased with age and in parallel with the increase in TZ volume. In a retrospective study, Matsugasumi et al ⁸ evaluated the relationship between morphology of the PZ and age in a cohort of 307 men (156 from Japan). The authors found that as the TZ increases in size the PZ becomes thinner as part of the stretched surgical capsule.

As described previously, ^7,15 we also found a positive correlation between serum PSA level, WG (p = 0.006) and TZ volume (p = 0.002). These findings are consistent with the concept that the TZ, which consists of a mixture of stromal and glandular hyperplasia developed in response to testosterone, is the principal determinant factor in benign prostatic hyperplasia and elevated serum PSA levels. ¹⁶

Because prostate cancer could extend in the AFMS, knowledge of its anatomy and evolution with age is important. The AFMS has low signal intensity on T ₂W and forms the entire anterior and convex surface of the prostate as a thick apron, extending from the bladder neck to the prostate apex. Like Hricak et al, ¹⁷ we found that the thickness of the AFMS decreases proportionally with the increase in TZ, possibly due to compression or stretching by the enlarged gland. The AFMS was most prominent in young patients with small prostate glands (4.4 mm in Group 1 vs 1.8 mm in Group 5).

The PPVP, which had a round, tubular structure on anterior and lateral aspects of the prostate, was generally prominent in young patients with small glands and became thinner with increasing age. Although this phenomenon may be due to venous compression by gland enlargement, the inverse correlation of venous caliber to patient age was much greater than the inverse correlation of venous caliber to gland size. These results are concordant with those of Allen et al. ⁶

The present study has some limitations. Morphometric analyses of prostate volume were done by one reader only. We did not monitor the longitudinal changes in zonal volume in our cohort or correlate morphologic changes with the International Prostate Symptom Score. Our results may be related to true anatomic differences in the prostate gland with age, as reflected in their parenchymal patterns, although other biological factors may be involved such as biochemical differences due to inflammatory or hormonal changes, unfortunately we could not have access to biological data as testoteronemia. The relationships between textural features and tissue composition (cytoplasm, stroma, luminal space) need to be explored with histopathological correlations.

Conclusion

Knowledge of the anatomic modifications of the prostate gland with age is a prerequisite to interpreting prostate MRIs. The results of this study show gradual variations in morphologic and textural features with age, mainly due to the increase in TZ volume, while the PZ tends to decrease. These modifications resulted in textural changes mainly at the expense of entropy.