Validation of Prostate Tissue Composition by Using Hybrid Multidimensional MRI: Correlation with Histologic Findings

Abstract

Background

Tissue estimates obtained by using microstructure imaging techniques, such as hybrid multidimensional (HM) MRI, may improve prostate cancer diagnosis but require histologic validation.

Purpose

To validate prostate tissue composition measured by using HM MRI, with quantitative histologic evaluation from whole-mount prostatectomy as the reference standard.

Materials and Methods

In this HIPAA-compliant study, from December 2016 to July 2018, prospective participants with biopsy-confirmed prostate cancer underwent 3-T MRI before radical prostatectomy. Axial HM MRI was performed with all combinations of echo times (57, 70, 150, and 200 msec) and b values (0, 150, 750, and 1500 sec/mm²). Data were fitted by using a three-compartment signal model to generate volumes for each tissue component (stroma, epithelium, lumen). Quantitative histologic evaluation was performed to calculate volume fractions for each tissue component for regions of interest corresponding to MRI. Tissue composition measured by using HM MRI and quantitative histologic evaluation were compared (paired t test) and correlated (Pearson correlation coefficient), and agreement (concordance correlation) was assessed. Receiver operating characteristic curve analysis for cancer diagnosis was performed.

Results

Twenty-five participants (mean age, 60 years ± 7 [standard deviation]; 30 cancers and 45 benign regions of interest) were included. Prostate tissue composition measured with HM MRI and quantitative histologic evaluation did not differ (stroma, 45% ± 11 vs 44% ± 11 [P = .23]; epithelium, 31% ± 15 vs 34% ± 15 [P = .08]; and lumen, 24% ± 13 vs 22% ± 11 [P = .80]). Between HM MRI and histologic evaluation, there was excellent correlation (Pearson r: overall, 0.91; stroma, 0.82; epithelium, 0.93; lumen, 0.90 [all P < .05]) and agreement (concordance correlation coefficient: overall, 0.91; stroma, 0.81; epithelium, 0.90; and lumen, 0.87). High areas under the receiver operating characteristic curve obtained with HM MRI (0.96 for epithelium and 0.94 for lumen, P < .001) and histologic evaluation (0.94 for epithelium and 0.88 for lumen, P < .001) were found for differentiation between benign tissue and prostate cancer.

Conclusion

Tissue composition measured by using hybrid multidimensional MRI had excellent correlation with quantitative histologic evaluation as the reference standard.

© RSNA, 2021

Online supplemental material is available for this article.

See also the editorial by Muglia in this issue.

Summary

Tissue composition measured by using hybrid multidimensional MRI demonstrated excellent correlation with quantitative histologic measures.

Key Results

• ■ Prostate tissue composition was similar when assessed with either hybrid multidimensional MRI or quantitative histologic evaluation: stroma (45% vs 44%), epithelium (31% vs 34%), and lumen (24% vs 22%) (all P > .05).

• ■ Agreement was excellent between MRI and histologic assessment of prostate tissue composition (concordance correlation coefficient: overall, 0.91; stroma, 0.81; epithelium, 0.90; lumen, 0.87).

• ■ Diagnostic performance for differentiation of benign tissue and cancer was high with both hybrid multidimensional MRI (area under the receiver operating characteristic curve [AUC], 0.96 for the epithelium and 0.94 for the lumen) and histologic evaluation (AUC, 0.94 for the epithelium and 0.88 for the lumen).

Introduction

Prostate cancer is the most common cancer among men (1). Multiparametric MRI is increasingly being used for prostate cancer diagnosis. MRI offers good soft-tissue contrast and noninvasive sampling of the complete organ, and it is highly effective in determining the presence, size, and aggressiveness of prostate cancer (2–5). However, approximately 15%–30% of clinically significant cancers are missed, even by expert radiologists (6–8), and benign disorders, such as benign prostatic hyperplasia or prostatitis, can mimic cancer (9). In addition, the interpretation of multiparametric MRI scans is subjective and demonstrates only moderate interreader reproducibility (10), which the Prostate Imaging Reporting and Data System has tried to improve. Although contrast on MRI scans heavily depends on tissue microstructure, widely accepted clinical MRI parameters for prostate cancer diagnosis—apparent diffusion coefficient (ADC) and other quantitative values (such as T2 relaxation time [11] and dynamic contrast-enhanced [Tofts model] parameters [12])—provide little information regarding the underlying complex microstructure, which remains the reference standard for cancer detection (13). As such, histologic evaluation remains the reference standard for prostate cancer diagnosis. However, prostate biopsy is an invasive procedure, and only 1% of the prostate is sampled by a standard 12-core needle biopsy. In addition, the aggressiveness of cancer based on the Gleason grading system is generally underestimated with biopsy compared with subsequent results for prostatectomy specimens (3).

Therefore, new microstructure MRI (tissue estimation) techniques, such as luminal water imaging (14), vascular, extracellular, and restricted diffusion for cytometry in tumors (VERDICT) (15), restriction spectrum imaging (16), and hybrid multidimensional (HM) MRI (17), have been developed to improve noninvasive prostate cancer diagnosis (18). Luminal water imaging uses luminal water fraction (the long component in T2 distribution), which is related to underlying tissue structure, for diagnosis. The VERDICT model estimates the vascular, extravascular-extracellular, and cellular spaces used to distinguish cancerous tissue. More recently, studies using ex vivo diffusion relaxation correlation spectrum imaging (19) have characterized the prostate microstructure. Despite the development of new methods, these methods have not been extensively validated in vivo by using direct correlation with prostate histologic findings, specifically the prostate tissue gland components.

HM MRI depicts tissue microstructure, specifically volume fractions of lumen, stroma, and epithelium through coupled T2 and ADC values (20). HM MRI measures the change in ADC and T2 as a function of echo time and diffusion weighting (b value), respectively (21–23), and uses these changes as a source of information about the underlying tissue microstructure. Studies have shown that these volume fractions change with the presence of prostate cancer (24) and its associated Gleason grade (25). Our previous feasibility study (17) showed that prostate tissue composition can be measured noninvasively by using HM MRI and has the potential to help improve prostate cancer diagnosis and determine its aggressiveness. However, that study lacked validation with histopathologic assessment as the reference standard. This validation is a critical step toward "virtual histopathology." Therefore, this study aimed to validate the tissue composition measured by using preoperative HM MRI with quantitative histologic results from whole-mount prostatectomy.

Materials and Methods

Study Participants

The institutional review board approved this prospective Health Insurance Portability and Accountability Act–compliant study. Participants provided informed written consent. Consecutive consenting participants with elevated prostate-specific antigen level and biopsy-confirmed prostate cancer who were scheduled for radical prostatectomy were recruited for this study at our research center between December 2016 and July 2018. Individuals with previous radiation or hormonal replacement therapy (leading to alterations in prostatic signal at MRI) were not eligible.

MRI Scan Acquisition

Participants underwent preoperative 3-T multiparametric MRI (Ingenia; Philips Healthcare) by using a combination of a six-channel cardiac phased-array coil placed around the pelvis and an endorectal coil (Medrad; Bayer Healthcare). The HM MRI sequence consisted of a spin-echo module with diffusion-sensitizing gradients placed symmetrically about the 180° pulse followed by single-shot echo-planar imaging readout. Axial HM MRI images were acquired with all combinations of echo times of 57, 70, 150, and 200 msec and b values of 0, 150, 750, and 1500 sec/mm², which resulted in a 4 × 4 array of data associated with each voxel. A larger parameter space was imaged in this study to improve the fitting of the HM MRI data with more data points compared with our previous work, which used three echo times and three b values (17) Axial HM MRI images were oriented perpendicular to the rectal wall, as guided by sagittal images to align MRI scans more closely with whole-mount histologic slices. Fat saturation was performed using spectral adiabatic inversion recovery. The MRI parameters were as follows: in-plane resolution, 1.5 × 1.5 mm; imaging matrix, 120 × 120; field of view, 180 × 180 mm; repetition time, 5 seconds; number of sections, 18; section thickness, 3 mm; and reconstruction matrix, 128 × 128. The acquisition time was 10–12 minutes. In addition, standard clinical multiparametric MRI sequences, including T2-weighted (axial, coronal, sagittal), diffusion-weighted (axial), and dynamic contrast-enhanced (axial) imaging, were also performed.

Histopathologic Evaluation

Participants subsequently underwent radical prostatectomy, and the excised prostate was fixed in formalin and serially sliced in approximately the same plane as the MRI scans. Whole-mount tissue slices were stained with hematoxylin-eosin to create histologic slides. The slides were evaluated for prostatic adenocarcinoma by an expert pathologist (T.A., 12 years of experience with genitourinary pathology). Areas of tumor were marked on the histologic slides. The slides were scanned at 20× magnification by using a whole-mount digital microscope (Olympus VS120; Olympus) and saved as Olympus Virtual Slide images. Histologic findings and MRI scans were coregistered visually by consensus between an expert radiologist (A.O., 15 years of experience with prostate MRI), a pathologist (T.A.), and a medical physicist (A.C., 8 years of experience with prostate MRI and pathology). Regions of interest (ROIs), including clinically significant cancers (Gleason grade ≥ 3 + 4) and benign tissue, from peripheral and transition zones were marked on histologic images and ADC maps at sites of prostatectomy-verified malignancy and benign tissue and were propagated to all parameter maps.

HM MRI Tissue Composition

The prostate tissue component volumes were calculated by using compartmental analysis of HM MRI data. The HM MRI signals were modeled as unmixed pools of water in three tissue components: stroma, epithelium, and lumen (17) (Appendix E1 [online]). The fractional volumes of tissue component were calculated on a voxel-by-voxel basis by fitting the following equation using the nonlinear least-squares method with an in-house MATLAB (MathWorks) program:

where V_n, T2_n, and ADC_n are the volume fraction, T2, and ADC value for each tissue (stroma, epithelium, and lumen) compartment; S is the signal intensity at each combination of echo time (TE) and b value; and S₀ is the signal intensity at the lowest echo time and b value. Quantitative T2 values (from multiecho data at a b value of 0 sec/mm²) and ADC (diffusion data using multiple b values at an echo time of 57 msec) values were calculated from HM MRI data by using a monoexponential signal decay model.

Figures 1 and 2 show representative examples of prostate tissue composition maps for stroma, epithelium, and lumen along with a combined tissue composition map, predicted cancer map, and corresponding histologic, ADC, and T2 maps. Predicted cancer was any region with conjoint voxels with elevated epithelium (>40%) and reduced lumen (<20%) based on our previous work (17) and meeting the minimum size requirement of 25 mm² on an axial HM MRI section.

Figure 1:

Images in a 65-year-old man with a prostate-specific antigen level of 64 ng/mL (64 μg/L) with confirmed Gleason 4+5 cancer in the right peripheral zone. The representative images include apparent diffusion coefficient (ADC) map; T2 map; tissue composition maps for stroma, epithelium, and lumen; combined tissue composition map (stroma = green, epithelium = red, and lumen = blue); predicted cancer map; and corresponding histologic findings. (Hematoxylin-eosin stain; original magnification ×20.) Cancer was predicted in any voxel with a fractional volume of lumen of less than 20% and a fractional volume of epithelium of more than 40%. The red region of interest in the right peripheral zone is Gleason 4+5, and the green region of interest is benign tissue. The tissue composition for prostate cancer (stroma = 19%, epithelium = 73%, lumen = 8%) and benign transition zone tissue (stroma = 44%, epithelium = 18%, lumen = 38%) was estimated with hybrid multidimensional MRI. The Gleason 4+5 cancer showed elevated epithelium and reduced stroma and lumen volume compared with benign tissue and was correctly identified as cancer by using hybrid multidimensional MRI.

Figure 2:

Images in a 68-year-old man with a prostate-specific antigen level of 7 ng/mL (7 μg/L) with confirmed Gleason 3+4 cancer in the left peripheral zone. The representative images include apparent diffusion coefficient (ADC) map; T2 map; tissue composition maps for stroma, epithelium, and lumen; combined tissue composition map (stroma = green, epithelium = red, and lumen = blue); predicted cancer map; and corresponding histologic findings. (Hematoxylin-eosin stain; original magnification ×20.) The red region of interest in the left peripheral zone highlights the Gleason 3+4 cancer (the benign tissue region of interest was taken on another section [not shown]). The tissue composition for prostate cancer (stroma = 31%, epithelium = 56%, lumen = 13%) was estimated by using hybrid multidimensional MRI. The cancer showed elevated epithelium and reduced stromal and lumen volume compared with benign tissue and correctly identified as cancer by using hybrid multidimensional MRI. The sparse Gleason 3+3 cancer in the right peripheral zone was not predicted by using hybrid multidimensional MRI.

The data collected were independent, and no repeated measurements were taken; only one ROI from each distinct cancer lesion or one ROI representing benign tissue from each prostatic zone was taken for each participant. ROIs for cancer lesions smaller than 5 × 5 mm were not marked and were excluded from further analysis.

Quantitative Histologic Evaluation for Measuring Tissue Composition

Cancerous and benign tissue ROIs from histologic analysis were saved as Tag Image File Format files using ImageJ (National Institutes of Health). Luminal secretions (mucin secretions, crystalloids, and corpora amylacea) were removed from the glandular lumen manually by using Microsoft Paint (Microsoft) because inclusion would affect subsequent segmentation (25). Quantitative histologic evaluation was performed to calculate volumes of tissue components (stroma = green, epithelium = red, lumen = blue) on rectangular ROIs from regions corresponding to the MRI ROIs. Images were segmented semiautomatically by using Image Pro Premier (Media Cybernetics) on the basis of color, intensity, morphologic features, and background with the Smart Segment tool. Segmentation was regarded as successful when segmentation was approved by the consensus of pathologist and medical physicist and the segmented image was visually inspected and judged to have errors of less than approximately 5%. The gland component volumes were defined as percentages of image area and were calculated by using the "Count" tool. Prostate tissue composition measured with this tool has been validated and used in previous studies (25–27). Representative examples of segmentation of cancer and benign tissue are shown in Figures 3 and 4.

Figure 3:

Representative images for quantitative histologic evaluation in which tissue was segmented into stroma (green), epithelium (red), and lumen (blue) for the patient shown in Figure 1. Tissue composition estimated by using hybrid multidimensional (HM) MRI and quantitative histologic evaluation showed good agreement for both prostate cancer (HM MRI vs histologic evaluation: stroma, 19% vs 18%; epithelium, 73% vs 68%; and lumen, 8% vs 14%) and benign tissue (stroma, 44% vs 38%; epithelium, 18% vs 28%; and lumen, 38% vs 34%). H&E = hematoxylin and eosin, ROI = region of interest.

Figure 4:

Representative images for quantitative histologic evaluation in which tissue was segmented into stroma (green), epithelium (red), and lumen (blue) for the patient shown in Figure 2. The tissue composition estimated by using hybrid multidimensional (HM) MRI and quantitative histologic evaluation showed good agreement for both prostate cancer (HM MRI vs histology: stroma, 31% vs 33%; epithelium, 56% vs 54%; and lumen, 13% vs 12%) and benign tissue (stroma, 42% vs 46%; epithelium, 18% vs 15%; and lumen, 40% vs 38%). H&E = hematoxylin and eosin, ROI = region of interest.

Statistical Analysis

The minimum sample size (ROIs) needed for us to claim with evidence of statistical significance that the mean difference in tissue volume fraction measures between HM MRI and quantitative histologic evaluation was no more than 5% (where the variance of tissue measures that is possible is 0%–100%) was determined to be 63 ROIs for a one-sided α of .05 (95% confidence) and statistical power of 0.80. Statistical analysis was performed by using SPSS software, version 24 (IBM). A paired t test was performed to assess the differences between prostate tissue composition: stroma, epithelium, and lumen measured by using HM MRI and reference standard quantitative histologic evaluation. A P value of .05 was used to determine statistical significance. Bland-Altman plots were used to assess the reliability of HM MRI in measuring prostate tissue composition in comparison with reference standard histologic evaluation. Pearson correlation coefficient was calculated between prostate tissue composition measured by using these methods. The Lin concordance correlation coefficient (CCC) was calculated to quantify the agreement between the two measurements, with values above 0.8 considered to indicate excellent agreement; this was used as the primary end point of this study (28). Receiver operating characteristic curve analysis was used to evaluate the performance of fractional volume of tissue components in the differentiation of cancer from benign prostatic tissue.

Results

Participant and Tumor Characteristics

A total of 25 consecutive consenting men participated in this study. All the patients screened were recruited and underwent MRI and subsequent radical prostatectomy, thus meeting all the criteria for inclusion in this study; no patients were excluded. Participant and tumor characteristics are summarized in Table 1. The mean age of these men was 60 years ± 7 (standard deviation). The mean prostate-specific antigen level was 10.5 ng/mL ± 12.6 (10.5 µg/L ± 12.6). The median time between MRI and prostatectomy was 21 days (interquartile range, 14–28 days). A total of 30 tumor ROIs (15 with Gleason grade of 3 + 4, 11 with Gleason grade of 4 + 3, one with Gleason grade of 4 + 4, and three with Gleason grade of 4 + 5) and 45 benign tissue ROIs (22 from the peripheral zone and 23 from the transition zone) were included in the analysis. Because of extensive cancer in the prostatectomy specimens in some cases, benign tissue from certain zones was not included (thus, the number of benign ROIs included in the study is lower than the number of patients). The cancers were primarily located in the peripheral zone (n = 24), followed by the transition zone (n = 6).

Table 1:

Participant and Tumor Characteristics

Tissue Composition at HM MRI and Histologic Evaluation

Overall, the measured prostate tissue composition did not differ between HM MRI and histologic evaluation: stroma, 45% ± 11 versus 44% ± 11 (P = .23); epithelium, 31% ± 15% versus 34% ± 15 (P = .08); and lumen, 24% ± 13 versus 22% ± 11 (P = .80). Figure 5 shows the Bland-Altman plots for prostate tissue composition measured with HM MRI and quantitative histologic evaluation. Minimal bias was well within acceptable limits (within ±5% mean bias and with 95% limits of agreement within ±15%), with no proportional bias on linear regression analysis (stroma, 1% ± 7, with 95% limits of agreement of –12% to 14%; epithelium, –3% ± 5, with 95% limits of agreement of –14% to 7%; lumen, 2% ± 6, with 95% limit of agreement of –9% to 14%) found between tissue composition estimated by using HM MRI and histologic evaluation.

Figure 5:

Bland-Altman plots for prostate tissue composition measured by using hybrid multidimensional (HM) MRI and quantitative histologic evaluation. Minimal bias, which is well within acceptable limits, with no proportional bias on linear regression analysis (stroma: 1% ± 7 [mean ± standard deviation] with 95% limit of agreement of –12% to 14%; epithelium: –3% ± 5 with 95% limit of agreement of –14% to 7%; lumen: 2% ± 6 with 95% limit of agreement of –9% to 14%) was found between the tissue composition estimated by using HM MRI and histologic evaluation. Bold line represents mean bias and dotted lines represent 95% limits of agreement (mean ± 1.96 standard deviation).

Detailed results can be found in Table 2 for cancer and benign tissue, respectively. For cancerous tissue, no difference was found between tissue composition measured with HM MRI and that measured with histologic evaluation. However, when HM MRI was compared with histologic evaluation, in benign tissue, epithelium was underestimated (mean, 21% ± 7 vs 26% ± 6; P = .01), whereas lumen volume was overestimated (mean, 32% ± 11 vs 28% ± 10; P = .01).

Table 2:

Comparison of Prostate Tissue Composition Measured by Using Hybrid Multidimensional MRI and Quantitative Histologic Evaluation

Correlative Analyses between HM MRI and Histologic Evaluation

The Pearson correlation coefficient was excellent for prostate tissue composition measured by using these methods (overall, 0.91; stroma, 0.82; epithelium, 0.93; lumen, 0.90; P < .05) (Fig 6). Most important, there was excellent agreement between histologic evaluation and HM MRI based on the Lin CCC, with a CCC of 0.91 (95% CI: 0.87, 0.94). Individual CCC was higher than the excellent agreement threshold of 0.8 in each of three prostatic components: stroma, 0.81 (95% CI: 0.64, 0.91); epithelium, 0.90 (95% CI: 0.82, 0.95); and lumen, 0.87 (95% CI: 0.76, 0.93).

Figure 6:

Correlation plot shows excellent Pearson correlation (overall, 0.91; stroma, 0.82; epithelium, 0.93; and lumen, 0.90; all P < .05) and Lin concordance correlation coefficient (CCC) (overall, 0.91; stroma, 0.81; epithelium, 0.90; lumen, 0.87; all P < .05) between prostate tissue composition measured by using hybrid multidimensional (HM) MRI and quantitative histologic evaluation.

Differences were found between epithelial and luminal volumes in cancerous versus benign tissue, but this was not the case for stromal volume. Prostate cancers were characterized by increased epithelium compared with benign prostatic tissue at both HM MRI and histologic evaluation (HM MRI: 44% ± 12 vs 21% ± 7; histologic evaluation: 46% ± 12 vs 26% ± 6; both P < .05) and decreased lumen (HM MRI: 14% ± 6 vs 32% ± 11; histologic evaluation: 14% ± 6 vs 28% ± 10; both P < .05). The area under the receiver operating characteristic curve values for distinguishing between prostate cancer and benign prostate tissue was high for both HM MRI (epithelium, 0.96; lumen, 0.94; both P < .001) and histologic evaluation (epithelium, 0.94; lumen, 0.88; both P < .001) measurements. Detailed analysis is provided in Table 3.

Table 3:

Receiver Operating Characteristic Curve Analysis for Differentiation of Prostate Cancer from Benign Tissue

Discussion

Tissue estimates obtained by using microstructure imaging techniques, such as hybrid multidimensional (HM) MRI, may improve prostate cancer diagnosis but require histologic validation. We demonstrated that tissue composition measured noninvasively with HM MRI matched closely with the reference standard of quantitative histologic evaluation. Prostate tissue composition did not significantly differ when evaluated with HM MRI or quantitative histologic evaluation: stroma, 45% versus 44% (P = .23); epithelium, 31% versus 34% (P = .08); and lumen, 24% versus 22% (P = .08). Of note, correlation and agreement between prostate tissue composition measured using these methods were excellent. The overall Pearson correlation coefficient was greater than 0.9. The overall Lin concordance correlation coefficient was 0.91, which was well above the excellent agreement threshold of 0.8.

We used the CCC as the primary end point to provide a quantitative measure of reliability in addition to the Pearson correlation coefficient; this is in contrast to other studies trying to validate tissue measures from new MRI microstructure imaging techniques with pathologic evaluation, which used only Pearson or Spearman rank correlation (19,28,29). Whereas the Pearson correlation coefficient simply provides a measure of linear covariation between two sets of measurements, CCC measures the degree of correspondence between the two sets of values based on covariation and correspondence. In addition, Bland-Altman analysis showed minimum bias, suggesting good accuracy, and the low standard error (<7.5%) suggests good precision. Together, these results indicate that HM MRI should be investigated further for noninvasive prostate cancer diagnosis.

Prostate cancers were characterized by increased epithelium and decreased lumen. This trend, as well as the fractional volumes of tissue components (stroma, epithelium, and lumen), agree with the findings of our previous study (17) and the findings of previous quantitative histologic studies (24,25,30). Receiver operating characteristic analysis showed a high area under the curve for fractional volumes of the tissue components measured with both HM MRI and histologic evaluation, suggesting that prostate cancer can be differentiated from benign tissue by increased epithelium and reduced luminal volume in prostate cancer compared with benign tissue. The area under the curve for differentiating between benign tissue and prostate cancer was highest for the epithelium (range, 0.94–0.96), followed by the lumen volume (range, 0.88–0.94). Therefore, both measurements have excellent potential as diagnostic markers. However, we found no significant difference in stromal volume between cancer and benign tissue, unlike a previous study that reported reduced stromal volume in cancer (17). However, stroma volume in cancer was nominally lower than benign tissue in our study, which was also reported in another histologic study (24).

Other prostate microstructure imaging studies have demonstrated correlation of MRI markers with histologic findings. In 17 patients in whom luminal water imaging was performed in vivo, moderate correlation (ρ = 0.75) was found between luminal water fraction and histologic lumen fractional volume (28). Luminal water fraction was lower in benign peripheral zone tissue (27.2%) than in cancer tissue (11.6%), a finding that matches the results in this study. However, a possible drawback of luminal water imaging is the inability to differentiate between epithelium and stroma, which have similar T2 values. These similar T2 and luminal water fraction values may lead to misclassification as cancers of regions with low luminal water fraction, such as anterior fibromuscular stroma (luminal water fraction, 4.8%) and transition zone, including benign prostatic hyperplasia (mean luminal water fraction range, 14%–16%) (14,31), particular in stromal benign prostatic hyperplasia (32).

VERDICT measurements of intracellular (r = 0.96), extracellular-extravascular (r = 0.96), and vascular (r = 0.39) volume fraction in three patients in vivo showed moderate to strong correlation (29). Similar analysis on the excised fresh and fixed prostate showed that the intracellular volume fraction estimated with VERDICT correlated with histologic indicators of cellularity (33).

Restriction spectrum imaging (16) has been validated in 10 patients; higher mean cellularity detected with restriction spectrum imaging correlated with the presence of cancer (1.81 in cancer vs –0.32 in benign tissue) and increasing Gleason grade. Cellularity evaluated visually at histologic evaluation correlated with restriction spectrum imaging cellularity index. However, previous work has shown that volume fractions of stroma, epithelium, and lumen are better predictors of prostate cancer and MRI signal changes than is cellularity (25).

Like HM MRI, diffusion-relaxation correlation spectrum imaging (19) helps predict volume fractions of stroma, lumen, and epithelium; it has been reported to have moderate correlation with measurements of stroma (ρ = 0.32), epithelium (ρ = 0.80), and lumen (ρ = 0.57) at histologic evaluation. In that study, diffusion-relaxation correlation spectrum imaging validation was limited to nine ex vivo samples and has not been evaluated in vivo (19). Therefore, more detailed histologic correlation is needed for these new techniques. Our study used a larger data set than the ones used in the studies cited previously. In addition, the correlation between tissue composition measured with HM MRI and histologic evaluation as reported in our study was stronger than the correlation reported for other MRI methods measuring tissue microstructure.

Our study had a few limitations. The tissue composition measured in vivo by using HM MRI was compared with measurements from fixed prostate tissue ex vivo. Formalin fixation reduces the size of the prostate compared with the in vivo prostate (approximately 15% reduction in volume) (34). However, the effect of fixation on volume fractions of individual gland components has not been studied. During surgery, the prostatic fluid leaks out, which could reduce the lumen volume (collapsed lumen) measured at histologic evaluation compared with in vivo HM MRI measurements, as suggested by our results. This effect was more pronounced in benign tissue, as lumen volume was overestimated with HM MRI. Nonetheless, we expect these effects to be minimal (<5%), as shown by our results.

Benign features, such as inflammation, benign prostatic hyperplasia, inflammation, and prostatitis that can mimic cancer (9,35), were not specifically included in the analysis as separate groups. However, benign prostatic hyperplasia regions were included as part of benign transition zone regions because they are more representative of the tissue in this zone. Nonetheless, it would be desirable to test the validity of HM MRI in measuring tissue composition specifically in these benign features in future studies.

Quantitative histologic evaluation was performed semiautomatically. This included a subjective interpretation, in which segmentation was regarded as successful when erroneous segmentation was visually estimated to be less than approximately 5% of the image area. Comparison with results obtained with the same software with manual segmentation in an article suggested that any resultant bias was minor (25).

Another limitation of our model was that estimation of other histologic features (eg, inflammation and blood vessels) that are not prominent features in prostate tissue (typically <5%) were not part of MRI modeling or histologic analysis. Similar validation using different MRI vendors and without the use of an endorectal coil is also needed. Additional studies to assess the reproducibility of tissue estimates by using HM MRI are also needed.

In conclusion, this study has demonstrated that tissue composition measured with hybrid multidimensional MRI correlates with reference standard quantitative histologic measures. There was excellent agreement between prostate tissue composition measured with these methods. In addition, the high area under the receiver operating characteristic curve suggests that prostate cancer may be differentiated from benign tissue through use of hybrid multidimensional MRI.