Reproducibility and correlation between quantitative and semiquantitative dynamic and intrinsic susceptibility-weighted MRI parameters in the benign and malignant human prostate

Authors


Abstract

Purpose:

To assess the reproducibility of relaxivity- and susceptibility-based dynamic contrast-enhanced magnetic resonance imaging (MRI) in the benign and malignant prostate gland and to correlate the kinetic parameters obtained.

Materials and Methods:

Twenty patients with prostate cancer underwent paired scans before and after androgen deprivation therapy. Quantitative parametric maps for T1- and T2*-weighted parameters were calculated (Ktrans, kep,ve, IAUC60, rBV, rBF, and R2*). The reproducibility of and correlation between each parameter were determined using standard methods at both timepoints.

Results:

T1-derived parameters are more reproducible than T2*-weighted measures, both becoming more variable following androgen deprivation (variance coefficients for prostate Ktrans and rBF increased from 13.9%–15.8% and 42.5%–90.8%, respectively). Tumor R2* reproducibility improved after androgen ablation (23.3%–11.8%). IAUC60 correlated strongly with Ktrans, ve, and kep (all P < 0.001). R2* did not correlate with other parameters.

Conclusion:

This study is the first to document the variability and repeatability of T1- and T2*-weighted dynamic MRI and intrinsic susceptibility-weighted MRI for the various regions of the human prostate gland before and after androgen deprivation. These data provide a valuable source of reference for groups that plan to use dynamic contrast-enhanced MRI or intrinsic susceptibility-weighted MRI for the assessment of treatment response in the benign or malignant prostate. J. Magn. Reson. Imaging 2010;32:155–164. © 2010 Wiley-Liss, Inc.

RELAXIVITY-WEIGHTED DYNAMIC CONTRAST-ENHANCED magnetic resonance imaging (DCE-MRI) techniques utilizing low molecular weight contrast media have become mainstream clinical tools with recognized indications in the imaging of prostate cancer (1). Current roles of T1-weighted techniques include tumor staging (depiction of capsular penetration and seminal vesicle invasion) and the detection of suspected tumor recurrence following definitive treatment. Roles in primary diagnosis, the assessment of treatment response, and for radiotherapy planning are currently under evaluation. Susceptibility-weighted techniques, which include T2*-weighted dynamic susceptibility contrast MRI (DSC-MRI) and intrinsic susceptibility-weighted MRI (ISW-MRI) (which is also known as blood oxygen level-dependent [BOLD] MRI) are largely experimental at present, although there is increasing evidence that prostate tumor blood flow and oxygenation status can be determined using these methods (2–6).

All quantitative biological measurements exhibit variability. This is caused by natural variations in the quantity being measured, variations in the measurement process, and errors introduced during analysis. For any measurement to be useful in the determination of whether a change has occurred, for example, as a result of treatment, knowledge of its reproducibility is required. With regard to dynamic and intrinsic susceptibility-weighted MRI studies of the prostate, variability can occur at several stages of the examination. Patient movement can be minimized with careful explanation of the procedure and ensuring comfort and relaxation during the examination but internal prostate movement is largely unavoidable (7). Cardiac output and local variations in patient blood flow introduce error when data are analyzed using a pooled arterial input function. For comparative studies, the ability to image the same volume through the prostate on successive examinations will influence reproducibility. This in turn is influenced by the extent to which any intervention between examinations alters the morphological appearance, size, and shape of the prostate gland. Measurement factors such as MRI scanner type and settings, variations in the injection procedure, data acquisition technique, and number of imaging sites may introduce variability between departments and between studies. Analysis methods, including the mathematical model used, parameter derived, assumptions made, and inter- and intraobserver variability in region of interest (ROI) placement will also affect the overall reproducibility of measurements.

There are limited numbers of published reports documenting the reproducibility of dynamic and susceptibility-weighted MRI in normal human tissues and tumors. A study of 19 patients assessed the reproducibility of the T1-weighted DCE-MRI parameters; transfer constant (Ktrans), leakage space (ve), and maximum contrast medium accumulation (MCMA) in pelvic muscle, bone marrow, and fat (8). Results showed high variability in bone marrow and fat, making these tissues unsuitable for obtaining kinetic parameter estimates. This study also demonstrated low, statistically nonsignificant, variance ratios for pelvic muscle, suggesting that the within-subject variance was large compared to the between-subject variance. In a similar study of 21 patients with a variety of tumor types, the reproducibility of semiquantitative (gradient, enhancement, and area under the curve [AUC]) and quantitative (Ktrans, ve, and rate constant, kep) parameters were compared (9). The semiquantitative measures were shown to be highly reproducible DCE-MRI parameters. Quantitative measures had greater variability, with larger changes in individuals required to be statistically significant, but were nevertheless sufficiently reproducible to detect changes greater than 14%–17% from baseline for larger patient cohorts.

The reproducibility of T1- and T2*-weighted parameters, including R2* derived from ISW-MRI studies, has been reported in pelvic tumors (predominantly ovarian or uterine in origin) (10). These are the only published data regarding the reproducibility of T2* or R2*-weighted parameters in extracranial human tumors. The susceptibility-weighted parameters proved to be less reproducible than those derived from relaxivity-weighted DCE-MRI techniques.

All studies conclude that measures of reproducibility are essential for valid assessment of treatment effect and should be incorporated into clinical research protocols that utilize dynamic relaxivity or susceptibility-weighted MRI parameters for response assessments. Thus, the purpose of this study was to determine the reproducibility of parameters derived from DCE, DSC, and ISW-MRI studies, for the prostate as a whole and then specifically for prostate tumor, normal prostatic peripheral zone, and benign prostatic transition zone, before and after 3 months of androgen deprivation. These data provide a benchmark of reproducibility values necessary for the design of future quantitative functional studies of the prostate gland.

MATERIALS AND METHODS

Patients, Treatment, and Examination Schedule

Following Institutional Review Board approval, 20 patients with localized prostate cancer (age 57–78 years old, stage T1c–T3b, Gleason grade 6–9, prostate-specific antigen [PSA] 3.7–34.0 ng/mL) who were due to be treated with neoadjuvant androgen deprivation prior to radical radiotherapy were recruited prospectively. Informed consent was obtained after the nature of the study had been fully explained. A diagnosis of prostate cancer was made by transrectal core biopsy in all 20 patients. Each patient was examined with two scans on consecutive days prior to androgen suppression and two further scans after 3 months of endocrine therapy. After completing the two baseline scans, patients were commenced on Bicalutamide (Casodex, AstraZeneca, Wilmington, DE) 50 mg daily for 28 days and Goserelin (Zoladex LA, AstraZeneca) 10.8 mg subcutaneous injection given after 14 days of bicalutamide therapy, repeated every 12 weeks. Each patient received at least 3 months of androgen deprivation therapy alone, before proceeding to any further local or systemic treatment for prostate cancer. No other treatments were administered during the study period.

MRI Methods

Patients were imaged in the same Symphony 1.5T MRI scanner (Siemens Medical Systems, Erlangen, Germany) using a phased array pelvic coil. Anatomical images were acquired, followed by functional images in the following sequence: 1) ISW-MRI, 2) T1-weighted DCE-MRI, and 3) T2*-weighted DSC-MRI.

Anatomical Scanning

Small field of view (FOV) T2-weighted anatomical scans perpendicular to the urethra were used to stage tumors and to identify tumor slice locations (TR = 3600 msec, TE = 127 msec, ETL = 29, FOV = 200 mm, 3 mm thickness, matrix size = 256 × 256). Images were inspected for the presence of an abnormality consistent with cancer. The central slice through tumor was chosen in order to plan the functional MRI sequences as below.

Functional Imaging

ISW-MRI.

Five spoiled gradient-echo images were acquired for three 8-mm slices through the prostate with varying TE = 5–60 msec, TR = 100 msec, flip angle = 40°, FOV = 200 mm, 2562 matrix, acquisition time 26 seconds each, from which R2* maps were calculated.

DCE-MRI.

Proton density-weighted spoiled gradient-recalled echo (FLASH) sequences images were acquired first (TE = 10 msec, TR = 350 msec, flip angle = 10°, three slices of 8 mm thickness, FOV = 200 mm, 2562 matrix). T1-weighted gradient-recalled echo FLASH sequences (TE = 5 msec, TR = 74 msec, flip angle = 70°, 3 slices of 8 mm thickness, FOV = 200 mm, 2562 matrix) were then obtained sequentially with a time resolution of 12 seconds (40 timepoints over 8 minutes). A bolus of 0.1 mmol/kg b.w. of gadopentetate dimeglumine (Gd-DTPA, Magnevist, Bayer-Schering, Burgess Hill, UK) contrast agent was administered at 4 mL/s during the fifth image acquisition (ie, beginning after 48 seconds) using a power injector, followed by a 20 mL bolus of normal saline at the same rate.

DSC-MRI.

A gradient-recalled echo T2*-weighted sequence was used to acquire data every 2 seconds over 2 minutes (TE = 20 msec, TR = 30 msec, flip angle = 40°, one slice of 8 mm thickness, FOV = 200 mm, 1282 matrix). The central slice from the ISW-MRI and DCE-MRI examinations was chosen as the plane of imaging. A bolus of 0.2 mmol/kg b.w. of Gd-DTPA was administered at 4 mL/s after 20 seconds, followed by a 20-mL bolus of normal saline at the same rate. The time interval between contrast injections for parts 2 and 3 was 10–12 minutes to allow for equilibrium of plasma Gd-DTPA following the first bolus.

Region of Interest (ROI) Definition

Voxel-based calculations were performed using two customized analysis software packages developed at the Institute of Cancer Research, London, UK. For the DCE and DSC-MRI imaging, Magnetic Resonance Imaging Workbench (MRIW) v. 4.2.1 was used (11). For the R2* analysis, DiffusionView v. 2.1.3 was used.

Four regions of interest were outlined for each prostate (whole prostate, tumor, peripheral zone, and transition zone). ROIs were defined using information from several acquisition sequences. The whole prostate ROIs were defined on T2-weighted and T1-weighted images. Tumor ROIs were defined using combinations of T2-weighted and contrast-enhanced T1-weighted images. In general, an irregular mass of low signal intensity in the peripheral zone seen on the T2-weighted images was considered to represent tumor. When an obvious malignant peripheral zone tumor was contiguous with homogeneous low signal intensity in the transition zone the ROI was extended appropriately. An oncologically trained radiologist with a specialist interest in prostate cancer MRI independently verified these regions (A.R.P.). Normal peripheral zone was determined on the basis of homogeneous high signal intensity on noncontrast-enhanced T2-weighted sequences. Transition zone ROIs were also outlined on noncontrast-enhanced T2-weighted sequences, based on anatomical details. The MRI features of benign prostatic hyperplasia were uniform areas of low signal intensity (glandular benign prostatic hyperplasia) or areas of nodular whorls with low signal intensity. An ROI was also placed over the obturator internus muscle as a normal tissue for R2* analysis. This process was repeated for each patient and for all slices per MRI examination. ROI definition was performed on the same day for all four scans for each patient in order to minimize intraobserver variability.

MRI Data Analysis

Signal changes on ISW-MRI were used to calculate intrinsic T2* relaxivity. R2* (=1/T2*) values were calculated pixel-by-pixel from a straight line fit plot of lnSt against TE for each pixel using a least-squares approach, of which the gradient is −R2* (units: s−1). Pixels with either negative or zero values were excluded from analysis. Median values were reported for each ROI.

Signal intensity enhancement on the T1-weighted DCE-MRI images were assessed quantitatively using Tofts' pharmacokinetic model. Quantitative kinetic parameters were derived from mathematically fitted concentration–time curves (12). A pooled arterial input function (modified Fritz-Hansen) was used for these analyses (13–15). Derived modeling parameters include the volume transfer constant of the contrast agent (Ktrans—formally called permeability-surface area product per unit volume of tissue; unit: min−1) (Fig. 1), leakage space as a percentage of unit volume of tissue (ve: unit %), and the rate constant (kep: unit min−1), which were calculated for every scan and median values were reported for each ROI. These standard parameters are related mathematically (kep = Ktrans/ve) (16). The initial area under the Gd-DTPA concentration–time curve for the first 60 seconds (IAUC60, unit: mmol.s) was also derived. Voxels that could not be fitted and those with poor fit were identified as previously described (11). Briefly, the model reported an enhancing fail if ve > 90% or the Ktrans > 20 min−1. It reported a nonenhancing fail if the gadolinium concentration >30 mmol or <5 mmol, or if a voxel failed to enhance over a calculated threshold after injection. The program reported a general fail if it was unable to fit the model to the data. Enhancing fails were excluded; nonenhancing fails could be indicative of treatment response and so were included.

DSC-MR images were analyzed by calculating ΔR2* values for each timepoint of the T2* weighted DSC-MRI dataset and fitted using a gamma-variate function. The relative blood volume (rBV; arbitrary units) and mean transit time (MTT; approximated by measuring the width of the ΔR2*-time curve at half its maximum value, units: s) were derived and median values were reported for each ROI. Relative blood flow (rBF) was then obtained by substituting in the transit time equation (MTT = rBV/rBF, units: au).

Statistical Methods for Reproducibility Calculations

Spontaneous Variability of Parameter Estimates

Within-subject coefficient of variation (wCV).

wCV is the within-subject standard deviation divided by the mean, multiplied by 100 to give a percentage. Variance ratio (F) is the ratio of the between-patient variance and within-patient variance. A parameter with a larger variance in the patient population, but a small variance within individual patients (wCV) would have a higher variance ratio.

Measurement Error

Intraclass correlation coefficient (ICC).

ICC is a measure of the precision of parameter estimates. It is the average correlation coefficient across all possible orderings of the data when there is no obvious choice of X and Y axis for each of the pairs of measurements. Repeatability statistic: threshold values (as a percentage change) beyond which the absolute difference between two measurements on the same patient (n = 1) is expected to lie for 95% of the pairs of observations. From this, one can calculate the number of patients in the group that achieved a parameter change that would be considered significant, at the 95% confidence level, as a real change rather than simply being due to measurement variability.

The standard consensus approach to assessing reproducibility was used (17–19). For each patient the difference between the measurements of a parameter at each reproducibility scan, d, was calculated. The distribution of d was tested for normality using the Shapiro–Wilk test. In order to establish whether the size of d was dependent on the parameter value, Kendall's tau for correlation of the absolute value of d against the mean parameter value for the two scans was calculated (18). If this test demonstrated that error was indeed proportional to the mean, at the 95% confidence level, then the data were transformed using natural logarithms (ln). The Shapiro–Wilk and Kendall's tau test were then repeated. The following statistical measures of reproducibility were then obtained from a one-way analysis of variance (ANOVA) on the original or transformed data:

The mean squared difference (dsd) was calculated:

equation image

The within-patient standard deviation (wSD) was calculated:

equation image

The within-patient coefficient of variance was calculated:

equation image
equation image

The repeatability parameter, r, was calculated as follows:

equation image
equation image

The ratio of the between-patient variance to the within-patient variance was derived for each parameter and tested for a significant difference in these variances. A parameter with a large variance in the patient population tested but a small variance within individual patients would have a high value of this ratio.

The ICC was calculated as follows:

equation image

where m is the number of observations per subject, SSB is the sum of squared between subjects, and SST is the total sum of squares (as per one-way ANOVA above).

In order to determine whether there were any significant correlations between any of the baseline kinetic parameter values, Spearman's ρ (rho) was calculated. Because multiple statistical testing was used in these exploratory analyses, a significance level of 0.01 was used.

Statistical analysis was performed using the StatsDirect statistical software package (StatsDirect, Cheshire, UK) and Microsoft Excel 2000 (Microsoft, Redmond, WA).

RESULTS

All 20 patients underwent the MRI studies according to schedule. One patient moved excessively (several centimeters) during the T1 dynamic experiment of his first scan resulting in no data being available from this study. Due to operator error, one patient did not receive a T2*-weighted imaging sequence during his second scan. One patient had no tumor visible on any scan, despite positive histology on prostate biopsy, reducing the number of patients evaluable for tumor ROI changes to 19. Two patients had no discernable normal peripheral zone and another two had no definable transition zone due to previous transurethral resection. No adverse events occurred during the running of the study. The median time between the first two scans was 2 days (range 1–8 days) and between the last two scans was 1.5 days (range 1–14 days). The median time between starting androgen deprivation and the third scan was 91.5 days (range 82–105 days).

The baseline means of the median kinetic parameter values for whole prostate are given in Table 1. In general these values are comparable with literature values for normal human tissues and human tumors (8–10, 20, 21). Reproducibility statistics for the prostate regions are also shown in Table 1 and illustrated graphically by Bland–Altman plots (Fig. 2). Results of the R2* analyses in obturator internus muscle are shown in Table 2. No parameter, either for baseline pairs or pairs after 3 months of androgen deprivation, had a mean difference significantly different from zero. In other words, there was no systematic change in parameter value, in either direction, between the paired sets of scans either at baseline or after treatment.

Table 1. Reproducibility Statistics for All Prostate Regions
 Ktrans (mins−1)ve(%)kep (mins−1)IAUC60(mmol.s)rBV (AU)rBF (AU)R2* (s−1)
Baseline3 monthsBaseline3 monthsBaseline3 monthsBaseline3 monthsBaseline3 monthsBaseline3 monthsBaseline3 months
Whole prostateNumber of patients1920192019201920192019202020
Mean ROI size (pixels)6725522267255222672552226725522264844864844868045534
Mean value0.2060.14641.647.30.5130.32710.847.46141.4247.942.370.9817.2420.73
95% CI of mean0.24–0.170.17–0.1237.0–46.242.2–52.40.46–0.570.29–0.369.08–12.66.11–8.8093.2–18935.9–60.01.64–3.100.75–1.2115.3–19.218.8–22.7
Repeatability for n = 1 (%)−30.3 to 43.4−33.4 to 50.2−24.7 to 24.7−33.1 to 33.1−30.3 to 43.6−39.1 to 64.2−32.3 to 47.7−41.0 to 69.4−68.6 to 218.3−84.0 to 525.2−62.5 to 166.7−83.3 to 499.0−60.1 to 60.1−34.1 to 34.1
Variance ratio (F)12.510.013.47.65.74.110.98.63.51.63.71.32.55.1
P-value for F<.0001<.0001<.0001<.00010.00020.0014<.0001<.00010.0050.150.0030.300.0250.0003
ICC0.840.810.850.760.690.590.820.780.530.210.560.090.400.66
wCV13.9%15.8%8.9%11.9%13.9%19.6%15.1%21.0%51.9%93.8%42.5%90.8%21.7%12.3%
TumorNumber of patients1919191919191919181818181919
Mean ROI size (pixels)10466141046614104661410466141266812668924668
Mean value0.2730.13041.245.30.6800.31114.237.28251.7734.184.200.7816.5923.41
95% CI of mean0.20–0.340.10–0.1635.4–47.038.7–51.80.56–0.810.24–0.3910.7–17.85.34–9.23163–340.421.3–47.12.89–5.510.49–1.0814.6–18.621.6–25.2
Repeatability for n = 1 (%)−39.8 to 66.2−38.6 to 62.8−34.6 to 34.6−48.6 to 48.6−44.8 to 81.3−60.6 to 153.7−36.3 to 57.1−49.4 to 97.7−81.1 to 427.7−81.2 to 430.6−77.9 to 351.5−78.8 to 371.9−64.6 to 64.6−32.8 to 32.8
Variance ratio (F)11.223.211.05.94.92.623.015.72.54.62.24.22.33.6
P-value for F<.0001<.0001<.00010.00020.00060.023<.0001<.00010.0330.0010.050.0020.0390.004
ICC0.830.910.820.700.650.420.910.870.400.630.350.600.370.55
wCV20.1%19.2%12.5%17.5%24.0%39.9%17.7%27.9%82.3%82.7%72.3%75.1%23.3%11.8%
Benign peripheral zoneNumber of patients1818181818181818171717171818
Mean ROI size (pixels)16211616211616211616211645304530155121
Mean value0.1260.06929.235.20.4610.2177.324.8486.9723.691.540.5214.0919.66
95% CI of mean0.09–0.160.05–0.0824.0–34.428.0–42.50.36–0.570.19–0.255.47–9.183.61–6.0638.9–14017.9–29.50.85–2.220.42–0.6111.1–17.117.0–22.3
Repeatability for n = 1 (%)−62.1 to 163.8−41.0 to 69.4−41.1 to 41.1−76.7 to 76.7−66.8 to 201.4−58.4 to 140.4−58.5 to 140.8−45.0 to 81.9−88.2 to 747.21−80.0 to 399.7−83.8 to 516.0−80.1 to 401.3−115.7 to 115.7−61.0 to 61.0
Variance ratio (F)3.313.511.74.52.01.54.56.05.51.62.90.92.13.1
P-value for F0.008<.0001<.00010.0010.008<.00010.00140.00020.00060.170.0170.590.060.011
ICC0.510.850.830.620.300.180.620.700.670.210.47−0.090.340.49
wCV41.9%21.0%14.8%27.7%48.9%37.2%37.3%24.1%116.3%78.7%92.8%79.0%41.8%22.0%
Benign transition zoneNumber of patients1818181818181818171717171818
Mean ROI size (pixels)17412617412617412617412643364336190145
Mean value0.2530.20632.642.20.8080.50013.0212.01164.8072.602.821.2210.3513.73
95% CI of mean0.20–0.310.16–0.2525.9–39.036.1–48.40.68–0.940.41–0.5910.5–15.69.5–14.5102–22739.1–1061.85–3.780.72–1.728.0–12.711.2–16.3
Repeatability for n = 1 (%)−40.5 to 68.2−50.8 to 103.3−36.1 to 36.1−35.1 to 35.1−35.8 to 55.8−58.1 to 138.9−32.8 to 48.9−44.5 to 80.2−81.5 to 440.7−93.3 to 1384.4−75.9 to 314.5−90.9 to 1000.0−140.4 to 140.4−81.4 to 81.4
Variance ratio (F)14.96.619.310.79.52.616.28.43.71.43.91.61.73.3
P-value for F<.00010.0001<.0001<.0001<.00010.0001<.0001<.00010.0050.260.0040.180.150.008
ICC0.870.730.900.820.800.430.880.780.560.130.570.190.220.51
wCV20.6%29.2%13.0%12.7%17.4%37.0%15.5%23.7%83.9%164.8%67.1%137.7%50.7%29.4%
Figure 1.

A 67-year-old patient with T3b, Gleason 5+4 carcinoma of the prostate which can be seen on the T2-weighted MRI image as a region of low signal intensity within the otherwise bright peripheral gland (arrow, A). B: The plot of gadolinium concentration against time; contrast medium injection took place after the fourth data point. Time signal intensity data is first converted into contrast agent concentration and then the model fitting procedure (continuous green line) is performed using the Tofts' model. The data plotted in panel B was used to calculate the parametric map representing transfer constant (Ktrans; color scale 0–1 min−1) shown in C. C,D: Ktrans maps on consecutive days prior to the start of androgen deprivation therapy and illustrate the reproducibly high tumor permeability surface area product at baseline. After 3 months of androgen withdrawal, tumor Ktrans decreases markedly as depicted in the scans shown in E,F, also performed on consecutive days.

Figure 2.

Bland–Altman plots for Ktrans, rBV, and R2* in the four prostate regions (whole prostate, tumor, benign peripheral zone, and benign transition zone). The mean value for each parameter is shown as the green line with the 95% limits of agreement shown as dotted black lines.

Table 2. R2* - Normal Tissue (Obturator Internus Muscle)
Obturator Internus Muscle - R2* BaselineObturator Internus Muscle - R2* 3 Months
Number of patients20Number of patients20
Mean ROI size (pixels)11632Mean ROI size (pixels)10956
Mean value (s−1)25.02Mean value (s−1)25.01
95% CI of mean24.07–25.9795% CI of mean24.16–25.86
Repeatability for n = 1 (%)−11.2 to 11.2Repeatability for n = 1 (%)−3.3 to 3.3
Variance ratio (F)8.07Variance ratio (F)71.7
P-value for F<0.0001P-value for F<0.0001
ICC0.77ICC0.97
wCV4.0%wCV1.2%

The distribution of each parameter was not significantly different from normal, either at baseline or at 3 months. However, Kendall's tau test showed the proportionality of error and mean value for Ktrans, kep, AUC60, rBV, and rBF, and these data were natural log (ln) transformed. The difference between values for ve and R2* showed no dependence on the mean and therefore no transformation was required.

Correlation between the different parameters at baseline is shown in Table 3. For whole prostate ROIs, AUC60 was found to correlate very strongly with Ktrans (ρ = 0.95, P < 0.0001) and strongly with ve (ρ = 0.74, P = 0.0003). Stronger correlations were observed for tumor measurements: AUC60 vs. Ktrans (ρ = 0.98, P < 0.0001) and AUC60 vs. kep (ρ = 0.83, P < 0.0001). Relative blood flow and relative blood volume exhibit extremely high correlation (ρ = 0.98 for both whole prostate and tumor, P < 0.0001) reflecting their close mathematical relationship and the narrow range of mean transit time (MTT) values observed (rBF = rBV/MTT). R2* did not correlate with any other parameter for whole prostate, tumor, transition zone, or peripheral gland. Correlations after 3 months of androgen deprivation are shown in Table 4. The most notable differences in correlation prior to and after androgen deprivation were that after androgen deprivation the correlation between the T1-weighted and T2* weighted parameters in tumor were lost and a significant correlation developed in benign transition zone.

Table 3. Correlation Coefficients (ρ) Between DCE, DSC, and ISW-MRI Parameters at Baseline
 AUC60rBVrBFR2*
ρP-valueρP-valueρP-valueρP-value
  1. Figures in bold are significant at the P ≤ 0.01 level.

Whole prostate
 Ktrans0.95<0.00010.680.00120.660.0020.170.4682
 ve0.740.00030.380.09880.310.17910.100.6766
 kep0.620.00460.680.00130.700.00090.260.264
 AUC60  0.700.00080.670.00160.220.344
 rBV    0.98<0.00010.060.8016
 rBF      0.100.6673
Tumor
 Ktrans0.98<0.00010.560.01380.590.00860.290.2171
 ve0.490.0330.330.16780.350.14530.190.4227
 kep0.83<0.00010.580.0110.590.00820.180.4619
 AUC60  0.590.00940.620.00550.360.1289
 rBV    0.98<0.0001−0.120.6309
 rBF      −0.090.7048
Normal peripheral zone
 Ktrans0.99<0.00010.380.11350.380.11350.500.035
 ve0.610.00790.190.44630.160.51930.220.3838
 kep0.650.00460.260.29410.290.24230.470.05
 AUC60  0.410.08740.410.09260.500.0354
 rBV    0.98<0.0001−0.170.4979
 rBF      −0.100.6804
Normal transition zone
 Ktrans0.96<0.00010.470.05110.570.01370.290.2388
 ve0.82<0.00010.140.58040.270.27110.110.6444
 kep0.530.02440.520.02770.520.0270.250.302
 AUC60  0.420.08080.550.01980.220.3792
 rBV    0.95<0.0001−0.090.7357
 rBF      −0.080.7544
Table 4. Correlation Coefficients (ρ) Between DCE, DSC, and ISW-MRI Parameters at 3 Months
 AUC60rBVrBFR2*
ρP-valueρP-valueρP-valueρP-value
  1. Figures in bold are significant at the P ≤ 0.01 level.

Whole prostate
 Ktrans0.92<0.00010.560.01130.580.0082−0.080.7431
 ve0.660.00190.150.50870.220.35070.040.8762
 kep0.730.00030.710.00070.600.0055−0.200.3998
 AUC60  0.570.00900.580.00800.060.7967
 rBV    0.91<0.0001−0.190.4257
 rBF      −0.060.7869
Tumor
 Ktrans0.99<0.00010.290.23150.350.14110.240.3130
 ve0.560.01280.240.31300.130.59020.180.4530
 kep0.80<0.00010.260.28520.370.11390.230.3386
 AUC60  0.290.22860.340.15410.220.3576
 rBV    0.94<0.0001−0.180.4709
 rBF      −0.120.6361
Normal peripheral zone
 Ktrans0.96<0.00010.380.11560.260.29020.060.8178
 ve0.83<0.00010.160.55780.010.98030.180.4718
 kep0.420.08570.430.07310.430.0792−0.220.3838
 AUC60  0.420.08400.260.2863−0.020.9541
 rBV    0.80<0.0001−0.080.7419
 rBF      0.020.9214
Normal transition zone
 Ktrans0.98<0.00010.710.00130.700.0016−0.050.8306
 ve0.700.00150.210.40250.180.46150.230.3480
 kep0.760.00040.760.00040.770.0003−0.030.9214
 AUC60  0.600.00900.590.0105−0.040.8629
 rBV    0.99<0.0001−0.080.7419
 rBF      −0.110.6563

DISCUSSION

This analysis has shown considerable difference in the variability of the various kinetic parameters for the four prostate regions of interest, namely, whole prostate, tumor, normal peripheral zone, and benign transition zone. Tumor ROIs are generally the most relevant, given that the tumor is the intended target for prostate cancer therapy. Differentiation of tumor from benign peripheral zone is often difficult and a frequent subject of radiological publications concerning the prostate gland. In this study great care was taken to utilize all the available information for tumor ROI delineation, including combinations of T2-weighted and contrast-enhanced T1-weighted images. Given the challenges in defining tumor regions, it is encouraging that the tumor ROI baseline reproducibility was generally not substantially inferior to whole prostate regions for T1 parameters (wCV values of 12.5%–24% for tumor compared with 8.9%–15.1% for whole prostate), T2* parameters (wCV values of 72.3%–82.3% for tumor compared with 42.5%–51.9% for whole prostate), and for R2* (wCV value of 23.3% for tumor vs. 21.7% for whole prostate). This is despite fewer pixels per ROI. The coefficients of variance are comparable to values published in the literature. For example, in this study the wCV for tumor Ktrans was 20.1% compared with 20.3% reported by Lankester et al (10) and 24% reported by Galbraith et al (9) even though both of these studies investigated much more readily definable tumors such as ovarian, uterine, and renal carcinomas as well as pelvic or thoracic sarcomas. Similar comparisons can be made for the other T1 parameters.

The baseline reproducibility of the prostate T2* parameters rBV and rBF was poor, with wCV values ranging from 42.5%–116.3%. This is much worse than the only other reported reproducibility measures for T2* parameters, in gynecological malignancy, of 12.1% for rBV and 16.8% for rBF (10). Susceptibility-weighted MRI methods are more liable to be affected by susceptibility artifacts generated by changes in the volume of rectal gas. These artifacts tend to influence the posterior prostate due to its close proximity to the rectum. As a result, tumor and peripheral zone ROIs are likely to be most affected by this phenomenon. Accordingly, wCV for tumor and peripheral zone at baseline (72.3%–116.3%) were worse than wCV for whole prostate (42.5%–51.9%). Transition zone T2* reproducibility is also likely to be affected by the fact that this region of the prostate is very heterogeneous, with nodules of benign hyperplasia interspersed with normal glandular tissue, calcification, and fibrosis. As a result blood flow estimates are likely to vary considerably between these tissue types.

The reproducibility of R2* in pelvic muscle was extremely good (wCV of 4%, F = 8.07 at baseline; Table 2). This suggests that the measurement technique itself exhibits very low variability. However, when R2* was measured in tumors, variability was greater (wCV of 23.3%, F = 2.3). This may be explained by our understanding of tumor vasculature, which has chaotic structure, poorly formed, fragile vessels with high permeability to macromolecules. There is arteriovenous shunting, high vascular tortuosity and vasodilatation leading to intermittent or unstable flow due to transient rises in already raised interstitial pressure. This results in intermittent changes in blood flow causing periods of transient acute hypoxia. The variability of R2* may therefore reflect the intrinsic biology of prostate tumors rather than measurement error. This hypothesis is supported by the consistent improvement in the reproducibility of tumor R2* following androgen deprivation (pretreatment wCV of 21.7%–50.7% vs. 11.8%–29.4% posttreatment). Androgen withdrawal causes tumor vascular regression and a reduction in the proportion of abnormal blood vessels, which explains the reduction in mean Ktrans, kep, rBV, and rBF values seen after therapy.

The declines in reproducibility of the T1- and T2*-weighted parameters following androgen deprivation were modest and mixed in comparison with the consistent improvements seen in R2*. Using the same argument as for R2*, one would expect an improvement in the reproducibility of the values of relative blood flow and volume. This may be due to a “floor effect” following the marked decreases in rBV and rBF following androgen deprivation because error is calculated using the mean value as the denominator. The technical difficulties of T2* methods due to susceptibility artifacts may also mask any physiological stabilization to the tumor blood flow. The deterioration in reproducibility for whole prostate, tumor, and transition zone ROIs following androgen withdrawal may have a simpler explanation. Androgen deprivation causes a reduction in prostate volume (by an average of 34% in this study), thereby reducing the number of pixels for whole prostate analysis. The distinction between tumor and normal peripheral zone becomes even harder to discern following the reduction in peripheral zone signal intensity on T2-weighted sequences, making the definition of these regions less reliable. Similarly, the altered prostate morphology made accurate recreation of transition zone ROIs from examination to examination more difficult, thereby increasing variability.

It is interesting that at baseline, correlations were demonstrated between DSC-MRI and DCE-MRI parameters in regions of prostate cancer but not in the benign transition and peripheral zones (Table 3). This provides evidence supporting the view that the enhancement of prostate tumors is flow-dominated. In other words, in malignant tissue the transfer constant approximates to the blood plasma flow per unit volume of tissue because of the greatly increased microvessel permeability (12) and by the vascular endothelial growth factor (VEGF)-induced increase in blood vessel density (22). The lack of correlation between R2* and rBF/rBV in prostate cancer prior to androgen deprivation was in stark contrast with the strong negative correlation that has been previously demonstrated in breast cancer. This probably reflects differences in microvessel density and maturity, with breast cancer having a greater proportion of mature vessels, thereby decreasing its flow-dependency (22). Three months of androgen deprivation causes a profound reduction of VEGF and consequently a decline in the neovasculature, leading to a more mature vascular phenotype. This is reflected by the loss of correlation between the T1-weighted and T2*-weighted parameters in prostate cancer following androgen withdrawal (Table 4) as the tumor vasculature becomes less permeable and the T1-enhancement becomes less dependent on blood flow.

The scope of this analysis is limited to the two timepoints studied, namely, baseline and after 3 months of androgen deprivation. It is likely that the effects of androgen withdrawal occur over a period of days and that the vascular changes are established by 1 month at most. Therefore, there is a short period of rapid change during which these results may be unreliable and further studies may be required to describe the reproducibility and correlation between the various parameters during the first few days after androgen deprivation. However, the data presented here can guide the design of studies concerning patients that are either naïve to androgen deprivation or alternatively for those who are well established on hormonal therapy.

In conclusion, this study is the first to document the variability and repeatability of T1- and T2*-weighted dynamic MRI and ISW-MRI for the various regions of the human prostate gland. Reproducibility statistics for each kinetic parameter in every prostate region have been documented and will provide a valuable source of reference for any group that plans to use these measures as a method for assessing treatment response in the benign or malignant prostate. Future studies may continue to unravel the complexity of biological variability and measurement error in the prostate by providing a greater understanding of the relative contributions from the numerous causative factors. Awareness of the intrinsic variability of any biomarker is vital for accurate study design. In particular, knowledge of the degree a particular parameter has to change, on an individual patient basis, for the change to be considered statistically significant, will facilitate accurate power calculations. T1-derived kinetic parameters exhibit a high degree of reproducibility compared to the T2*-weighted measures. Both sets of parameters become more variable following androgen deprivation. In contrast, the reproducibility of R2* improves with androgen ablation therapy.

Ancillary