Differentiation of prostate cancer and stromal hyperplasia in the transition zone with histogram analysis of the apparent diffusion coefficient

Abstract

Background

Prostate cancer and stromal hyperplasia (SH) in the transition zone (TZ) are difficult to discriminate by conventional magnetic resonance imaging (MRI) and diffusion-weighted imaging (DWI).

Purpose

To investigate the apparent diffusion coefficient (ADC) of prostate cancer and SH in the TZ with histogram analysis and the ability of ADC metrics to differentiate between these two tissues.

Material and Methods

Thirty-three cancer and 29 SH lesions in the TZ of 54 patients undergoing preoperative DWI (b-value 0, 1000 s/mm²) were analyzed. All the lesions on the MR images were localized based on histopathologic correlations. The 10th, 25th, and 50th percentiles, and the mean ADC values were calculated for the two tissues and compared. The efficiencies of the 10th, 25th, and 50th ADC percentiles in differentiating the two tissues were compared with that of the mean ADC with receiver operating characteristic (ROC) analysis.

Results

The 10th, 25th, and 50th percentiles and mean ADC values (×10⁻³ mm²/s) were 0.86 ± 0.15, 0.89 ± 0.16, 0.94 ± 0.16, and 1.03 ± 0.17 in SH and 0.64 ± 0.12, 0.69 ± 0.12, 0.72 ± 0.16, and 0.83 ± 0.15 in TZ cancer, respectively. The parameters were all significantly lower in cancer than SH. The 10th ADC percentile yielded an area under the ROC curve (AUC) of 0.87 for the differentiation of carcinomas from SH, which was higher than the mean ADC (0.80) (P < 0.05), and the AUCs of the 25th (0.82) and 50th (0.83) percentiles exhibited no differences from those of the mean ADC (P > 0.05).

Conclusion

Histogram analysis of ADC values may potentially improve the differentiation of prostate cancer from SH in the TZ.

Keywords

Magnetic resonance (MR) diffusion/perfusion prostate neoplasms – primary

Introduction

Up to 30% of prostate cancers arise in the transition zone (TZ) (1); this prevalence is less than in the peripheral zone (PZ), but those in the TZ are generally considered more difficult to diagnose using conventional magnetic resonance imaging (MRI) (2,3). The T2-weighted (T2W) features such as homogeneous and low T2 signal intensity, ill-defined margins, and lack of a capsule were commonly considered as the criteria for the identification of TZ cancer. However, in a study by Yoshizako et al. (2), the sensitivity and specificity for the detection of TZ cancer with these criteria were only 61.5% and 68.8%, respectively. The major obstacle to the identification of TZ cancer is considered to be caused by benign prostatic hyperplasia (BPH), particularly stromal hyperplasia (SH), which presents T2W imaging features resembling those of TZ carcinoma (4).

Diffusion-weighted imaging (DWI) has also been employed in the diagnosis of TZ cancer in the last decade, and some studies have revealed a lower apparent diffusion coefficient (ADC) value in TZ cancer compared with BPH (5). However, a previous study noted that SH also exhibited a lower ADC value than normal and glandular BPH tissue because of its denser structure resulting from increased cellularity and reduced glandular tissue (6). Oct et al. (4) also reported an obvious overlap between the ADC values of SH and TZ cancer, leading in a relatively lower area under the receiver operating characteristic curve (AUC) of 0.75 for the differentiation of the two tissues and indicating that the differential diagnosis of TZ cancer was still complicated even under the assistance of DWI exam. DWI images with a high b value (e.g. 1500 and 2000 s/mm²) were considered to have contrast ratios significantly higher than conventional DW images (7). A previous study reported that incorporating the ADC value with a b value of 2000 s/mm² into the T2W image plus the ADC value with a b value of 1000 s/mm² could further increase the sensitivity for detection of TZ cancer (3); however, more studies are needed to assess its capability in the differential diagnosis of TZ cancer.

Recently, because of its capacity of volumetric analysis of the whole lesion, ADC histogram analysis has been applied in cancer diagnosis. Several previous studies have proved that the ADC percentiles could better differentiate cancerous from benign lesions or tumors with different grades (8,9). To our knowledge, there are currently no studies of the application of an ADC histogram to cancer diagnosis in the TZ of the prostate. Therefore, the aim of the present study was to determine the role of ADC value histogram analysis in the differentiation of TZ cancer and SH; the efficiencies of ADC percentiles for differentiation were evaluated and compared to the mean ADC.

Material and Methods

Patients

From July 2012 to July 2015, we retrospectively analyzed 106 consecutive patients with prostate cancer proven by biopsy who underwent prostate MRI in our hospital before prostatectomy. All patients were retrieved from our radiology report and history system; none of these 106 patients received neoadjuvant hormonal, chemotherapy, or radiation therapy before the MRI scan and prostatectomy. Thirty of these patients had pathologically confirmed TZ cancer and four of them had PZ cancer at the same time. The remaining 55 patients had cancer only in the PZ, and among these patients, 43 with pathologically proven BPH nodules with dimensions ≥ 5 mm and without obvious deformation in the TZ were also assessed in the study. Therefore, 73 patients were included for further study. The mean interval was 20 days (range, 15–28 days) between biopsy and MRI exam, and four days (range, 2–7 days) between MRI and prostatectomy.

MRI acquisition

All MR images were acquired by a 3.0-T MRI system (Signa HDx; GE Healthcare, Milwaukee, WI, USA) with an eight-channel phased-array surface coil.

Sequences included transverse, sagittal, and coronal T2W imaging, transverse T1-weighted (T1W) imaging, and DWI through the prostate and seminal vesicles. The T2W imaging was performed with a fast-recovery fast-spin echo (FSRSE) sequence (TR/TE, 5000/87.9 ms; number of excitations (NEX), 4; slice thickness, 4 mm; space, 1 mm) and T1W imaging with a fast spoiled gradient echo (FSPGR) sequence (TR/TE, 150/3 ms; slice thickness, 4 mm; space, 1 mm). The DWI images were acquired with a single-shot echoplanar imaging (SS-EPI) sequence (TR/TE, 4000/71.9 ms; field of view [FOV], 260 × 260 mm; matrix, 128 × 128; NEX, 4, slice thickness, 4 mm; space, 1 mm) with identical slice location as the transverse T2W imaging, the b values were between 0 and 1000 s/mm², and the acquisition time was approximately 168 s. To reduce the scan time and magnetic susceptibility artefacts, a parallel imaging technique, Array Spatial Sensitivity Encoding Technique (ASSET), was performed.

MRI–histopathologic correlations

Whole-mount sections (thickness, 7–8 µm) of the prostate were prepared as previously described (10). Hematoxylin and eosin (H&E) stained slices of the patients were reviewed by our faculty pathologist (eight years of experience in genitourinary pathology) who was unaware of the MRI findings. The TZ cancer and SH lesions were identified on the slice; in particular, the SH lesions were defined as well-defined nodular hyperplasia lesions that were composed of more than 50% stromal tissue. The sizes, locations of all lesions, and the Gleason score of the TZ cancer were recorded on a schematic prostate diagram.

The MRI–histopathologic correlations were then performed by two radiologists (with seven and 17 years of experience, respectively, in imaging diagnosis of prostate disease) in consultation with the pathologist on a workstation (Advantage Windows, version 4.6, GE Healthcare).

Cancer lesion data were obtained from patients with TZ cancer and SH lesion data were obtained from patients with cancer only in the PZ. The locations of the lesions were identified on the T2W images referring to pathological diagrams, the landmarks for alignment of T2W images and pathological diagrams, including the morphologic features of the PZ, TZ, the distances from the base or apex, the locations of the ejaculatory ducts, urethra, and any explicit BPH nodules. To be considered a match, the TZ cancer and SH must be a lesion at the same supero-inferior level of the prostate and in the same quadrant of the TZ. All TZ cancer and SH foci with the largest histological dimensions of ≥ 5 mm were matched to T2W images by consensus among the radiologists and the pathologist and were included in the subsequent analyses.

Image processing and data acquisition

The analysis of the T2W and DWI images were also performed by the two radiologists, and T1W images were consulted to exclude hemorrhage (hyperintense) in the tissue. The size and MRI characteristics of the included TZ cancer and SH foci were interpreted, and their signal intensity (hyper- or hypointense), homogeneity (homo- or heterogeneous), margins (well- or ill-defined) on T2W images, and the signal intensity (hyper- or hypointense) on the DWI images were recorded.

After the ADC maps were reformatted by the Reformat software of the workstation, the mean and ADC value percentiles were calculated with the histogram analysis. To measure the mean and percentile ADC values of the lesions, the region of interest (ROI) circles were placed together by the two radiologists on the b = 0 mm/s² DWI images with reference to the T2W images because they have a better tumor visibility than the b = 1000 s/mm² DWI and ADC maps (11). The circles were drawn as large as possible in the suspected lesion areas. The data from each lesion were measured three times within the same site and then averaged. Differences in measurement were resolved by consensus.

Statistical analysis

Statistical analyses of the data were carried out with STATA 10.0 software (Stata Corporation, College Station, TX, USA). Student’s paired t-tests were used to examine the histogram parameter differences between the tissue types. The discrimination efficiencies of the 10th, 25th, and 50th ADC percentiles were evaluated using receiver operating characteristics (ROC) curve analyses and then compared to the mean ADC by AUCs. The differences in the T2W and DWI image characteristics between the different prostate tissues were assessed by Fisher’s test. A P value of < 0.05 was considered to be a significant level for all analyses.

Results

In total, 28 patients (mean age, 64.6 years; age range, 51–73 years; median serum prostate-specific antigen [PSA] level, 18.2 ng/mL; range, 7.2–33 ng/mL) with 33 TZ cancer (average greatest dimension on pathology, 10 mm; range, 5–26 mm; average greatest dimension on T2W imaging, 11.6 mm; range, 6–28 mm; median Gleason score, 4+3; range, 3+3–5+5) and 26 patients (mean age, 58.6 years; age range, 49–71 years; median PSA level, 14.5 ng/mL; range, 8–43 ng/mL) with 29 SH (average greatest dimension on pathology, 9.8 mm, range, 5–18 mm; average greatest dimension on T2W imaging, 11.1 mm; range, 6–24 mm) were included in the final study cohort, and the remaining two patients with TZ cancer and 17 patients with only PZ cancer were excluded because of the absence of qualified lesions.

The T2W and DWI (b = 1000 s/mm²) image characteristics of TZ cancer and SH were compared in Table 1. All of the cancer and a majority of the SH lesions exhibited homogeneously low signal intensities on T2W images, and most of the cancer and SH lesions had ill-defined margins. On DWI, all of the cancers showed significantly high signals, the proportion of SH with hyperintensity was significantly lower than cancer but still reached nearly two-thirds.

Table 1.

The T2W and DW (b = 1000 s/mm²) image features of the prostate cancer and SH.

Tissue	TZ cancer (n = 33)	SH (n = 29)
Signal intensity on T2W image
Hypointense	33	26
Hyperintense	0	3
Homogeneity
Homogeneous	33	23*
Heterogeneous	0	6
Margin
Ill-defined	25	20
Well-defined	8	9
DWI signal
Significantly hyperintense	33	19*
Slightly hyperintense or isointense	0	10

Significant difference compared with TZ cancer

DWI, diffusion-weighted imaging; T2W, T2-weighted; TZ, transition zone.

Histogram analysis of the ADC value revealed that both tissues exhibited unique parameters (Table 2, Figs. 1 –3). The ADC value of the SH was significantly higher than that of cancer (P < 0.05). However, a notable overlap between the ADC values of the two tissues was observed. The ADC percentiles for the cancers were also significantly lower than those of the SH.

Table 2.

The 10th, 25th, and 50th percentile and mean ADC values (×10^–3 mm²/s) of prostate cancer and SH.

Tissue	TZ cancer (n = 33)	SH (n = 29)
10th percentile ADC	0.64 ± 0.12	0.86 ± 0.15*
25th percentile ADC	0.69 ± 0.12	0.89 ± 0.16*
50th percentile ADC	0.72 ± 0.16	0.94 ± 0.16*
Mean ADC	0.83 ± 0.15	1.03 ± 0.17*

Significant difference compared with TZ cancer.

ADC, apparent diffusion coefficient; SH, stromal hyperplasia; TZ, transition zone.

Fig. 1.

Mean and percentile ADC values for TZ cancer and SH. The mean and 50th, 25th, and 10th percentile ADC values of TZ cancer were all significantly lower than those of SH.

Fig. 2.

Images of a 58-year-old patient with prostate cancer in the TZ. (a) The transverse T2W image showed a homogeneously hypointense lesion in the anterior TZ (arrow). (b) The lesion showed a moderately high signal on the DW image (b = 1000 s/mm²) (arrow). (c) The ROI (circle) was drawn on the DW image (b = 0 s/mm²). (d) The ADC value of the lesion was remarkably lower compared to the surrounding normal tissue on the ADC map. (e) The histogram of the whole lesion revealed a large portion of pixels with low ADC values; the 10th, 25th, 50th percentile and mean ADC values (×10⁻³ mm² /s) were 0.65, 0.79, 0.84 and 0.94, respectively. (f, g) Photomicrograph of histopathologic slice (H&E stain; original magnification, ×20 and ×200) shows a prostate cancer lesion with a Gleason score of 3+4, and the magnification of cancer tissue in the circle of (f) is shown in (g).

Fig. 3.

Images of a 65-year-old PZ cancer patient with an SH nodule in the TZ. (a) The transverse T2W image showed an ill-defined nodule with a low signal in the left side of the TZ (arrow). (b) The nodule showed slightly high signal on the DW image (b = 1000 s/mm²) (arrow). (c) ROI (circle) was placed on b = 0 s/mm² DW image. (d) The ADC of SH was lower than the normal tissue around it. (e) The histogram of the lesion displayed relatively larger ADC values compared with the prostate cancer in Fig. 2; the 10th, 25th, and 50th percentiles and mean ADC values (×10⁻³ mm² /s) were 0.85, 0.88, 0.94 and 1.04, respectively. (f, g) The histopathologic slide (×20 and ×200) demonstrated a predominantly stromal component of the nodule.

The AUC of the ADC for the discrimination of TZ cancer from SH was 0.80, and at a cutoff of 0.93 × 10^–3 mm²/s, it provided a sensitivity of 72.41% and a specificity of 78.79% in the differentiation of cancer from SH. The 25th and 50th ADC percentiles produced AUCs of 0.82 and 0.83, respectively; these values were close to the mean ADC value. The 25th ADC percentile resulted in a sensitivity of 75.86% and a specificity of 81.82% at the cutoff of 0.78 × 10^–3 mm²/s, and the 50th ADC percentile produced values of 72.41% and 81.82%, respectively, at the cutoff of 0.83 × 10^–3 mm²/s. However, the 10th ADC percentile yielded a higher AUC (0.87) than the mean ADC value (P < 0.05) in the differentiation of cancer from SH (Fig. 4). At the cutoff of 0.77 × 10^–3 mm²/s, a sensitivity of 75.86% and a specificity of 84.85% were observed in the differentiation of cancer from SH.

Fig. 4.

ROC curves for the differential diagnosis of TZ cancer and SH based on diffusion parameters. The 10th ADC percentile produced a significantly greater AUC than the mean ADC in the discrimination of TZ cancer from SH, but the AUCs of the 25th and 50th percentiles showed no significant difference from that of the mean ADC value.

Discussion

The aim of the present study was to evaluate the efficiencies of ADC percentiles for differentiation between TZ cancer and SH compared to the mean ADC.

The results revealed that the 10th ADC percentile exhibited better performance than the mean ADC in the differentiation of cancer from SH and in the TZ, suggesting that histogram analyses of ADC value could provide a new approach for the diagnosis of prostate cancer.

In our study, SH and TZ cancer share many common T2W image characteristics (mainly hypointensity, ill-defined margin), as reported in previous studies (4,12), indicating that conventional MRI was insufficient for the diagnosis of TZ cancer.

DWI revealed that nearly two-thirds (19/29) of the SH nodules exhibited high signal intensities, which was similar to cancer. Although the cancers exhibited a lower ADC value than the SH, the overlap between them made differentiation still difficult, with a sensitivity of 72.41% and a specificity of 78.79%. These results were similar to those of previous studies (4,10). Oct et al. (4) observed a similar overlap between the ADC of cancer and SH and reported an AUC for the differentiation of the tissues of 0.75. Liu et al. (10) also confirmed this obstacle to the differential diagnoses of TZ cancer and SH.

The parameters in the histogram analyses of the ADC values were also lower for cancer than for SH, and the 10th ADC percentile provided greater accuracy than the mean ADC in the differential diagnosis of TZ cancer. The use of ADC percentiles has been applied in some previous studies and is considered a new approach to differentiate cancer and benign lesions or different grades of tumors. Xue et al. (8) reported that the 5th and 45th ADC percentiles rather than the mean ADC value could discriminate well or moderately differentiated cervical squamous cell carcinoma (SCC) from poorly differentiated SCC. Moreover, in a breast cancer study, the AUCs for the differentiation of malignant and benign lesions of the minimum and 25th percentile ADCs were significantly higher than those of the mean and median ADCs, indicating that the minimum and 25th percentile ADCs may play an important role in breast lesion discrimination (9). There are also a few previous studies of the application of histogram analysis ADC values in the prostate. In one recent study, Donati et al. (13) proved that the 10th percentile ADC value correlated with the Gleason score better than the other ADC parameters, and another study proved that the 10th pixel-wise ADC percentile is more effective than the average ADC in the differentiation of prostate cancer from normal PZ tissue (14). Our study produced results similar to those of the latter study in that the 10th ADC percentile provided a greater accuracy in the differential diagnosis of prostate cancer. Donati et al. (13) suggested that this difference may be because the 10th ADC percentile was more effective than the average ADC in the identification of areas in which cancerous glands were intermixed with benign prostatic tissue rather than completely replacing the prostatic tissue.

In our study, the superior performance of the 10th ADC percentile relative to the mean ADC may be explained by the heterogeneous histological characteristics of prostate cancer (15) and SH foci (16). On the one hand, in cancer tissue, the increased cellularity and acinar destruction result in a dense structure and a lower ADC value. In SH, the increase in the stromal component and reduction in the acinar component also limit the diffusion of water and reduce the ADC value to a level similar to that of cancer. On the other hand, large proportions of interspersed normal glandular prostatic tissues within the two tissue types influence the mean ADC to artificially increase the values of the former parameter, resulting in the two tissues being more difficult to distinguish. However, in the histogram analysis, within the tumors/SH with heterogeneous cellularities, focal areas of high cellularity were represented to a greater extent by the 10th ADC percentile than by the mean ADC, indicating that the 10th ADC percentile could partially reduce the effects of the acinar components and emphasize the cellular and stromal components; thus, the complexity of the differentiation was decreased.

However, it should be noted that although the difference between the accuracies of the mean ADC and the 10th percentile ADC was statically significant, the sensitivity was only increased by 3%, and the specificity was only increased by 6% via the use of the 10th percentile ADC. These findings indicated that even with the assistance of histogram analysis, the influence of the interspersed normal glandular tissues and perfusion components were not satisfactorily reduced, or the difference between SH and prostate cancer remained subtle with our current technique. Thus, future work is required to improve the DWI-based diagnosis of prostate cancer in the TZ.

There were some limitations in our study. First, the relatively small sample size and the criteria of the lesions (i.e. the diameter of lesion) in the study may potentially result in some selection bias. Second, the image analysis and data measurements were done by two radiologists together; thus, the inter-reader variability was not evaluated. Third, the reliability of the correlation between MRI and whole-mount step-section specimens is crucial. Although we attempted to minimize inaccuracies by performing the correlations based on a consensus and carefully avoided the inclusion of healthy tissue in the tumor ROIs, uncertainty remains.

In conclusion, our results suggest that when ADC parameters were determined based on whole-lesion histogram analysis, the 10th ADC percentile provided improved accuracy in the differentiation of prostate cancer from SH in the TZ. Therefore, the 10th ADC percentile may improve the diagnosis of prostate cancer.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by grants from Fudan University Shanghai Cancer Center (No. YX 201501).

References

McNeal

Redwine

Freiha

et al.

Zonal distribution of prostatic adenocarcinoma. Correlation with histologic pattern and direction of spread. Am J Surg Pathol 1988; 12: 897–906.

Yoshizako

Wada

Hayashi

et al.

Usefulness of diffusion-weighted imaging and dynamic contrast-enhanced magnetic resonance imaging in the diagnosis of prostate transition-zone cancer. Acta Radiol 2008; 49: 1207–1213.

Rosenkrantz

Kim

Campbell

et al.

Transition zone prostate cancer: revisiting the role of multiparametric MRI at 3T. Am J Roentgenol 2015; 204: W266–W272.

Oto

Kayhan

Jiang

et al.

Prostate cancer: differentiation of central gland cancer from benign prostatic hyperplasia by using diffusion-weighted and dynamic contrast-enhanced MR imaging. Radiology 2010; 257: 715–723.

Elbuluk

Muradyan

Shih

et al.

Differentiating transition zone cancers from benign prostatic hyperplasia by quantitative multiparametric magnetic resonance imaging. J Comput Assist Tomogr 2016; 40: 218–224.

Noworolski

Vigneron

Chen

et al.

Dynamic contrast-enhanced MRI and MR diffusion imaging to distinguish between glandular and stromal prostatic tissues. Magn Reson Imaging 2008; 26: 1071–1080.

Wetter

Nensa

Lipponer

et al.

High and ultra-high b-value diffusion-weighted imaging in prostate cancer: a quantitative analysis. Acta Radiol 2015; 56: 1009–1015.

Xue

Ren

Yang

et al.

Histogram analysis of apparent diffusion coefficient for the assessment of local aggressiveness of cervical cancer. Arch Gynecol Obstet 2014; 290: 341–348.

Suo

Zhang

Cao

et al.

Characterization of breast masses as benign or malignant at 3.0T MRI with whole-lesion histogram analysis of the apparent diffusion coefficient. J Magn Reson Imaging 2016; 43: 894–902.

10.

Liu

Zhou

Peng

et al.

Differentiation of central gland prostate cancer from benign prostatic hyperplasia using monoexponential and biexponential diffusion-weighted imaging. Magn Reson Imaging 2013; 31: 1318–1324.

11.

Rosenkrantz

Kong

Niver

et al.

Prostate cancer: comparison of tumor visibility on trace diffusion-weighted images and the apparent diffusion coefficient map. Am J Roentgenol 2011; 196: 123–129.

12.

Ishida

Sugimura

Okizuka

et al.

Benign prostatic hyperplasia: value of MR imaging for determining histologic type. Radiology 1994; 190: 329–331.

13.

Donati

Mazaheri

Afaq

et al.

Prostate cancer aggressiveness: assessment with whole-lesion histogram analysis of the apparent diffusion coefficient. Radiology 2014; 271: 143–152.

14.

Peng

Jiang

Yang

et al.

Quantitative analysis of multiparametric prostate MR images: differentiation between prostate cancer and normal tissue and correlation with Gleason score–a computer-aided diagnosis development study. Radiology 2013; 267: 787–796.

15.

Borren

Moman

Groenendaal

et al.

Why prostate tumour delineation based on apparent diffusion coefficient is challenging: an exploration of the tissue microanatomy. Acta Oncol 2013; 52: 1629–1636.

16.

Chagas

Babinski

Costa

et al.

Stromal and acinar components of the transition zone in normal and hyperplastic human prostate. Bju Int 2002; 89: 699–702.