FULL PAPER Magnetic Resonance in Medicine 75:2130–2140 (2016)

Relaxation Along Fictitious Field, Diffusion-Weighted Imaging, and T2 Mapping of Prostate Cancer: Prediction of Cancer Aggressiveness Ivan Jambor,1* Marko Pesola,1 Harri Merisaari,2,3 Pekka Taimen,4 Peter J. Bostr€ om,5 6,7 1,8 Timo Liimatainen, and Hannu J. Aronen INTRODUCTION

Purpose: To evaluate the performance of relaxation along a fictitious field (RAFF) relaxation time (TRAFF), diffusion-weighted imaging (DWI)-derived parameters, and T2 relaxation time values for prostate cancer (PCa) detection and characterization. Methods: Fifty patients underwent 3T MR examination using surface array coils before prostatectomy. DWI was performed using 14 and 12 b values in the ranges of 0–500 s/mm2 and 0–2000 s/mm2, respectively. Repeated MR examination was performed in 16 patients. TRAFF, DWI-derived parameters (monoexponential, kurtosis, biexponential models), and T2 values were measured and averaged over regions of interest placed in PCa and normal tissue. Repeatability of TRAFF and DWI-derived parameters were assessed by coefficient of repeatability and intraclass correlation coefficient ICC(3,1). Areas under the receiver operating characteristic curve (AUCs) for PCa detection and Gleason score classification were estimated. The parameters were correlated with Gleason score groups using Spearman correlation coefficient (r). Results: ICC(3,1) values for TRAFF were in the range of 0.82– 0.92. TRAFF values had higher AUC values for Gleason score classification compared with DWI-derived parameters and T2. The RAFF method demonstrated the highest r value (0.65). Conclusion: In a quantitative region of interest–based analysis, RAFF outperformed DWI (“low” and “high” b values) and T2 mapping in the characterization of PCa. Magn Reson Med C 2015 Wiley Periodicals, Inc. 75:2130–2140, 2016. V Key words: rotating frame relaxation; prostate cancer; diffusion weighted imaging; Gleason score

1

Department of Radiology, University of Turku, Turku, Finland. Department of Information Technology, University of Turku, Turku, Finland. 3 Turku PET Centre, University of Turku, Turku, Finland. 4 Department of Pathology, University of Turku and Turku University Hospital, Turku, Finland. 5 Department of Urology, University of Turku and Turku University Hospital, Turku, Finland. 6 Department of Biotechnology and Molecular Medicine, A. I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland. 7 Imaging Centre, Kuopio University Hospital, Kuopio, Finland. 8 Medical Imaging Centre of Southwest Finland, Turku University Hospital, Turku, Finland. lius Grant sponsor: Instrumentarium Research Foundation, Sigrid Juse Foundation, Academy of Finland, Turku University Hospital, TYKS-SAPA research fund, Finnish Cultural Foundation, University of Eastern Finland strategic funding. *Correspondence to: Ivan Jambor, MD, Department of Radiology, University of Turku, Kiinamyllynkatu 4-8, P.O. Box 52, FI-20521 Turku, Finland. E-mail: [email protected] 2

Received 5 March 2015; revised 20 May 2015; accepted 21 May 2015 DOI 10.1002/mrm.25808 Published online 22 June 2015 in Wiley Online Library (wileyonlinelibrary. com). C 2015 Wiley Periodicals, Inc. V

Prostate cancer (PCa) continues to be the second most common cause of cancer-related death among men in the western world (1). However, approximately 50% of patients with newly diagnosed PCa by means of systematic transrectal ultrasound (TRUS)-guided prostate biopsy have a low risk of disease progression or death due to PCa (2,3). Due to a wide range of PCa aggressiveness, patient tailed treatment is crucial in order to prevent overtreatment and limit PCa-related deaths. Patients with indolent disease might not require whole gland treatment, such as prostatectomy or external beam radiotherapy, and could benefit from active surveillance or focal therapy (4,5). Gleason score, consisting of Gleason grade, is the most commonly used marker of PCa aggressiveness (6). Gleason score has an important role in clinical nomograms used for risk stratification of patients with PCa (7,8). Unfortunately, a Gleason score based on systematic TRUS-guided biopsy often does not provide an accurate estimation of true Gleason score (ie, based on prostatectomy specimens). Thus, research attention is shifting from PCa detection to PCa characterization and risk stratification. Several previous studies have demonstrated a correlation of apparent diffusion coefficient (ADCm), applying the monoexponential model to diffusion-weighted imaging (DWI) data sets, with Gleason score based on prostatectomy sections (9–14) and biopsy cores (15–17). However, there is still a need for development and validation of novel noninvasive method for PCa characterization. Recently, a novel method called relaxation along a fictitious field (RAFF) has been developed (18–20). Rotating frame relaxation measurements using the RAFF method are made under subadiabatic conditions with approximately 40% less radiofrequency power deposition than a continuous wave rotating frame relaxation time (T1r) radiofrequency pulse with the same peak power and duration (21,22). In a preclinical glioma gene therapy study (23), relaxation values using RAFF (TRAFF) represented the only parameter having significant association with cell density in all tumor parts of a glioma rat model, outperforming DWI. We have already demonstrated that TRAFF values significantly (P < 0.01) differ between PCa lesion with Gleason score of 3 þ 3 version >3 þ 4 (24). However, our initial study consisted of only 36 patients and RAFF was performed using a voxel size of 2.47  3.08  5.00 mm3. No comparison with DWIderived parameters was performed.

2130

RAFF, DWI, and T2 Mapping of Prostate Cancer

2131

The study was approved by the local ethics committee, and each patient provided written informed consent. Between April 2013 and October 2014, 60 patients with histologically confirmed PCa and scheduled for roboticassisted laparoscopic prostatectomy underwent 3T MR examination performed using a 3T MR scanner (Ingenuity PET/MR; Philips, Cleveland, Ohio, USA). In one PCa patient, a gonadotropin-releasing hormone antagonist (Degarelix; Ferring Pharmaceuticals, Parsippany, New Jersey, USA) was started 10 days before the MR examination, whereas none of remaining patients had any hormonal, surgical, and/or radiotherapy treatment related to prostate before or at the time of MR examination. Ten patients were excluded from the final data set due to gross motion during MR examination and/or severe susceptibility artifacts due to rectal air. The patients’ characteristics are summarized in Supporting Table S1. Sixteen (32%, 16/50) of the 50 patients underwent two repeated MR examinations to evaluate short-term repeatability. Following the first MR examination, a patient was taken out of the MR bore and asked to rest for 10–15 min. After re-positioning of the patient on the MR table, the second MR examination was performed. The second MR examination consisted of RAFF and DWI. DWI data sets (obtained using b values in the range of 0–2000 s/mm2) of 37 patients were used in previous studies (25–27), whereas T2 data sets of 27 patients were included in a study evaluating feasibility of RAFF and continuous wave T1r imaging of PCa (24).

250 mm2; acquisition matrix size ¼ 124  124; reconstruction matrix size ¼ 256  256; slice thickness ¼ 5.0 mm; no intersection gaps; diffusion gradient timing (D) ¼ 21.204 ms; diffusion gradient duration (d) ¼ 6.6 ms; diffusion gradients applied in three directions (gradient overplus option on); SENSE (28) factor ¼ 2; partial Fourier acquisition ¼ 0.69; SPAIR fat suppression; number of signal averages for each b value ¼ 2; b values ¼ 0, 2, 4, 6, 9, 12, 14, 18, 23, 28, 50, 100, 300, and 500 s/mm2; and acquisition time ¼ 3 min, 45 s. The high b value set was acquired using the following parameters: TR/TE ¼ 3141/ 51 ms; FOV ¼ 250  250 mm2; acquisition matrix size ¼ 100  99; reconstruction matrix size ¼ 224  224; slice thickness ¼ 5.0 mm; intersection gaps ¼ 0.5 mm; diffusion gradient timing (D) ¼ 24.5 ms; diffusion gradient duration (d) ¼ 12.6 ms; diffusion gradients applied in three directions (gradient overplus option on); SENSE (28) factor ¼ 2; partial Fourier acquisition ¼ 0.69; SPAIR fat suppression; b values (number of signal averages) ¼ 0 (2), 100 (2), 300 (2), 500 (2) (2), 700 (2), 900 (2), 1100 (2), 1300 (2), 1500 (2), 1700 (3), 1900 (4), and 2000 (4) s/mm2; and acquisition time ¼ 8 min, 48 s. T2 relaxation values were measured using GraSE sequence with the following parameters: TR/TE ¼ 686/20, 40, 60, 80, and 100 ms; FOV ¼ 230  183 mm2; acquisition matrix size ¼ 256  163; reconstruction matrix size ¼ 512  400; slice thickness ¼ 5.0 mm; no intersection gaps; and acquisition time ¼ 1 min, 35 s. Furthermore, T2-weighted images were obtained using a single-shot TSE sequence with the following parameters: TR/TE ¼ 4668/130 ms; FOV ¼ 250  320 mm2; acquisition matrix size ¼ 250  320; reconstruction matrix size ¼ 512  672; slice thickness ¼ 2.5 mm; partial Fourier factor ¼ 0.6; SENSE (28) factor ¼ 2; and acquisition time ¼ 1 min, 10 s. The RF field homogeneity (B1 field) was evaluated using an actual flip angle imaging method (29) with the following parameters: TR/ TE ¼ 30, 150/2.2 ms; flip angle ¼ 60 ; FOV ¼ 400  400 mm2; and slice thickness ¼ 6.0 mm.

MRI

Data Analyses

A two-channel volume whole body RF coil was used for excitation, whereas signal was measured using the manufacturer’s 32-channel cardiac coil. A second-order rotating frame (RAFF) was used in the current study (18) with pulse train durations of 0, 45, and 90 ms and 500 Hz (gB1/2p) RF peak amplitude, which corresponds to 11.74 mT (B1). 3D T1-FFE sequence was used as a readout with the following parameters: repetition time (TR)/ echo time (TE) ¼ 4.0/2.3 ms; field of view (FOV) ¼ 250  250 mm2; acquisition matrix size ¼ 168  144; reconstruction matrix size ¼ 384  384; slice thickness ¼ 5.0 mm; number of slices ¼ 6; flip angle ¼ 20 ; TFE factor ¼ 10; centric k-space coding; partial Fourier factor ¼ 0.625; SENSE (28) factor ¼ 2; RAFF pulse interval ¼ 3000 ms; and acquisition time ¼ 5 min, 51 s. DWI data sets were acquired using a single-shot spin-echo based sequence with a monopolar diffusion gradient scheme and an echo-planar readout in two separate acquisitions consisting of 14 (low b value set) and 12 (high b value set) b values. The low b value set was acquired using the following parameters: TR/TE ¼ 1394/44 ms; FOV ¼ 250 

As a preprocessing step before region of interest (ROI) drawing and data analysis, averaging filter with kernel size of 3  3 voxels was used only for the DWI reconstructed data. The filter was applied to the reconstructed trace DWI data obtained using low and high b values. Regions of interest (ROIs) were drawn on TRAFF, trace DWI, and T2 using anatomical T2-weighted images and prostatectomy sections as the reference. Anatomical structures visible on TRAFF, trace DWI, T2, and T2weighted images were used to match whole mount prostatectomy sections to the imaging data sets. The ROI positions for all 50 patients are shown in Supporting Figs. S1–S50. One squared-shaped ROI (4.56  4.56  5.00 mm3 for TRAFF; 4.89  4.89  5.00 mm3 for low b value set DWI; 5.6  5.6  5.0 mm3 for high b value set DWI; 4.93  4.93  5.00 mm3 for T2) was placed in the center of the PCa area while the the same-sized ROI was placed in the peripheral zone (PZ) and central gland (CG) not containing PCa. The mean signal intensity of each ROI was fitted using a two-parameter monoexponential function (23) for RAFF and T2 data.

Thus, the purpose of the current study was to evaluate association of TRAFF, DWI-derived parameters, T2 relaxation time values with Gleason score, and the short-term repeatability of region of interest analysis. Differences in these quantitative values between PCa and normal tissues were estimated as well. Furthermore, we investigated cross-correlation of the parameters. METHODS

2132

Jambor et al.

The following models were applied to DWI data sets: 1) the monoexponential model (30): sðbÞ ¼ s0 ebADCm ;

[1]

2) the biexponential model (31): sðbÞ ¼ s0 ðfebDp þ ð1  f ÞebDf Þ;

[2]

and 3) the kurtosis model (32): 1 2

sðbÞ ¼ s0 eðbADCk þ6b

ADCk2 KÞ

;

[3]

where S(b) is the signal intensity for a particular b value, S0 is the signal intensity at b ¼ 0 s/mm2, ADCm is the diffusion coefficient of the monoexponential model, Dp is the intravoxel incoherent motion component of IVIM model (31) (“pseudo-diffusion”), f is the proportion of “pseudo-diffusion”, Df is the “fast” diffusion component, ADCk is the diffusion coefficient of the kurtosis model, and K is the kurtosis. Low b value DWI data sets were fitted using monoexponential and biexponential models while high b value DWI data sets were fitted using monoexponential and kurtosis models. ADCm values calculated using low b value DWI data sets are designed as ADCmL, whereas ADCmH is used for values calculated using high b value DWI data sets. In addition, the monoexponential model was fitted using b values higher or equal to 100 s/mm2. ADC values calculated using these b value distributions are designed as ADCmLEx and ADCmHEx for low and high b value DWI data sets, respectively. The biexponential model for low b value DWI data sets was fitted in a segmented fashion (33,34). Specifically, the first step consisted of fitting the monoexponential model to data with b values equal to or higher than 100 s/mm2 to obtain Df. Subsequently, the biexponential model was fitted using the Df parameter value from the first step. The fitting procedure for RAFF and T2 was performed using the “lsqnonlin” function in MATLAB (MathWorks Inc., Natick, Massachusetts, USA), whereas DWI modeling was performed using an in-house written Cþþ code using the Broyden–Fletcher–Goldfarb–Shanno algorithm (35) in dlib library (36). Multiple initialization values were used to limit the possible effect of local minima in the fitting procedure (27). In order to estimate short-term repeatability, intraclass correlation coefficient (ICC) values—specifically ICC(3,1)— were estimated (37). In ICC(3,1) calculation, differences between repeated measurements were compared with the differences between patients (37). Higher ICC(3,1) values mean better reliability and repeatability. If ICC(3,1) values are negative, it means that the random error as a source of variation is greater than the actual investigated target variable. Ninety-five percent confidence intervals for ICC(3,1) were calculated using 100,000 bootstrap samples. Moreover, the difference between two measurements (d), root mean squared difference (rmsd), and coefficient of repeatability (CR) values were calculated (38): sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n X rmsd ¼ di2 ðn  1Þ1 ; CR ¼ 1:96  rmsd i¼1

Ninety-five percent confidence intervals for CR were bootstrapped using 100,000 bootstrap samples. Normality of the fitted parameters was evaluated using the KolmogorovSmirnov test. Normally distributed values were compared using analysis of variance with a Bonferroni test, whereas the Kruskal–Wallis test and Dunns post hoc test were used for those parameters that did follow normal distribution. Receiver operating characteristic curve (ROC) analysis using 100,000 bootstrap samples was used to evaluate the ability of TRAFF, f, Dp, Df, ADCmL, ADCmH, ADCk, K, and T2 relaxation time to correctly classify PCa into Gleason score groups (n ¼ 3). Area under the curve (AUC) values were calculated using the trapezoid rule. Ninety-five percent confidence intervals for AUC values were calculated from 100,000 bootstrap samples. Spearman correlation coefficient (r) was calculated between the parameter (TRAFF, f, Dp, Df, ADCmL, ADCmH, ADCk, K, and T2) values and Gleason score groups (n ¼ 3). Ninety-five percent confidence intervals for r values were estimated using the Fisher transformation. In order to investigate crosscorrelation of the parameters, Pearson’s correlation coefficient was calculated. Two-sided P values were calculated; P < 0.05 was considered statistically significant. Statistical analyses were performed using code written in-house using MATLAB and/or GraphPad Prism, version 5.0 (GraphPad Software, San Diego, California, USA). The postprocessing codes as well as all MR sequences are freely available upon request. Histopathologic Analysis Whole mount prostatectomy sections were prepared as described previously (25–27) and were analyzed by one experienced genitourinary pathologist in conjunction with another staff pathologist who had at least 5 years of experience in general pathology (including genitourinary pathology). The extent of PCa and the Gleason score were based on a consensus of both pathologists in order to limit misinterpretation of the Gleason grade. The Gleason score was assigned as a combination of primary, secondary, and tertiary Gleason grade according to the 2005 International Society of Urological Pathology Modified Gleason Grading System (6). Tertiary Gleason grade was assigned in the presence of Gleason grade pattern higher than the primary and secondary, where the tertiary component was estimated visual to account for less than 5% of the tumor (39). Gleason scores were classified into three groups. The low Gleason score group consisted of PCa lesions with a Gleason grade 2 or 3 only; the intermediate Gleason score group consisted of PCa lesions with a Gleason grade 4 secondary or tertiary (without any Gleason grade 5); and the high Gleason score group consisted of PCa lesions with a Gleason grade 4 primary and/or Gleason grade 5 primary, secondary, or tertiary. RESULTS For 50 patients included in the final analysis, mean 6 standard deviation (range) values for age and serum prostate-specific antigen were 64 6 6 (45–73) y and 11.1 6 7.3 (1.3–30.0) ng/mL, respectively. The mean 6 standard deviation (range) time between MR examination and robotic-assisted laparoscopic prostatectomy was 14 6 12 (1–60) days. In total, 7, 20, and 23 patients had PCa

RAFF, DWI, and T2 Mapping of Prostate Cancer

2133

Table 1 Repeatability of ROI Level Analysis CR as Percentage of Median Value, %

ICC(3,1) (95% Confidence Interval) Parameter TRAFF ADCml ADCmlEx f Dp Df ADCmh ADCmhEx K ADCk

0.92 0.72 0.91 0.04 0.25 0.91 0.77 0.77 0.65 0.75

PZ (0.72, 0.98) (0.46, 0.97) (0.75, 0.99) (0.13, 0.47) (0.36, 0.72) (0.75, 0.99) (0.16, 0.93) (0.05, 0.93) (0.06, 0.89) (0.29, 0.90)

0.83 0.55 0.49 0.57 0.36 0.49 0.88 0.84 0.83 0.70

CG (0.61, 0.93) (0.09, 0.75) (0.05, 0.85) (0.08, 0.87) (0.73, 0.19) (0.05, 0.85) (0.63, 0.95) (0.51, 0.96) (0.53, 0.92) (0.23, 0.90)

lesions in the low, intermediate, and high risk Gleason score groups, respectively. In 11 patients (22%, 11/50), ROIs for PCa were placed in tumors located in CG, whereas the remaining 39 ROIs were placed in PZ tumors (Supporting Figs. S1–S50). Due to the presence of susceptibility artifacts, DWI data of the first or second MR examination of three (low b set) and four (high b set) patients were excluded from the final analysis. In one patient, T2 relaxation times were not measured. The median (95% confidence interval) of the B1 field on voxel level in prostate at the level of TRAFF measure-

0.82 0.71 0.60 0.17 0.16 0.60 0.91 0.91 0.82 0.85

PCa (0.49, 0.94) (0.35, 0.90) (0.31, 0.87) (0.32, 0.03) (0.27, 0.18) (0.31, 0.87) (0.69, 0.97) (0.77, 0.96) (0.53, 0.97) (0.30, 0.97)

PZ 4.9 6.7 3.6 167.6 76.1 3.6 5.3 5.2 5.1 4.2

CG 3.6 4.5 5.7 20.4 103.4 5.7 4.2 3.6 4.6 3.9

PCa 4.2 10.7 12.2 141.7 162.7 12.2 6.7 7.4 6.3 6.8

ments was 4% (2%–12%), expressed as a percentage of nominal flip angle (60 ). Repeatability of RAFF and DWI The quantitative parameters derived from RAFF (TRAFF), high b value DWI (ADCmh, K, ADCk), and T2 data sets demonstrated high repeatability with ICC(3,1) and CR as a percentage of the median value in the ranges of 0.65– 0.92 and 3.6%–6.8%, respectively. In contrast, f and Dp parameters had substantially lower ICC(3,1) and high CR

FIG. 1. Box plots show (A) TRAFF, (B) ADCml, (C) f, (D) Dp, (E) Df, (F) ADCmh, (G) K, (H) ADCk, and (I) T2 relaxation time values for PZ, CG, and PCa. Boxes represent the 25th to 75th percentiles, and the error bars extend from the minimal to maximal values. The lines within the boxes represent the median values.

2134

Jambor et al.

Table 2 AUC Values for Gleason Score Classification of PCa Tumors and Spearman Correlation Coefficient Values Gleason Score Group Parameter TRAFF ADCml ADCmlEx f Dp Df ADCmh ADCmhEx K ADCk T2

Low Versus Intermediate and High 0.92 (0.82, 0.98) 0.85 (0.72, 0.95) 0.80 (0.62, 0.95) 0.65 (0.51, 0.84) 0.53 (0.50, 0.78) 0.80 (0.62, 0.95) 0.76 (0.54, 0.96) 0.75 (0.54, 0.96) 0.74 (0.52, 0.96) 0.75 (0.53, 0.95) 0.85 (0.67, 0.98)

Low and Intermediate Versus High 0.83 (0.70, 0.94) 0.73 (0.57, 0.86) 0.75 (0.61, 0.89) 0.55 (0.50, 0.71) 0.53 (0.50, 0.69) 0.75 (0.61, 0.89) 0.68 (0.53, 0.82) 0.68 (0.53, 0.82) 0.66 (0.52, 0.80) 0.65 (0.52, 0.80) 0.54 (0.50, 0.70)

Spearman(r) 0.65 (0.79, 0.44)*** 0.47 (0.68, 0.21)*** 0.48 (0.69, 0.22)*** 0.01 (0.26, 0.28) 0.02 (0.31, 0.25) 0.48 (0.69, 0.22)*** 0.36 (0.61, 0.08)** 0.36 (0.62, 0.04)** 0.32 (0.03, 0.57)* 0.33 (0.57, 0.04)* 0.21 (0.46, 0.10)

Values are presented as the AUC (95% confidence interval). *P < 0.05; **P < 0.01; ***P < 0.001.

values. Specifically, ICC(3,1) and CR as a percentage of the median value were in the range of 0.36–0.57 and 20.4%–162.7% for f and Dp parameters, respectively. ADCml and Df parameters, which are the parameters derived from the low b value set, had ICC(3,1) and CR as a percentage of the median value in the ranges of 0.49– 0.91 and 4.2%–12.2%, respectively. The ICC(3,1) and CR values are summarized in Table 1.

PCa Detection and Characterization Statistically significant differences (P < 0.001) were present between TRAFF, ADCml, Df, ADCmh, K, ADCk, and T2 values of PCa and PZ (Fig. 1). Following Bonferroni correction (0.05/9), statistically significant differences between PCa and CG were present for TRAFF, ADCml, Df, ADCmh, K, and ADCk parameters (Fig. 1). The differences between PZ and CG reached a level of statistical

FIG. 2. Box plots show (A) TRAFF, (B) ADCml, (C) f, (D) Dp, (E) Df, (F) ADCmh, (G) K, (H) ADCk, and (I) T2 relaxation time values of low (3 þ 3), intermediate (3 þ 4), and high (>3 þ 4) Gleason score group tumors. Boxes represent the 25th to 75th percentiles, and the error bars extend from the minimal to maximal values. The lines within the boxes represent the median values.

RAFF, DWI, and T2 Mapping of Prostate Cancer

2135

FIG. 3. Parametric maps (no interpolation) of (A) TRAFF, (B) ADCml, (C) f, (D) Dp, (E) Df, (F) ADCmh, (G) K, (H) ADCk, and (I) T2 of a 62-yold patient with serum prostate-specific antigen values of 4.1 ng/mL. Decreased parameter values (red arrow) were present in the right peripheral zone on (A) TRAFF, (B) ADCmL, (E) Df, (F) ADCmH, (H) ADCk, (I) T2 parametric maps, whereas no clearly identifiable lesion was present on T2-weighted images (J). In the corresponding area, K parameter values were increased (G). PCa is outlined in green on the corresponding whole mount prostatectomy section (K). The Gleason score of the lesions was 4 þ 5. The parametric maps are scaled as follows: (A) TRAFF 0–185 ms, (B) ADCmL 0–3.0 mm2/ms, (C) f 0–0.5, (D) Dp 0–30.0 mm2/ms, (E) Df 0–3.0 mm2/ms, (F) ADCmH 0–3.0 mm2/ ms, (G) K 0–2.0, (H) ADCk 0–3.0 mm2/ms, and (J) T2 0–185 ms.

2136

Jambor et al.

FIG. 4. Cross-correlations of PCa parameters. Pearson correlation coefficient values are shown.

significance (P < 0.05/9) only for TRAFF and T2 parameters (Fig. 1). Differences between tumors belonging to low and high Gleason score groups were statistically significant following Bonferroni correction (P < 0.05/9) only for TRAFF (Fig. 2). Furthermore, RAFF outperformed DWI-derived parameters and T2 mapping in tumor classification based on AUC values (Table 2). The strongest correlations between the parameter values and Gleason score groups (n ¼ 3) were present for TRAFF, ADCmL, and Df with Spearman correlation coefficient (95% confidence interval) values of 0.65 (0.79, 0.44), 0.47 (0.66, 0.22), and 0.48 (0.67, 0.22), respectively (Table 3). Representative imaging finding on a voxel level for all parameters are shown in Figure 3. Cross-Correlation of the Parameters In PCa lesions (Fig. 4), TRAFF values demonstrated the highest Pearson correlation coefficient with T2 values (0.6710), whereas the corresponding values for the corre-

lation with DWI-derived parameters were in the range of 0.5380 (correlation with K was negative) to 0.6239. The corresponding values for PZ (Fig. 5) and CG (Fig. 6) were in the range of 0.6237–0.7452 and 0.3055–0.4953, respectively. Pearson correlation coefficient values of 0.6301– 0.9816 were present between different DWI-derived parameters. These values were lower in CG compared with PCa and PZ. DISCUSSION In this prospective study involving 50 patients with histologically confirmed PCa, we demonstrated that TRAFF and DWI-derived parameter values correlated with Gleason score based on whole mount prostatectomy sections. The RAFF method demonstrated the highest r value compared with TRAFF, f, Dp, Df, ADCmL, ADCmH, ADCk, K, and T2 parameters. However, it is also of clinical importance to correctly classify PCa lesions into low, intermediate, and high Gleason score groups. In the current study, ROC analysis was used to evaluate the potential of the fitted

RAFF, DWI, and T2 Mapping of Prostate Cancer

2137

FIG. 5. Cross-correlations of PZ parameters. Pearson correlation coefficient values are shown.

parameters to correctly classify tumors into Gleason score groups. Similarly to r values, RAFF outperformed DWIderived parameters in ROC analysis for PCa characterization. Sixteen patients underwent repeated MR examination to evaluate short-term repeatability of TRAFF and DWI-derived parameter values. TRAFF values demonstrated a repeatability comparable with the DWI parameters derived using the monoexponential model. Due to wide range of PCa aggressiveness, accurate estimating of Gleason score is of utmost clinical importance (5,8). Gleason score is a histopathological scoring system of PCa aggressiveness in biopsy and prostatectomy specimens (6) and was shown to be an independent predictor of tumor recurrence (40). PCa is still most commonly diagnosed by the means of systematic TRUS-guided biopsy. However, Gleason score based on cores of systematic TRUS-guided biopsy commonly underestimates true Gleason score detected in prostatectomy specimens (41–43). Gleason score based on prostatectomy samples should, ideally, be used for validation of an imaging technique’s potential to predict Gleason score. In the

current study, Gleason score was based on a consensus review of whole mount prostatectomy sections by two experienced pathologists. Several previous studies have demonstrated correlation of ADCm with Gleason score (12,16,44,45). The reported r values were in the range of 0.30/0.36 (12,44,45) to 0.60 (16,46). Similar to those studies, ADCm (low b set, ADCml; high b set, ADCmh) demonstrated significant correlation (P < 0.001) with Gleason score in the current study. A large proportion (46%, 23/ 50) of PCa patients had high Gleason score tumors in contrast to the previous studies. This fact could explain higher AUC and r values for PCa detection and characterization in the current study compared with previous reports. Furthermore, the mean signal intensity per ROIs was fitted in contrast to the use of statistical measures (eg, mean, median, percentile, kurtosis, etc) from voxelby-voxels fits. The true biological correlation of TRAFF values remains to be explored. Dipolar interactions and/or exchange relaxation pathways are the main contributors to RAFF

2138

Jambor et al.

FIG. 6. Cross-correlations of CG parameters. Pearson correlation coefficient values are shown.

signal (18,19,23,46). However, changes in TRAFF values between tissues are not only driven by relaxation, but water diffusion may have some contribution as well (23). Nevertheless, relaxation mechanisms of RAFF remain to be fully explored and are beyond the scope of the current study. In the current study, Pearson correlation coefficient values for the correlations of TRAFF with DWIderived parameters were higher in PZ compared with PCa. Moreover, TRAFF demonstrated higher AUC and r values compared with DWI-derived parameters. Thus, our findings suggest that changes in PCa related to different Gleason score are not adequately measured by DWI (using the imaging parameters and postprocessing as in the current study) and RAFF could provide valuable additional information. However, statistical methods such as principal component analysis should be applied to explore the added value of RAFF in addition to DWI. Moreover, the data quantification should ideally be fully automatic allowing reproducibility across different sites. Our study is limited by a relatively small number of patients. Only 11 tumors were located in CG, so we did

not perform a separate analysis for CG tumors. Correlation of whole mount prostatectomy section to in vivo imaging data sets is known to be challenging. In the current study, all of ROIs were placed by one research fellow. Thus, interreader variability was not evaluated. Development of fully automatic tools for ROI positioning and data extraction from multiparametric MRI is needed to allow high reproducibility of results (47,48). Dynamic contrast-enhanced MRI and proton MR spectroscopy, which are commonly a part of prostate multiparametric MRI (49), have not been performed in the current study. Comparison of RAFF with DCE-MRI and 1H-MRS remains to be performed. The accuracy of TRAFF measurements depends on homogeneity of the B1 field, which was evaluated in the current study using an actual flip angle imaging method (29). This potentially presents a challenge for future multi-institutional trials evaluating the role of RAFF in PCa imaging. However, the introduction of multichannel RF excitation MR systems led to improved B1 field homogeneity. Moreover, the prostate is a relatively small organ, and obtaining homogenous B1

RAFF, DWI, and T2 Mapping of Prostate Cancer

field across the whole prostate is less of a challenge than in larger organs such as the brain or liver. In conclusion, TRAFF, Df, ADCmL, ADCmH, ADCk, and K parameter values of PCa were significantly associated with Gleason score groups. RAFF demonstrated the repeatability of ROI analysis comparable with Df, ADCmL, and ADCmH, parameters. Relaxation values obtained using RAFF outperformed DWI-derived parameters and T2 relaxation time values in Gleason score classification of PCa tumors. Our findings suggest that RAFF can provide valuable information for PCa characterization. ACKNOWLEDGEMENTS We thank Jaakko Liippo (Turku University Hospital, Turku, Finland) for help in scanning the histological slides and Anitta Entonen (Turku University Hospital, Turku, Finland) for help with patient enrollment and assistance during MRI examinations.

2139

15.

16.

17.

18.

19.

20.

REFERENCES 1. Siegel R, Ma J, Zou Z, Jemal A. Cancer statistics, 2014. CA Cancer J Clin 2014;64:9–29. 2. Walsh PC, DeWeese TL, Eisenberger MA. Clinical practice. Localized prostate cancer. N Engl J Med 2007;357:2696–2705. 3. Draisma G, Boer R, Otto SJ, van der Cruijsen IW, Damhuis RA, Schr€ oder FH, de Koning HJ. Lead times and overdetection due to prostate-specific antigen screening: estimates from the European Randomized Study of Screening for Prostate Cancer. J Natl Cancer Inst 2003;95:868–878. 4. Ahmed HU, Akin O, Coleman JA, et al. Transatlantic Consensus Group on active surveillance and focal therapy for prostate cancer. BJU Int 2012;109:1636–1647. 5. Bill-Axelson A, Holmberg L, Filen F, et al; Scandinavian Prostate Cancer Group Study Number 4. Radical prostatectomy versus watchful waiting in localized prostate cancer: the Scandinavian prostate cancer group-4 randomized trial. J Natl Cancer Inst 2008;100: 1144–1154. 6. Epstein JI, Allsbrook WC Jr, Amin MB, Egevad LL. The 2005 International Society of Urological Pathology (ISUP) Consensus Conference on Gleason Grading of Prostatic Carcinoma. Am J Surg Pathol 2005;29:1228–1242. 7. Stephenson AJ, Scardino PT, Eastham JA, Bianco FJ Jr, Dotan ZA, Fearn PA, Kattan MW. Preoperative nomogram predicting the 10-year probability of prostate cancer recurrence after radical prostatectomy. J Natl Cancer Inst 2006;98:715–717. 8. Partin AW, Mangold LA, Lamm DM, Walsh PC, Epstein JI, Pearson JD. Contemporary update of prostate cancer staging nomograms (Partin Tables) for the new millennium. Urology 2001;58:843–848. 9. Hambrock T, Somford DM, Huisman HJ, van Oort IM, Witjes JA, Hulsbergen-van de Kaa CA, Scheenen T, Barentsz JO. Relationship between apparent diffusion coefficients at 3.0-T MR imaging and Gleason grade in peripheral zone prostate cancer. Radiology 2011; 259:453–461. 10. Vargas HA, Akin O, Franiel T, Mazaheri Y, Zheng J, Moskowitz C, Udo K, Eastham J, Hricak H. Diffusion-weighted endorectal MR imaging at 3 T for prostate cancer: tumor detection and assessment of aggressiveness. Radiology 2011;259:775–784. 11. Peng Y, Jiang Y, Antic T, Sethi I, Schmid-Tannwald C, Eggener S, Oto A. Apparent diffusion coefficient for prostate cancer imaging: impact of B values. AJR Am J Roentgenol 2014;202:W247–W253. 12. Donati OF, Mazaheri Y, Afaq A, Vargas HA, Zheng J, Moskowitz CS, Hricak H, Akin O. Prostate cancer aggressiveness: assessment with whole-lesion histogram analysis of the apparent diffusion coefficient. Radiology 2014;271:143–152. 13. Boesen L, Chabanova E, Logager V, Balslev I, Thomsen HS. Apparent diffusion coefficient ratio correlates significantly with prostate cancer Gleason score at final pathology. J Magn Reson Imaging 2015;42:446– 453. 14. Rosenkrantz AB, Triolo MJ, Melamed J, Rusinek H, Taneja SS, Deng FM. Whole-lesion apparent diffusion coefficient metrics as a marker

21.

22.

23.

24.

25.

26.

27.

28. 29.

30. 31. 32.

33.

34.

35.

of percentage Gleason 4 component within Gleason 7 prostate cancer at radical prostatectomy. J Magn Reson Imaging 2015;41:708–714 Tamada T, Sone T, Jo Y, Toshimitsu S, Yamashita T, Yamamoto A, Tanimoto D, Ito K. Apparent diffusion coefficient values in peripheral and transition zones of the prostate: comparison between normal and malignant prostatic tissues and correlation with histologic grade. J Magn Reson Imaging 2008;28:720–726. Turkbey B, Shah VP, Pang Y, et al. Is apparent diffusion coefficient associated with clinical risk scores for prostate cancers that are visible on 3-T MR images? Radiology 2011;258:488–495. Itou Y, Nakanishi K, Narumi Y, Nishizawa Y, Tsukuma H. Clinical utility of apparent diffusion coefficient (ADC) values in patients with prostate cancer: can ADC values contribute to assess the aggressiveness of prostate cancer? J Magn Reson Imaging 2011;33: 167–172. Liimatainen T, Sorce DJ, O’Connell R, Garwood M, Michaeli S. MRI contrast from relaxation along a fictitious field (RAFF). Magn Reson Med 2010;64:983–994. Liimatainen T, Hakkarainen H, Mangia S, Huttunen JM, Storino C, Idiyatullin D, Sorce D, Garwood M, Michaeli S. MRI contrasts in high rank rotating frames. Magn Reson Med 2015;73:254–262. Hakkarainen H, Sierra A, Mangia S, Garwood M, Michaeli S, Gr€ ohn O, Liimatainen T. MRI relaxation in the presence of fictitious fields correlates with myelin content in normal rat brain. Magn Reson Med 2016;75:161–168. Aronen HJ, Ramadan UA, Peltonen TK, Markkola AT, Tanttu JI, J€ a€ askel€ ainen J, H€ akkinen AM, Sepponen R. 3D spin-lock imaging of human gliomas. Magn Reson Imaging 1999;17:1001–1010. Ramadan UA, Markkola AT, Halavaara J, Tanttu J, Hakkinen AM, Aronen HJ. On- and off-resonance spin-lock MR imaging of normal human brain at 0.1 T: possibilities to modify image contrast. Magn Reson Imaging 1998;16:1191–1199. Liimatainen T, Sierra A, Hanson T, Sorce DJ, Yl€ a-Herttuala S, Garwood M, Michaeli S, Gr€ ohn O. Glioma cell density in a rat gene therapy model gauged by water relaxation rate along a fictitious magnetic field. Magn Reson Med 2012;67:269–277. Jambor I, Pesola M, Taimen P, Merisaari H, Bostr€ om P, Minn H, Liimatainen T, Aronen HJ. Rotating frame relaxation imaging of prostate cancer: repeatability, cancer detection and Gleason score prediction. Mag Res Med 2016;75:337–344. Jambor I, Merisaari H, Taimen P, Bostr€ om P, Minn H, Pesola M, Aronen HJ. Evaluation of different mathematical models for diffusion-weighted imaging of normal prostate and prostate cancer using high b-values: a repeatability study. Magn Reson Med 2015;73: 1988–1998. Merisaari H, Jambor I. Optimization of b-value distribution for four mathematical models of prostate cancer diffusion-weighted imaging using b values up to 2000 s/mm2: simulation and repeatability study. Magn Reson Med 2015;73:1954–1969. Toivonen J, Merisaari H, Pesola M, et al. Mathematical models for diffusion-weighted imaging of prostate cancer using b values up to 2000 s/mm2: correlation with Gleason score and repeatability of region of interest analysis. Magn Reson Med 2015;74:1116–1124. Pruessmann KP, Weiger M, Scheidegger MB, Boesiger P. SENSE: sensitivity encoding for fast MRI. Magn Reson Med 1999;42:952–962. Yarnykh VL. Actual flip-angle imaging in the pulsed steady state: a method for rapid three-dimensional mapping of the transmitted radiofrequency field. Magn Reson Med 2007;57:192–200. Stejskal EO. Use of spin echoes in a pulsed magnetic-field gradient to study restricted diffusion and flow. J Chem Phys 1965;43:3597–3603. Le Bihan D, Turner R. The capillary network: a link between IVIM and classical perfusion. Magn Reson Med 1992;27:171–178. Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K. Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means of magnetic resonance imaging. Magn Reson Med 2005;53: 1432–1440. Marzi S, Piludu F, Vidiri A. Assessment of diffusion parameters by intravoxel incoherent motion MRI in head and neck squamous cell carcinoma. NMR Biomed 2013;26:1806–1814. Cho GY, Moy L, Zhang JL, Baete S, Lattanzi R, Moccaldi M, Babb JS, Kim S, Sodickson DK, Sigmund EE. Comparison of fitting methods and b-value sampling strategies for intravoxel incoherent motion in breast cancer. Magn Reson Med 2015;74:1077–1085. Shanno, David F. On Broyden-Fletcher-Goldfarb-Shanno method. J Optim Theory Appl 46.1;1985:87–94.

2140 36. King DE. Dlib-ml: a machine learning toolkit. J Mach Learn Res 2009; 10:1755–1758. 37. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater reliability. Psychol Bull 1979;86:420–428. 38. Bland JM, Altman DG. Measuring agreement in method comparison studies. Stat Methods Med Res 1999;8:135–160. 39. Epstein JI. An update of the Gleason grading system. J Urol 2010;183: 433–440. 40. D’Amico AV, Whittington R, Malkowicz SB, Fondurulia J, Chen MH, Tomaszewski JE, Wein A. The combination of preoperative prostate specific antigen and postoperative pathological findings to predict prostate specific antigen outcome in clinically localized prostate cancer. J Urol 1998;160:2096–2101. 41. Nepple KG, Wahls TL, Hillis SL, Joudi FN. Gleason score and laterality concordance between prostate biopsy and prostatectomy specimens. Int Braz J Urol 2009;35:559–564. 42. Steinberg DM, Sauvageot J, Piantadosi S, Epstein JI. Correlation of prostate needle biopsy and radical prostatectomy Gleason grade in academic and community settings. Am J Surg Pathol 1997;21:566–576. 43. Rajinikanth A, Manoharan M, Soloway CT, Civantos FJ, Soloway MS. Trends in Gleason score: concordance between biopsy and prostatectomy over 15 years. Urology 2008;72:177–182. 44. Oto A, Yang C, Kayhan A, Tretiakova M, Antic T, Schmid-Tannwald C, Eggener S, Karczmar GS, Stadler WM. Diffusion-weighted and dynamic contrast-enhanced MRI of prostate cancer: correlation of quantitative MR parameters with Gleason score and tumor angiogenesis. AJR Am J Roentgenol 2011;197:1382–1390. 45. Peng Y, Jiang Y, Yang C, Brown JB, Antic T, Sethi I, SchmidTannwald C, Giger ML, Eggener SE, Oto A. Quantitative analysis of multiparametric prostate MR images: differentiation between prostate cancer and normal tissue and correlation with Gleason score—a computer-aided diagnosis development study. Radiology 2013;267: 787–796. 46. Liimatainen T, Mangia S, Ling W, Ellermann J, Sorce DJ, Garwood M, Michaeli S. Relaxation dispersion in MRI induced by fictitious magnetic fields. J Magn Reson 2011;209:269–276. 47. Ginsburg SB, Viswanath SE, Bloch BN, Rofsky NM, Genega EM, Lenkinski RE, Madabhushi A. Novel PCA-VIP scheme for ranking MRI protocols and identifying computer-extracted MRI measurements associated with central gland and peripheral zone prostate tumors. J Magn Reson Imaging 2015;41:1383–1393.

Jambor et al. 48. Stember JN, Deng FM, Taneja SS, Rosenkrantz AB. Pilot study of a novel tool for input-free automated identification of transition zone prostate tumors using T2- and diffusion-weighted signal and textural features. J Magn Reson Imaging 2014;40:301–305. 49. Jambor I, Kahkonen E, Taimen P, Merisaari H, Saunavaara J, Alanen K, Obsitnik B, Minn H, Lehotska V, Aronen HJ. Prebiopsy multiparametric 3T prostate MRI in patients with elevated PSA, normal digital rectal examination, and no previous biopsy. J Magn Reson Imaging 2015;41:1394–1404.

SUPPORTING INFORMATION Additional Supporting Information may be found in the online version of this article. Table S1. Patient Characteristics Figures S1–S16. The positions of ROIs placed in the PCa (red color), PZ (green color), and CG (blue color) are shown for RAFF [(A) 90 ms duration of the weighing pulse train], low b set DWI [(C) trace image of b value 0 s/ mm2], high b set DWI [(I) trace image of b value 0 s/mm2], and T2 [(K) echo time of 100 ms]. The images without overlaid ROIs for RAFF, DWI, and T2 are shown in panels B (90 ms duration of the weighing pulse train), D (trace image of b value 500 s/mm2), F (trace image of b value 2000 s/mm2), and L (echo time of 100 ms), respectively. The images of the second repeated scan for RAFF, low b set DWI, and high b set DWI are shown in panels E and F, G and H, and M and N, respectively. The corresponding T2-weighted image and whole mount prostatectomy section are shown in panels O and P, respectively. Prostate cancer is outlined in green/blue on the corresponding whole mount prostatectomy section. Figures S17–S50. The positions of ROIs placed in the PCa (red color), PZ (green color), and CG (blue color) are shown for RAFF [(A) 90 ms duration of the weighing pulse train], low b set DWI [(C) trace image of b value 0 s/ mm2], high b set DWI [(E) trace image of b value 0 s/mm2], and T2 [(G) echo time of 100 ms]. The images without overlaid ROIs for RAFF, DWI, and T2 are shown in panels B (90 ms duration of the weighing pulse train), D (trace image of b value 500 s/mm2), F (trace image of b value 2000 s/ mm2), and H (echo time of 100 ms), respectively. The corresponding T2weighted image and whole mount prostatectomy section are shown in panels I and J, respectively. PCa is outlined in green/blue on the corresponding whole mount prostatectomy section.