Prostate cancer vs. post-biopsy hemorrhage: Diagnosis with T2- and diffusion-weighted imaging

Authors


  • This work was presented at the 17th annual meeting of the International Society for Magnetic Resonance Imaging, April 2009, Honolulu, Hawaii.

Abstract

Purpose:

To assess the value of quantitative T2 signal intensity (SI) and apparent diffusion coefficient (ADC) to differentiate prostate cancer from post-biopsy hemorrhage, using prostatectomy as the reference.

Materials and Methods:

Forty-five men with prostate cancer underwent prostate magnetic resonance imaging (MRI), including axial T1-weighted imaging (T1WI), T2WI, and single-shot echo-planar image (SS EPI) diffusion-weighted imaging. Two observers measured, in consensus, normalized T2 signal intensity (SI) (nT2, relative to muscle T2 SI), ADC, and normalized ADC (nADC, relative to urine ADC) on peripheral zone (PZ) tumors, benign PZ hemorrhage, and non-hemorrhagic benign PZ. Tumor maps from prostatectomy were used as the reference. Mixed model analysis of variance was performed to compare parameters among the three tissue classes, and Pearson's correlation coefficient was utilized to assess correlation between parameters and tumor size and Gleason score. Receiver-operating characteristic (ROC)-curve analysis was used to determine the performance of nT2, ADC, and nADC for diagnosis of prostate cancer.

Results:

nT2, ADC, and nADC were significantly lower in tumor compared with hemorrhagic and non-hemorrhagic benign PZ (P < 0.0001). There was a weak but significant correlation between ADC and Gleason score (r = −0.30, P = 0.0119), and between ADC and tumor size (r = −0.40, P = 0.0027), whereas there was no correlation between nT2 and Gleason score and tumor size. The areas under the curve to distinguish tumor from benign hemorrhagic and non-hemorrhagic PZ were 0.97, 0.96, and 0.933 for nT2, ADC, and nADC, respectively.

Conclusion:

Quantitative T2 SI and ADC/nADC values may be used to reliably distinguish prostate cancer from post-biopsy hemorrhage. J. Magn. Reson. Imaging 2010;31:1387–1394. © 2010 Wiley-Liss, Inc.

MAGNETIC RESONANCE IMAGING (MRI) is often used for staging of prostate cancer in order to assess for extra prostatic disease, which traditionally precludes radical prostatectomy given an increased likelihood of positive surgical margins and disease recurrence (1–3). However, given the emergence of an array of focal therapies for prostate cancer, such as high-intensity focused ultrasound (4), cryoablation (5), and photodynamic therapy (6), it is increasingly important to accurately localize tumor. Conventional T2-weighted imaging (T2WI) is of limited accuracy for tumor localization (7), partly because the vast majority of prostate MR examinations are performed after the patient has previously undergone transrectal ultrasound-guided biopsy to establish the diagnosis. Subsequent post-biopsy hemorrhage presents a diagnostic challenge for accurate tumor detection with MRI (8), as low T2 signal from hemorrhage may cause an overestimation of tumor presence in the initial post-biopsy period (9). Although several investigators have recommended delaying MRI after the initial biopsy by anywhere from 3 to 8 weeks to allow time for resolution of hemorrhage, such a delay may not achieve a full resolution of blood products and may lead to a delay in therapy and increased patient anxiety (10, 11).

Quantitative T2 signal intensity (SI) values (12, 13) and apparent diffusion coefficient (ADC) values measured with diffusion-weighted imaging (DWI) (14–20) have been applied for prostate cancer detection in several studies. However, to our knowledge, their utility in identifying tumor in the presence of hemorrhage has not been specifically assessed. Although one recent study showed improved accuracy for detecting prostate cancer in the presence of hemorrhage when using dynamic contrast-enhanced and DWI, the investigators applied DWI in a qualitative fashion without measurement of ADC values and did not use correlation with step-section prostatectomy (20). Although prostate cancer and hemorrhage may demonstrate a similar subjective visual appearance on T2WI and DWI, we hypothesize that quantitative measures may help in their differentiation.

The purpose of our study was to assess the diagnostic value of quantitative T2 SI and ADC values to differentiate prostate cancer from post-biopsy hemorrhage, using step-section prostatectomy as the reference standard.

PATIENTS AND METHODS

Patients

This retrospective study complied with the Health Information Portability Accountability Act and was approved by our institutional review board, with waiver of the requirement for written informed consent. We assessed 45 consecutive men (mean age 60 ± 7.7 years, range 46–80 years) with biopsy-proven diagnosis of prostate cancer who had undergone prostate MRI on the same scanner prior to radical prostatectomy. Patients were included regardless of the delay between biopsy and MRI. The mean interval between MRI and prostatectomy was 20 ± 27 days (range 1–144 days), and the mean interval between biopsy and MRI was 63 ± 45 days (range 10–241 days). No patient received hormonal or radiation therapy for their prostate cancer. The mean preoperative prostate-specific antigen (PSA) was 7.4 ± 10.7 ng/mL (range 0.6–69.4 ng/mL). Data from 38 of the patients in this study were also included in a previous report from this institution (21).

MRI

All patients underwent prostate MRI on a single 1.5-T clinical system (Magnetom Avanto; Siemens Healthcare, Malvern, PA). A six-element pelvic surface phased-array coil was used. The sequences included: axial turbo-spin–echo (TSE) T2WI through the prostate and seminal vesicles [TR/TE 4000/101, flip angle (FA) 160°, slice thickness 3 mm, no interslice gap, no fat suppression, field of view (FOV) 150 × 150 mm, matrix 192 × 123, no parallel imaging, three signal averages, acquisition time 4:32 minutes]; axial TSE T1-weighted imaging (T1WI) through the prostate and seminal vesicles using identical slice location as the T2WI [TR/TE 428/12, FA 160°, slice thickness 3 mm, no interslice gap, no fat suppression, FOV 150 × 150 mm, matrix 192 × 123, generalized autocalibrated partially parallel acquisition (GRAPPA) with parallel imaging factor of 2, two signal averages, acquisition time 2:17 minutes]; axial single-shot echo-planar image DWI through the pelvis [TR/TE 3300/84, FA 90°, motion-probing gradients applied in three orthogonal directions with b-values of 50–500–1000 sec/mm2 with in-line reconstruction of trace ADC map, slice thickness 6 mm, FOV 350 × 240 mm, matrix 192 × 132, fat suppression using spectral attenuated inversion recovery (SPAIR), GRAPPA with parallel imaging factor of 2, four signal averages, acquisition time 2:09 minutes].

Other sequences performed as part of the routine prostate MR examination protocol at our institution but not assessed in this study include sagittal and coronal TSE T2WI sequences through the prostate and seminal vesicles, axial TSE T1-weighted sequence through the entire pelvis, and dynamic contrast-enhanced sequences following intravenous administration of gadolinium-based contrast.

Image Interpretation

Two observers [B.T. (observer 1) and A.B.R. (observer 2), with 5 years and 1 year of experience, respectively, in interpretation of prostate MRI] reviewed the MR images for all cases, in consensus, using a commercial workstation (Leonardo; Siemens Healthcare). For each case, histologically defined tumor maps (see later) were used to define three tissue classes: peripheral zone (PZ) tumor; benign PZ hemorrhage; and benign non-hemorrhagic PZ. First, for each tumor focus depicted on the tumor maps, the observers identified an area of low SI located within a similar region of the same sextant on T2WI, which was considered to represent tumor on imaging. No tumor focus was recorded if no area of T2 hypointensity was identified in the sextant that contained tumor on the tumor maps. This same approach was also used to define tumor foci as areas of visually decreased ADC value on the ADC maps (6). Next, foci of benign hemorrhage were identified as foci of high signal on T1WI (9). Specifically, foci of high T1 SI within the PZ measuring at least 1 cm2 that did not overlap with a region previously regarded as tumor were considered to represent benign hemorrhage (up to three foci of benign hemorrhage were recorded per patient). For each area of benign hemorrhage identified on T1WI, a region of interest (ROI) was copied and pasted at the approximate corresponding slice location on T2WI (see later). Finally, one area of benign non-hemorrhagic PZ was identified for each case. For this purpose, a region measuring at least 1 cm2 that demonstrated homogeneous low T1 signal, homogeneous high T2 signal, and homogeneous high ADC, based on visual assessment of these sequences, and that did not correspond with tumor on the tumor maps, was selected. The central gland was not assessed.

For each selected focus of tumor, benign PZ hemorrhage, and benign non-hemorrhagic PZ, ROIs were placed over the focus on T2WI and ADC maps, with the mean T2 SI and mean ADC values recorded. For foci of tumor and hemorrhage, an ROI as large as possible was manually traced along the margin of tumor and area of hemorrhage, respectively. Although the FOV for DWI was larger than the FOV for T2WI, the images were viewed side-by-side during ROI placement to ensure that the ROIs corresponded to comparable areas on the two sequences. In addition, for each case, the T2 SI of the right obturator internus muscle was obtained using a 2-cm2 circular ROI, and the ADC of the urine within the bladder was obtained using a 5-cm2 ROI. Normalized T2 SI-to-muscle ratio (nT2) and normalized ADC-to-bladder ratio (nADC) were then calculated for all T2 SI and ADC measurements. As T2 SI values are in arbitrary units and may vary based on scanner and acquisition parameters, we calculated ratios of prostate T2 SI to muscle to attempt to normalize the T2 SI of the various tissue classes, as has been done by previous investigators (12, 13). Although ADC values are obtained in units of millimeters squared per second and in theory should be more reproducible, recent studies indicate substantial variability in ADC values as well (22, 23), likely due to a combination of technical and patient-related factors. Therefore, we attempted normalization of ADC values as well, using the urinary bladder as reference given the low SI of the muscle on DWI. Mean nT2, ADC, and nADC were calculated for tumor foci, foci of benign PZ hemorrhage, and foci of benign non-hemorrhagic PZ.

An additional analysis was performed to assess the frequency at which hemorrhage was observed within tumor foci. For each tumor focus, the observers reviewed T1WI and T2WI images side-by-side and recorded whether any T1 hyperintensity was present within the precise region defined to represent the tumor itself. T1 hyperintensity along the margin of the tumor was not included as a positive finding, given the possibility that such T1 hyperintensity may represent coexistent hemorrhage adjacent to and outlining the tumor, rather than representing hemorrhage intermixed with the substance of the tumor itself.

Reference Standard: Pathologic Analysis

All prostatectomy slides for each case were reviewed by two experienced pathologists in consensus (J.M. and X.K., with 19 and 11 years of experience, respectively, in prostate pathology). Whole-mount slides were available in 6 cases and standard step-section slides were prepared for the remaining cases. The two pathologists used these slides to generate diagrams depicting the location of tumor foci within the PZ; these diagrams were used as tumor maps during the retrospective review of MRI findings in the patient cohort, as described earlier. The largest focus of PZ tumor within each sextant (right and left base, right and left mid-gland, right and left apex) was depicted, along with the tumor size and Gleason score. The pathologists were unaware of the MRI findings.

Statistical Analysis

SAS version 9.0 (SAS Institute, Cary, NC) was used for all statistical computations.

Tissue classes were compared with respect to each of the three imaging measures (ADC, nADC, nT2) using mixed model analysis of variance based on ranks. The analysis was non-parametric and able to account for statistical dependencies among the multiple observations derived for a single patient. For each analysis of variance (ANOVA), the covariance structure was modeled by assuming observations to be correlated only when provided by the same patient and by allowing the error variance to differ across tissue classes. ANOVA was first conducted using all available data and with tissues classified as tumor, benign with hemorrhage, and benign without hemorrhage. The P-values from the ANOVA are reported with Tukey's multiple comparison correction. Spearman's rank correlation coefficients were used to characterize the association of the imaging measures derived for tumors with the Gleason score and pathologically determined tumor size. P-values for the associations were derived from a mixed model ANOVA based on ranks. The analysis followed the same outline as described earlier, except that tissue class was removed from the model and the model included tumor size or Gleason score as a numeric factor. Logistic regression for correlated data was used to derive an exact 95% confidence interval (CI) for the percentage of tumors that can be expected to show signs of hemorrhage using generalized estimating equations based on a binary logistic regression model and assuming observations to be correlated only when acquired from the same patient. ROC analyses were used to assess the diagnostic utility of ADC, nADC, and nT2 for the discrimination of tumors from benign tissue either with or without hemorrhage. In addition, Pearson's correlation coefficients were used to characterize the association of nT2, ADC, and nADC with delay between biopsy and MRI for all three tissue classes. For this analysis, a single mean nT2, ADC, and nADC was calculated for each tumor class for each patient. Finally, Spearman's correlation coefficient was used to characterize the association of tumor size and Gleason score for tumor foci. All reported P-values are two-sided and were declared statistically significant at P < 0.05.

RESULTS

Pathologic Findings

In the 45 patients, 107 foci of prostate cancer were marked on the tumor maps. These 107 tumor foci had a mean Gleason score of 6.8 (range 6–9) and mean size of 10 mm (range 3–26 mm). The T-stages for the 45 patients based on prostatectomy were as follows: T2a (N = 5); T2b (N = 5); T2c (N = 20); T3a (N = 11); and T3b (N = 4). All patients had a nodal stage of N0, except for 1 patient with T1c disease who had a nodal stage of N1.

MRI Findings

Of the 107 tumor foci diagnosed at prostatectomy, 86 (80.3%) were visible on T2WI as a region of hypointensity on which an ROI could be measured, and 73 (68.2%) were identifiable on the ADC map as a region of decreased ADC on which an ROI could be measured. In addition, 40 areas of benign hemorrhage were identified for analysis. While a corresponding ROI was placed on T2WI for all 40 (100%) of these hemorrhagic foci, it was possible to measure a corresponding ADC value on the ADC map for only 33 (83%) of these 40 foci due to differences in slice positioning. nT2, ADC, and nADC all demonstrated the largest values for benign non-hemorrhagic PZ, the smallest values for foci of tumor, and intermediate values for benign PZ hemorrhage (Table 1 and Fig. 1). The three measures all showed significant differences between all possible combinations of the three tissue classes (benign non-hemorrhagic PZ, benign PZ hemorrhage, and prostate cancer) (Table 2 and Figs. 2, 3, and 4).

Figure 1.

Boxplot distribution of nT2 (a), ADC (b), and nADC (c) for benign PZ, PZ hemorrhage, and PZ tumor. Asterisk represents outliers. Top and bottom of boxes represent 25–75th percentiles of data values. Line in box represents median value.

Figure 2.

A 62-year-old man with prostate cancer (Gleason score 7). Axial TSE T2-weighted image (a), axial TSE T1-weighted image (b), and axial ADC map (c) [obtained with single-shot echo-planar imaging diffusion sequence (SS EPI) using b-values of 50–500–1000 sec/mm2]. Histologically defined peripheral zone tumor at the left base shows low T2 signal intensity (arrow) (a). Tumor is spared from surrounding hemorrhage on T1-weighted image (arrow) (b), and shows low ADC (arrow) (c). Measured parameters in PZ tumor were as follows: nT2 = 4.05; ADC = 0.79 × 10−3 mm2/sec; and nADC = 0.32.

Figure 3.

A 46-year-old man with prostate cancer (Gleason score 7). Axial TSE T1-weighted image (a), axial TSE T2-weighted image (b), and axial ADC map (c) (obtained with SS EPI diffusion sequence using b-values of 50–500–1000 sec/mm2). There is a small area of post-biopsy hemorrhage at the left mid-gland (arrow) (a) that did not correspond to tumor on prostatectomy. This area demonstrates low T2 signal (arrow) (b) and low ADC (arrow) (c). Measured parameters in hemorrhagic focus were as follows: nT2 = 5.33; ADC = 1.11 × 10−3 mm2/sec; and nADC = 0.45—values higher than those measured in the patient presented in Fig. 2.

Figure 4.

A 63-year-old man with prostate cancer (Gleason score 8). Axial TSE T2-weighted image (a), axial TSE T1-weighted image (b), and axial ADC map (c) (obtained with SS EPI diffusion sequence using b-values of 50–500–1000 s/mm2). An area within the peripheral zone at the mid-line of mid-gland that corresponds with tumor on prostatectomy demonstrates low T2 signal (arrow) (a). The tumor contains significant internal hemorrhage as seen on T1-weighted image (arrow) (b) and shows a low ADC value (arrow) (c). Measured parameters were as follows: nT2 = 3.43; ADC = 0.98 × 10−3 mm2/sec; and nADC = 0.48. For comparison, benign hemorrhage in the left base of the same patient (not shown) demonstrates larger values for these parameters: nT2 = 8.04; ADC = 2.07 × 10−3 mm2/sec; and nADC = 1.01.

Table 1. Mean and SD of nT2, ADC, and nADC within each Tissue Class in 45 Patients with Prostate Cancera
TissueNumber of focinT2ADCnADC
  • nT2: normalized T2 signal intensity (SI) = SI of tissue/SI obturator muscle; ADC (apparent diffusion coefficient): × 10−3 mm2/sec; nADC: normalized ADC = ADC of tissue/ADC urine; PZ: peripheral zone.

  • a

    See Table 2 for P-values for each comparison.

  • *

    Number of foci for which nT2 was measurable.

  • **

    Number of foci for which ADC and nADC were measurable.

PZ tumor86*/73**4.16 ± 1.251.02 ± 0.220.41 ± 0.12
Benign PZ with hemorrhage40*/33**7.42 ± 2.101.52 ± 0.300.61 ± 0.15
Benign PZ with no hemorrhage45*/45**9.12 ± 2.011.74 ± 0.220.71 ± 0.13
Table 2. Tukey-Corrected P-Values from ANOVA Test to Compare Tissue Classes with Respect to each Imaging Measure
Tissues comparednT2ADCnADC
  1. ANOVA, analysis of variance; PZ, peripheral zone.

PZ tumorBenign PZ with hemorrhage<0.0001<0.0001<0.0001
PZ tumorBenign PZ with no hemorrhage<0.0001<0.0001<0.0001
Benign PZ with no hemorrhageBenign PZ with hemorrhage0.0056<0.0001<0.0001

ROC Analysis of nT2, ADC, and nADC for Diagnosis of Tumor

Receiver-operating characteristic (ROC) curve analysis for the diagnosis of tumor, when including all tissue classes, revealed no significant difference in the areas under the curve (AUCs) for nT2, ADC, and nADC (0.97, 0.93, and 0.93, respectively, P > 0.15) (Fig. 5). ROC-curve analysis was also used to identify the threshold value for each of the measures to optimize accuracy (Table 3). The determined threshold values with corresponding sensitivity and specificity were as follows: nT2, 5.75 (93% and 88.2%); ADC, 1.39 × 10−3 mm2/sec (97.3% and 82.1%); and nADC, 0.49 (86.3% and 89.7%).

Figure 5.

ROC curves to assess the diagnostic utility of ADC (black, solid), nADC (red, dashed), and nT2 (blue, dotted) for the discrimination of tumors from benign tissue (either with or without hemorrhage). The AUC was 0.97 for nT2, 0.96 for ADC, and 0.93 for nADC, with no significant differences among the measures in terms of AUC (P > 0.15).

Table 3. ROC-Curve Analysis of nT2, ADC, and nADC for Discriminating Prostate Cancer in PZ from Post-Biopsy PZ Hemorrhage and Benign Non-Hemorrhagic PZ
ParameterCriterionAUCSensitivitySpecificity*Positive predictive valueNegative predictive value
  • ROC, receiver-operating characteristic; PZ, peripheral zone; AUC, area under the curve.

  • *

    Reflects all benign PZ foci (hemorrhagic and non-hemorrhagic).

nT2≤5.750.9793.0% (80/86)88.2% (75/85)88.9% (80/90)92.6% (75/81)
ADC≤1.390.9697.3% (71/73)82.1% (64/78)83.5% (71/85)97.0% (64/66)
nADC≤0.490.9386.3% (63/73)89.7% (70/78)88.7% (63/71)87.5% (70/80)

Correlation of nT2, ADC, and nADC Values with Tumor Gleason Score, Tumor Size, and Delay between Biopsy and MRI

There was a weak, but statistically significant, correlation between each of ADC and nADC with tumor Gleason score (r = −0.30, P = 0.0119; and r = −0.27, P = 0.0315, respectively) and between each of ADC and nADC with tumor size (r = −0.40, P = 0.0027; and r = −0.34, P = 0.0209, respectively) (Table 4). The correlation was not statistically significant between nT2 and Gleason score (r = −0.13, P = 0.2293) or between nT2 and tumor size (r = −0.16, P = 0.2004). In addition, there was a moderate correlation that reached statistical significance between tumor size and Gleason score (r = 0.57, P < 0.0001).

The correlation between delay between biopsy and MRI and each of the three parameters (nT2, ADC, and nADC) was not statistically significant for any of the three tissue classes (P-values from 0.355 to 0.802 for benign non-hemorrhagic foci, from 0.136 to 0.159 for benign hemorrhage, and from 0.347 to 0.992 for tumor).

Table 4. Spearman Rank Correlation to Evaluate Association of each Imaging Measure with Tumor Grade and Size
MeasureGleason scoreTumor size
CorrelationPCorrelationP
  • a

    Statistically significant (P < 0.05).

nT2−0.130.2293−0.160.2004
ADC−0.300.0119a−0.400.0027a
nADC−0.270.0315a−0.340.0209a

Association between Tumor and Hemorrhage

Five of the 107 prostate cancer foci (4.6%) had internal hemorrhage. A 95% confidence interval for the prevalence of hemorrhage among tumor foci extends from 2.0% to 10.5% and implies that there is 95% confidence that no more than 10.5% of tumor foci will have hemorrhage.

DISCUSSION

In our study, we observed significant differences in quantitative T2 SI and ADC/nADC values between prostate cancer foci, benign post-biopsy PZ hemorrhage, and benign non-hemorrhagic PZ, with tumor demonstrating significantly lower values than the other two classes. It was possible to separate tumor from benign hemorrhagic and non-hemorrhagic PZ foci with an AUC of >0.95 either using ADC or using T2 SI relative to muscle. We also observed a very low rate of hemorrhage within tumor foci.

The presence of hemorrhage poses a major challenge in attempting to detect prostate cancer using MRI, as blood products can appear T2 hypointense, thereby obscuring the presence of tumor (8–10). Various functional MR techniques that have been applied for prostate imaging are also subject to limitation as a result of biopsy and subsequent hemorrhage. For instance, when performing MR spectroscopy of the prostate, the number of voxels showing spectral degradation was found to vary inversely with the delay after biopsy, despite a lack of correlation with a subjective qualitative assessment of the degree of hemorrhage (11). Furthermore, a study of the role of dynamic contrast enhancement in prostate cancer detection observed areas of hemorrhage to show abnormal enhancement, attributed to formation of granulation tissue, thereby posing a diagnostic challenge (7). Finally, although a recent study found that the use of DWI increased the accuracy of prostate cancer detection in patients with hemorrhage, an overall poor sensitivity for prostate cancer was attributed to susceptibility artifact and lowering of ADC of benign PZ resulting from hemorrhage (20). Therefore, to date, the most feasible approach has been to simply delay prostate MRI following biopsy to allow time for resolution of blood products. However, this option may be suboptimal for patients and referrers wishing to proceed with cancer staging and therapy and offers no solution for cases in which hemorrhage persists despite the delay. Another alternative may be to perform MRI prior to biopsy, although this practice is not widely accepted and currently performed in a very limited number of centers outside of the USA (24).

Our results suggest a practical and easy way to differentiate tumor from benign hemorrhage as a cause of T2 hypointensity or low ADC via the use of quantitative T2 SI and ADC/nADC values, respectively, and may impact prostate MRI in the clinical setting in several ways. First, the results illustrate the value of interpreting prostate MRI in a quantitative fashion, rather than performing a sole subjective visual assessment. By applying specific numerical thresholds, a high degree of accuracy in distinguishing prostate cancer from hemorrhage can be obtained. Second, our results would support not requiring a long delay between prostate biopsy and MRI to allow time for resorption of hemorrhage, given that hemorrhage may be reliably distinguished from tumor. Finally, an increased ability to distinguish prostate cancer and hemorrhage should assist in precisely localizing tumor within the gland and facilitate the ability of prostate MRI to guide focal therapies.

The observed decreases in T2 SI and ADC values of prostate cancer relate to its histologic nature, as prostate cancer consists of tightly packed glandular elements with increased cellularity and decreased fluid accumulation (2, 20). Such histologic features may account for the observed decreases in both T2 SI and ADC values, although these parameters reflect distinct tissue properties (25). It seems reasonable that, as prostate hemorrhage consists essentially of benign tissue with intermixed blood products, the quantitative measurements would not change as drastically from that of benign PZ, as seen with prostate cancer.

Normalization of ADC has been rarely used outside the brain. Recent studies in the pelvis have observed a larger difference between benign and malignant lymph nodes in uterine and cervical cancers using a normalized ADC rather than absolute ADC (26), as well as a higher AUC for differentiating metastatic from non-metastatic lymph nodes in cervical cancer using a normalized ADC (27). Despite the added value of using a normalized ADC in these previous studies, we observed no benefit of using nADC in the prostate in our study, as there was no significant difference between ADC and nADC in terms of AUC for differentiating benign from malignant PZ.

In addition, our results may seem to differ from those of Kaji et al (8), who observed T2 hypointensity within 80% of foci of hemorrhage. However, their evaluation was based entirely on a subjective visual assessment, without distinguishing between the degree of T2 hypointensity of prostate cancer and hemorrhage in either a qualitative or quantitative fashion (8). Although we indeed showed a slight decrease, on average, in the T2 SI of hemorrhage, as compared with normal PZ, it is only by a quantitative assessment that the degree of T2 hypointensity between hemorrhage and prostate cancer becomes apparent. The optimal threshold value that we determined for distinguishing benign from malignant PZ using nT2 of 5.75 is close to the threshold of 4 identified by Engelhard et al (12) to distinguish PZ tumor from benign conditions such as chronic prostatitis, fibrosis, or atrophy. Finally, the correlation between nT2, ADC, and nADC for hemorrhagic foci with the delay between biopsy and MRI did not reach statistical significance for any of these measures, indicating that a partial normalization of T2 SI and ADC values within hemorrhage that may occur over time following biopsy would not have accounted for our findings.

We observed a weak, but statistically significant, correlation of ADC and nADC with Gleason score. The correlations we observed (−0.30 and −0.27, respectively) are intermediate between the correlations between ADC and tumor grade in two earlier studies of −0.18 and −0.497 (20, 28). Zelhof et al (29) found a significant negative correlation between ADC and cellular density in the prostate, attributed to the decreased interstitial space that occurs in the setting of more compact malignant tissue, potentially providing an explanation for the relationship between ADC and tumor grade. This correlation of ADC with tumor grade further supports the role of a quantitative assessment in predicting the biologic aggressiveness of detected tumor, which in turn impacts patient prognosis and management. The additional correlation of ADC and nADC with tumor size is not surprising given the interdependence between the size and Gleason score of tumor foci that was observed both in this study as well as in previous work (30). Although we do not have a strong explanation for the lack of a significant correlation between nT2 and tumor grade and size, we do note that such a correlation has not been uniformly observed in previous investigations, as Engelbrecht et al (31) failed to observe a correlation between T2 relaxation rates and tumor Gleason score (r = −0.08) and size (r = 0.05).

We found that 5% of the 107 tumor foci contained hemorrhage intermixed within tumor, indicating that, with 95% confidence, fewer than 10.5% of tumor foci contain hemorrhage. Although it has been stated previously that prostate cancer has a resistance to being penetrated by hemorrhage (2, 15), the actual rate of association of these two processes has not previously been well documented. For instance, although Tamada et al (20) observed an inverse relationship between the degree of hemorrhage and tumor size, their analysis was performed on a per-sextant basis, such that adjacent but distinct areas of hemorrhage and prostate cancer could have been regarded as coexistent tumor and hemorrhage, when such instances may in fact have represented chance association of tumor and hemorrhage within the same sextant. By evaluating for hemorrhage within the precise margins of the tumor, as determined by unblinded review of T2WI in conjunction with histologically defined tumor maps, we believe that we have provided some of the strongest evidence to date of the low incidence of post-biopsy hemorrhage within foci of tumor. Regardless of this low rate of hemorrhage within tumor foci, review of T1WI alone would not be adequate

We acknowledge several limitations of our study. First, the AUCs that we observed for distinguishing tumor from benign PZ using nT2, ADC, and nADC are extremely high. This likely reflects a combination of the retrospective manner in which the ROIs were placed for tumor foci, unblinded to pathologic findings, the inclusion of only tumor foci that were visible on T2-weighted images or on ADC maps for placement of ROIs on these respective images, and the inclusion of benign non-hemorrhagic PZ in the ROC analysis. We believe that direct correlation with the tumor maps was the only possible way of ensuring that our manually placed ROIs occurred entirely within areas of tumor without partial volume averaging with adjacent benign prostate tissue. Based on our data, a future prospective study assessing ROIs placed on visually suspected foci of tumor and benign hemorrhage on T2WI and ADC maps, blinded to pathologic findings, seems warranted to evaluate the diagnostic value of this quantitative approach in a clinical context. Second, we used hand-drawn tumor maps as our reference standard; although whole-mount step-section histologic examination of prostatectomy specimens provides a more precise “gold standard,” this procedure is not routinely performed at our institution and was only available for 6 cases. In addition, we looked solely at the ADC maps without review of the primarily acquired diffusion-weighted images. Such an approach was used in three recent studies of the utility of DWI for prostate MRI (6, 28, 32); although we acknowledge the possible role of the DWI for prostate cancer detection, the purpose of our study was solely to try to use a quantitative method to distinguish hemorrhage and prostate cancer using T2 SI and ADC values and, accordingly, inclusion of the DWI in our analysis was not necessary. Also, we used T1WI as our reference standard for the presence of hemorrhage. While it has been postulated that proteinaceous secretions or other processes besides hemorrhage may cause high T1 signal in the PZ (10, 33), T1WI has served as a reference standard for localization of post-biopsy hemorrhage in numerous studies (8–11, 20). Finally, the patients in this study were imaged only with a pelvic phased-array coil; although use of an endorectal coil would have achieved a higher SNR, we do not routinely use this approach at our institution.

In conclusion, our study shows that quantitative T2 SI and ADC/nADC values may be used to distinguish reliably between prostate cancer and post-biopsy hemorrhage. ADC values, but not normalized T2 SI, showed a weak but significant correlation with tumor size and grade. In addition, we have documented an infrequent rate of post-biopsy hemorrhage within areas of prostate cancer.

Ancillary