UROGENITAL

Computed diffusion-weighted imaging of the prostate at 3 T:impact on image quality and tumour detection

Andrew B. Rosenkrantz & Hersh Chandarana &

Nicole Hindman & Fang-Ming Deng & James S. Babb &

Samir S. Taneja & Christian Geppert

Received: 27 March 2013 /Revised: 23 April 2013 /Accepted: 6 May 2013# European Society of Radiology 2013

AbstractObjectives To investigate the impact of prostate computeddiffusion-weighted imaging (DWI) on image quality andtumour detection.Methods Forty-nine patients underwent 3-T magnetic reso-nance imaging using a pelvic phased-array coil before prosta-tectomy, including DWI with b values of 50 and 1,000 s/mm2.Computed DW images with b value 1,500 s/mm2 were gen-erated from the lower b-value images. Directly acquired b-1,500 DW images were obtained in 39 patients. Two radiolo-gists independently assessed DWI for image quality measuresand location of the dominant lesion. A third radiologist mea-sured tumour-to-peripheral-zone (PZ) contrast. Pathologicalfindings from prostatectomy served as the reference standard.Results Direct and computed b-1,500 DWI showed bettersuppression of benign prostate tissue than direct b-1,000 DWIfor both readers (P≤0.024). However, computed b-1,500 DWI

showed less distortion and ghosting than direct b-1,000 anddirect b-1,500 DWI for both readers (P≤0.067). Direct andcomputed b-1,500 images showed better sensitivity and posi-tive predictive value (PPV) for tumour detection than direct b-1,000 images for both readers (P≤0.062), with no difference insensitivity or PPV between direct and computed b-1,500 im-ages (P≥0.180). Tumour-to-PZ contrast was greater on com-puted b-1,500 than on either direct DWI set (P<0.001).Conclusion Computed DWI of the prostate using b value≥1,000 s/mm2 improves image quality and tumour detectioncompared with acquired standard b-value images.Key Points•Diffusion weighted MRI is increasingly used for diagnosingand assessing prostate carcinoma.

• Prostate computed DWI can extrapolate high b-value im-ages from lower b values.

• Computed DWI provides greater suppression of benigntissue than lower b-value images.

• Computed DWI provides less distortion and artefacts thanimages using same b value.

• Computed DWI provides better diagnostic performancethan lower b-value images.

Keywords Prostate cancer . MRI . Diffusion-weightedimaging . b value . Prostatectomy

Introduction

Diffusion-weighted imaging (DWI) is a critical componentof multi-parametric prostate MRI protocols [1, 2] that hasbeen shown to substantially improve tumour detection andlocalisation [3–5], planning of targeted prostate biopsy [6]and formation of accurate risk estimates [7, 8]. Current MRsystems are able to perform DWI in a reliable and user-friendly fashion, facilitating the widespread adoption of this

A. B. Rosenkrantz :H. Chandarana :N. Hindman : J. S. BabbDepartment of Radiology, NYU Langone Medical Center,550 First Avenue, New York, NY 10016, USA

F.<M. DengDepartment of Pathology, NYU Langone Medical Center,550 First Avenue, New York, NY 10016, USA

S. S. TanejaDivision of Urologic Oncology, Department of Urology,NYU Langone Medical Center, 550 First Avenue,New York, NY 10016, USA

C. GeppertSiemens Medical Solutions MR R&D, 660 First Avenue,New York, NY 10016, USA

A. B. Rosenkrantz (*)Department of Radiology, Center for Biomedical Imaging,NYU School of Medicine, 660 First Avenue,New York, NY 10016, USAe-mail: Andrew.Rosenkrantz@nyumc.org

Eur RadiolDOI 10.1007/s00330-013-2917-8

technique. While the apparent diffusion coefficient (ADC)map serves as the primary data-set for tumour assessmentwhen evaluating DWI of the prostate, recent guidelines onthe interpretation of multi-parametric MRI also indicate aneed to evaluate the acquired DW images for lesions suspi-cious for tumour, identified as foci of increased signal inten-sity (SI) relative to surrounding benign prostate tissue [1].This contrast on acquired DW images between tumours andbenign prostate tissue can be enhanced by increasing thediffusion weighting of the sequence, as measured by the bvalue [9].

While numerous studies of DWI of the prostate havereported success using a maximal b value in the range of700-1,000 s/mm2, [10–13], benign prostate tissue often re-mains incompletely suppressed and shows mild SI at such bvalues. While the use of b-values higher than 1,000 s/mm2

remains controversial [14, 15], a number of recent studieshave shown improved tumour localisation when using great-er b values, ranging from 1,500 to 2,000 s/mm2, [16–19],attributable to greater suppression of benign prostate andsubsequent improved contrast between benign and malig-nant regions, compared with lower b values. However, ac-quisition of such high b values can be difficult owing tolowering of signal-to-noise ratio (SNR) and increasing imagedistortion and artefacts that occur at higher b values [20]. Theextent to which such issues will pose a challenge depends onthe choice of device, software platform, coil arrangementand a particular diffusion sequence being used, and in someinstances can create a clinical challenge in the direct acqui-sition of very high b-value images.

Computed DWI is a recently described technique in whichDW images using very high b values can be mathematicallyderived from lower b-value images, rather than directly ac-quired [21]. In this approach, the ADCmap is calculated fromthe directly acquired b values in standard fashion, and thenused in turn to extrapolate projected SI at non-acquired bvalues based upon a standard mono-exponential fit. Thisprocess can be performed on a voxel-by-voxel basis to con-struct a complete computed DW image set for higher, non-acquired, b values. The advantages of this process are twofold.First, the enhanced image contrast offered by very high bvalues can be attained while avoiding the distortion and othertechnical pitfalls encountered at such b values, thereby in-creasing the feasibility of incorporation of very high b-valueimages into clinical protocols. Second, when aiming to ac-quire very high b values in addition to inclusion of standardhigh b-value images in the range of 700-1,000 s/mm2, thecomputed DW image set is generated without any additionalimaging time via the post-processing of images obtained usingthe standard b values.

A previous study demonstrated the value of computedDWI for detection of metastatic lesions when included in awhole-body protocol in ten oncology patients [21]. Given

incomplete suppression of benign prostate at b values ofaround 1,000 s/mm2, it would seem reasonable that comput-ed DWI would also be of value for localisation of primarytumours within the prostate. Therefore, the purpose of ourstudy was to investigate the impact of computed DWI of theprostate on image quality and tumour detection, using path-ological findings from radical prostatectomy as the referencestandard.

Materials and methods

Patients

This retrospective study was U.S. Health Insurance Portabil-ity and Accountability Act (HIPAA) compliant and approvedby our institutional review board with a waiver of the re-quirement for written informed consent. We identified 50consecutive men who underwent multi-parametric prostateMRI that included computed DWI before radical prostatec-tomy for prostate cancer. One patient was subsequentlyexcluded owing to previous treatment with external-beamradiation and hormonal therapy. This left a final cohort of 49men (mean age 62±7 years; median age 63 years; age range44-75 years). The mean delay between MRI and prostatec-tomy was 32±33 days (median 14 days; range 4-165 days).The mean pre-operative prostate-specific antigen (PSA) was7.8±8.9 ng/ml (median 5.3 ng/ml; range 1.2-57.0 ng/ml). In41 patients, the MRI was performed for staging and riskstratification following a previous prostate biopsy positivefor cancer; in three patients, the MRI was performed follow-ing a previous negative prostate biopsy; in five patients, theMRI was performed before an initial prostate biopsy in thesetting of an abnormal PSA.

MRI technique

All patients underwent MRI using a 12-channel pelvicphased-array coil on a single 3.0-T system (MAGNETOMTrio; Siemens Healthcare, Erlangen, Germany). Examinationsincluded an axial turbo spin-echo (TSE) T2-weighted se-quence of the prostate and seminal vesicles (TR/TE4,960/105 ms; slice thickness [ST] 3 mm, no inter-slice gap;field of view [FOV] 180×180 mm; matrix 256×256; parallelimaging factor of 2; 3 signal averages), an axial TSE T1-weighted sequence of the prostate and seminal vesicles(TR/TE 683/10 ms; ST 3 mm; no inter-slice gap; FOV 180×180 mm; matrix 192×116; parallel imaging factor of 2; 2signal averages), and an axial fat-suppressed single-shotecho-planar DW sequence using tri-directional diffusion-sensitising gradients (TR/TE 4,100/86 ms; ST 3 mm; nointer-slice gap; FOV 200×200 mm; matrix, 100×100; parallelimaging factor of 2; 10 signal averages; b values of 50 and

Eur Radiol

1,000 s/mm2). Thirty-eight of the examinations also includeda DW acquisition using a single b value of 1,500 s/mm2,keeping all other imaging parameters identical, aside from aminimal increase in TE to 90 ms. In all 49 cases, ADC mapswere calculated from the directly acquired DW images using bvalues of 50 and 1,000 s/mm2, which were used to produce anadditional computed DW image set at a b value of1,500 s/mm2 based upon a standard mono-exponential fit.

Reference standard

A pathologist with 7 years of experience in uropathology(F.M.D.) and a radiologist with 4 years of experience inprostate MRI (A.B.R.) jointly reviewed step-section histo-logical slides from the radical prostatectomy specimens tolocalise the index lesion for each patient [22–24], which wasconsidered to represent the largest individual tumour focus.The pathologist recorded the Gleason score of each indexlesion. The pathologist and radiologist then recorded thelocation of the index lesion in three planes (right vs left;anterior vs posterior; base vs midgland vs apex) using stan-dard anatomical landmarks, such as the prostatic capsule,junction of the peripheral and transition zones, urethra andejaculatory ducts.

Image analysis

The previously noted radiologist reviewed all DW image setsfor each patient (directly acquired b-1,000; when available,directly acquired b-1,500; computed b-1,500), aware of theresults of the histopathological assessment. The radiologistidentified the index lesions on DWI as a focal area of in-creased signal in a region corresponding to the pathologicallocation of the index lesion. Then, a region of interest (ROI)was manually traced just within the outer margin of theidentified abnormality on each image set. If no lesion on agiven image set could be identified, then an ROI was man-ually traced to correspond to the location of the identifiedlesion on the other image sets. Eight patients (five withGleason score 3+3 tumour and three with Gleason score3+4 tumour) in whom the lesion could not be identified onany of the DW image sets were excluded from this portion ofthe analysis. For the remaining 41 patients, a small ROImeasuring 0.3–0.4 cm2 was also placed in a nearby regionof the PZ showing normal homogeneous signal intensity onall of the image sets as well as on T2-weighted images. Themean value of each ROI was recorded, and for each imageset, the tumour-to-PZ contrast was calculated as the absolutevalue of (SIPZ−SItumour)/(SIPZ+SItumour) [25, 26], whichyields a larger value for greater relative differences betweentumours and benign prostate.

Two other fellowship-trained radiologists (readers 1[N.H.] and 2 [H.C.], with 6 and 5 years of experience in

abdominal imaging, respectively) independently reviewedthe DW image sets. For this purpose, the three DW imagesets for each patient were assigned unique, anonymous iden-tifiers, and all DW image sets for all cases were thenreviewed in random order, such that the two readers wereunaware of which patient and which DW image set werebeing reviewed. The radiologists knew that all patients in thestudy had undergone radical prostatectomy, but were other-wise unaware of all clinical and pathological details. The tworeaders identified the single largest focus in the prostateshowing visually increased SI on the DW image set underreview and recorded its location in three dimensions usingthe same approach used to indicate the pathological locationof the index lesion. No lesion was recorded if the prostateshowed homogeneous SI on a given image set. Each indi-vidual DW image set was reviewed in conjunction with theaxial T1-weighted and axial T2-weighted images for thatcase to guide anatomical localisation of findings on DWimages; abnormalities identified only on T2-weighted imag-ing were not recorded. For each identified lesion, the readersrecorded the lesion’s subjective visual conspicuity comparedwith the surrounding normal prostate tissue (scale of 1–5;where 5 = maximal visual conspicuity).

When reviewing each DW image set, the two readers alsosubjectively evaluated each image set on a scale of 1–5 scale(where 5 = highest quality) in the following categories:suppression of benign prostate, absence of distortion of theprostate, absence of ghosting and overall image quality.

Imaging review was performed using a commercial work-station (iSite; Philips Healthcare, Foster City, CA, USA).

Statistical analysis

Exact Wilcoxon signed rank tests were used to compare DWimage sets in terms of tumour-to-PZ contrast as well as thereader assessments of suppression of benign prostate, distor-tion, ghosting and overall image quality. ExactMcNemar testswere used to compare DW image sets in terms of sensitivityfor the patient’s index lesion based on radical prostatectomyfindings. The lesion identified by the reader had to match thepathological location of the index lesion using the describedlocalisation scheme in order to constitute a true-positive result.Data from the 11 patients in whom direct b-1,500 images werenot acquired were omitted from comparisons involving thatimage set in order to avoid confounding the comparisons withbetween-subject differences. Lesion visual conspicuity amongtrue-positive lesions, as well as positive predictive value(PPV) for the index lesion, comprised data for each pair ofimage sets for each reader from overlapping but unequal sub-sets of patients. Thus, logistic regression for correlated datawas used to compare image sets in terms of PPV, while mixedmodel analysis of variance (ANOVA) was used to compareimage sets in terms of visual conspicuity among true-positive

Eur Radiol

lesions. In particular, generalised estimating equations (GEE)based on binary logistic regression were used to model PPVasa function of image sets. For the GEE and mixed modelanalyses, a compound symmetric correlation structure wasused to account for statistical dependencies among resultsfrom the same subject. Given that results from ANOVAexhibited strong agreement with those from pair-wiseWilcoxon tests to compare image sets when both analyseswere restricted to patients providing data for all three imagesets, and given that the assumptions underlying the Wilcoxontest are valid, it was considered appropriate to extend theANOVA analyses to include all available data. All P valuesare two-sided and considered statistically significant at P<0.05. Analysis was performed using SAS software (version9.0; SAS Institute, Cary, NC, USA).

Results

Subjective image quality

Table 1 shows results related to subjective image quality, asassessed by the two readers. Compared with direct b-1,000DW images, direct b-1,500 DW images and computed b-1,500 DW images had significantly better suppression ofbenign prostate for both readers (P≤0.024) and significantlybetter overall image quality for reader 1 (P≤0.018). In addi-tion, compared with direct b-1,000 DW images, computed b-1,500 images had significantly less distortion for both readers(P≤0.039) and less ghosting for reader 1 (P=0.001), and non-significant trends towards less ghosting and better overallimage quality for reader 2 (P≤0.075). Compared with directb-1,500 images, computed b-1,500 images had significantlyless distortion for both readers (P≤0.026), less ghosting forreader 2 (P=0.041) and overall image quality for reader 1 (P=0.018), as well as a non-significant trend towards less ghostingfor reader 2 (P=0.059). No other comparison among the threeDW image sets was significant for either reader (P≥0.104).

Tumour detection

The distribution of the Gleason score of the index lesion inthe 49 patients was as follows: 3+3 (n=18); 3+4 (n=17); 4+3 (n=8); 4+4 (n=1); 4+5 (n=2); 5+4 (n=2); 3+5 (n=1).Table 2 shows findings related to tumour detection withrespect to the index lesion. For direct b-1,000 images, directb-1,500 images and computed b-1,500 images, sensitivityfor all tumours was 46.9 %, 65.8 % and 69.4 % forreader 1 and 46.9 %, 60.5 % and 67.3 % for reader 2;sensitivity for tumours with Gleason score of 3+4 orgreater was 54.8 %, 76.9 % and 80.6 % for reader 1,and 58.1 %, 69.2 % and 80.6 % for reader 2; and PPVfor tumour detection was 59.0 %, 66.7 % and 70.8 % forreader 1, and 60.5 %, 71.9 % and 78.6 %, for reader 2.The greater sensitivity for all tumours or for Gleason 3+4 or greater tumours, as well as the greater PPV fortumours, for both direct and computed b-1,500 imagesin comparison with direct b-1,000 images was statistical-ly significant for both readers (P≤0.030), aside from thegreater sensitivity for high-grade lesions for direct b-1,500 images than for direct b-1,000 images for reader2 and the greater PPV for direct b-1,500 images than fordirect b-1,000 images for reader 1, both of whichshowed non-significant trends (P≤0.083). None of thedifferences between direct and computed b-1,500 imageswas statistically significant (P≥0.180).

Compared with direct b-1,000 images, there was signif-icantly greater conspicuity of the dominant lesion amongtrue-positive lesions for both direct and computed b-1,500images for reader 1 (P<0.001), as well as a non-significanttrend towards better conspicuity of the dominant lesion forcomputed b-1,500 images than for direct b-1,000 imagesfor reader 2 (P=0.093). No other difference in visualconspicuity between image sets was significant for eitherreader (P≥0.289).

Quantitative tumour-to-PZ contrast was significantly great-er on computed b-1,500 images than on either of the direct

Table 1 Comparison of subjective image quality between diffusion-weighted image sets

Parameter Reader Direct b-1,000 (A) Direct b-1,500 (B) Computed b-1,500 (C) P (A vs B) P (A vs C) P (B vs C)

Suppression of benign prostate 1 3.4±0.6 3.9±0.7 4.2±0.7 0.024* <0.001* 0.104

2 2.9±0.9 3.3±0.9 3.5±0.8 0.002* 0.016* 0.140

Absence of distortion 1 3.9±0.9 4.0±0.8 4.5±0.6 0.360 <0.001* 0.008*

2 3.7±0.8 3.7±0.7 4.0±0.8 0.660 0.039* 0.026*

Absence of ghosting 1 4.5±0.6 4.7±0.5 4.9±0.3 0.173 0.001* 0.059

2 3.9±0.7 3.9±0.7 4.2±0.8 0.560 0.067 0.041*

Overall image quality 1 3.7±0.7 4.1±0.7 4.5±0.7 0.018* <0.001* 0.018*

2 3.1±0.7 3.3±0.7 3.3±0.9 0.212 0.075 0.705

* P <0.05, statistically significant

Eur Radiol

image sets (P<0.001), and significantly greater on direct b-1,500 images than on direct b-1,000 images (P<0.001).

Representative cases are shown in Figs. 1 and 2.

Discussion

To date, there remains a lack of consensus regarding theoptimal b value for DW imaging of the prostate, in view ofboth the advantages and challenges presented by increasingb values. On the one hand, a large b value reflects strongerdiffusion-weighting, thereby suppressing benign tissue and

allowing for greater conspicuity of tumours [16]. On theother hand, implementation of the larger b value can betechnically challenging owing to greater artefacts resultingfrom eddy current distortions [21]. In this study, we havedemonstrated that the computed DWI may provide a solutionto this challenge by achieving the background suppression ofvery high b-value DW images, without encountering theartefacts that may be associated with these images. Indeed,all four subjective quality measures in our study (backgroundsuppression, distortion, ghosting and overall image quality)received better scores on computed b-1,500 images than ondirect b-1,000 images for two independent readers at a

Table 2 Comparison of measures related to tumour detection between diffusion-weighted sequences

Parameter Reader Direct b-1,000(A)

Direct b-1,500(B)

Computed b-1,500(C)

P (A vs B) P (A vs C) P (B vs C)

Sensitivity (all tumours) 1 46.9 % (23/49) 65.8 % (25/38) 69.4 % (34/49) 0.020* <0.001* 0.739

2 46.9 % (23/49) 60.5 % (23/38) 67.3 % (33/49) 0.025* 0.004* 0.180

Sensitivity (Gleason score ≥3+4tumours)

1 54.8 % (17/31) 76.9 % (20/26) 80.6 % (25/31) 0.014* 0.005* 1.0

2 58.1 % (18/31) 69.2 % (18/26) 80.6 % (25/31) 0.083* 0.020* 0.180

PPV for tumour detection 1 59.0 % (23/39) 66.7 % (24/36) 70.8 % (34/48) 0.062 0.007* 0.561

2 60.5 % (23/38) 71.9 % (23/32) 78.6 % (33/42) 0.030* 0.009* 0.278

Conspicuity of the dominant lesion 1 3.6±0.8 4.5±0.7 4.6±0.7 <0.001* <0.001* 0.723

2 3.9±0.9 4.4±1.0 4.2±1.0 0.093 0.289 0.432

Tumour-to-PZ contrast - 0.22±0.11 0.32±0.10 0.37±0.10 <0.001* <0.001* <0.001*

*P <0.05, statistically significant

Fig. 1 A 55-year-old man witha Gleason 4+3 tumour in theright anterior midgland transitionzone identified on radicalprostatectomy. a Axial turbospin-echo T2-weighted imageshows an ill-defined region ofhomogeneous decreased T2signal within the right transitionzone (arrow). b Direct b-1,000DW image fails to demonstratethe tumour as showing increasedsignal intensity. c Direct b-1,500DW image demonstratesincreased signal intensity withinthe tumour (arrow). d Computedb-1,500 DW image alsodemonstrates increased signalintensity within the tumour(arrow). In addition, there is lessdistortion of the computed b-1,500 image than of the direct b-1,500 image. All DW image setsare displayed using an identicalwindow and level

Eur Radiol

significance level of 0.075 or less. Furthermore, the comput-ed b-1,500 images received better scores in terms of distor-tion and ghosting than direct b-1,500 images, attributed tothe fact that the computed b-1,500 images are derived fromlower b-value images. Thus, globally across the varioussubjective quality measures, the computed b-1,500 imagesachieve the highest quality among the three DW image sets.

The better image quality for computed DW images trans-lated into improved diagnostic performance, as both readersachieved significantly better sensitivity and PPV for comput-ed b-1,500 DW images than for direct b-1,000 DW images.Although there was also higher sensitivity and PPV for com-puted b-1,500 than for direct b-1,500 DW images for bothreaders, these differences did not approach statistical signifi-cance. This greater diagnostic performance is supported by thehighest tumour-to-PZ contrast being observed on the direct b-1,500 DW images. While the readers’ sensitivity for both b-1,500 DW images remained under 70 % when including low-grade tumours, we note that a past study observed some low-grade tumours to be histologically sparse, comprising a sub-stantial fraction of intermixed benign tissue, thereby being“invisible” on DWI and placing limits on the sensitivity thatmay be achieved for tumour detection [27]. Lastly, past stud-ies of the role of DWI for prostate cancer localisation havelargely focused on the ADC map as the primary diagnosticsequence when interpreting DWI of the prostate [3, 4], givenrelatively less tumour conspicuity on standard DW images

[28]. Although warranting confirmation in other studies, ourfindings suggest that the use of computed DW images mayfacilitate greater incorporation of the high b-value DW imagesinto clinical interpretation schemes.

On one level, computed DWI can enhance the feasibility ofDWI using very high b values. Depending upon the gradientperformance, coil design and software platform, DWI with bvalues greater than 1,000 s/mm2may otherwise be prohibitive inclinical practice. However, even in settings in which very highb-value DW images can be reasonably acquired, our data indi-cate that computed DWI still remains of additive value given thereduced distortion compared with direct b-1,500 DW images.Furthermore, even if the direct and acquired images were tohave comparable quality and diagnostic performance, the com-puted DW images require no additional acquisition time on topof the imaging time used to obtain standard b-value images. Inaddition to these benefits, on some systems computed DWI cancurrently be applied in a straightforward fashion, with automaticgeneration of the derived high b-value images by the MRconsole at the time of image acquisition. Thus, when available,we suggest that it may be worthwhile incorporating the gener-ation and interpretation of computed DWI into standard clinicalmulti-parametric prostate MRI protocols.

To our knowledge, our findings have not been previouslyreported in the peer-reviewed literature. A small number ofstudies have shown improved diagnostic performance of di-rect b-1,500 or direct b-2,000 DW images in comparison with

Fig. 2 A 71-year-old man witha Gleason 3+4 tumour in theright posterior midglandperipheral zone identified onradical prostatectomy. a Axialturbo spin-echo T2-weightedimage shows crescenticsubcapsular region ofhomogeneous decreased T2signal within the posterior rightperipheral zone (arrow). bDirectb-1,000 DW image shows mildincreased signal intensity in theregion of the tumour (arrow). cDirect b-1,500 DW imagedemonstrates substantialincreased signal intensity withinthe tumour (arrow). d Computedb-1,500 DW image alsodemonstrates substantialincreased signal intensity withinthe tumour (arrow). All DWimage sets are displayed using anidentical window and level

Eur Radiol

direct b-1,000 DW images [16–18]. In addition, the originalreport of computed DWI by Blackledge et al. [21] demon-strated greater conspicuity of a single primary prostate tumouron an acquired b-2,000 DW image than on a direct b-1,000DWimage; otherwise, this previous report describes the utilityof computed DWI for the detection of metastatic sites. In viewof the paucity of data on this topic, we would encourageadditional studies to validate our observations regarding thebenefit of computed DWI for the detection of prostate cancer.

Several limitations of our study warrant a mention. Firstly,we only assessed computed DW images generated at a singlehigh b value (1,500 s/mm2). We selected this b-value in partbecause we wished to compare the computed DW imageswith direct DW images obtained at an identical b-value, andwe were able to obtain direct DW images at a b value of1,500 s/mm2 with otherwise comparable acquisition param-eters and while maintaining reasonable SNR. Following ourinitial study comparing direct and acquired DW images at anidentical b value of 1,500 s/mm2, future studies may assessthe utility of computed DW images using even higher bvalues of at least 2,000 s/mm2. Secondly, imaging wasperformed using a pelvic phased-array coil, such that theimpact of computed DWI when performing prostate MRIusing an endorectal coil is unknown. Given the greater imagedistortion and susceptibility artefacts that can occur whenusing an endorectal coil, it is possible that computed DWImight provide an even larger benefit when such a coil isused, although this requires evaluation in separate studies.Thirdly, we failed to observe an expected increase in distor-tion and ghosting on direct b-1,500 images in comparisonwith the direct b-1,000 images. We speculate that this mayrelate to the greater background suppression on the direct b-1,500 images leading to decreased conspicuity of such arte-facts. Fourthly, we did not evaluate ADC maps as part of thisstudy; however, the ADC maps corresponding to the stan-dard b-value images are used to generate the computed DWimages and do not themselves vary with the use of thistechnique. Fifthly, our analysis was based on an assessmentfor the index lesion, rather than for all tumour foci. However,data in support of the concept of the index lesion continue toemerge [22–24, 29, 30], and urologists are increasingly usingthis concept to support decisions regarding prognosticcounselling and selection of targeted treatments such as focalablative therapy. Finally, we acknowledge that one benefit ofthe direct acquisition of DW images with b values greaterthan 1,000 s/mm2 is to allow for computation of more ad-vanced non-Gaussian diffusional metrics that may not bepossible when simply deriving such very high b-value im-ages using a standard mono-exponential fit [31]; direct ac-quisition of DW images with b values greater than1,000 s/mm2 may be necessary when desiring to analysesuch metrics, for instance for the purpose of characterisingtumour aggressiveness [31]. In comparison, in our study we

used DWI solely for the purpose of the discrimination ofprostate cancer and benign prostate; a recent study analysinga spectrum of mathematical models for fitting data fromDWI of the prostate concluded that a mono-exponentialmodel is sufficient when using DWI for the purpose oftumour detection [32].

In conclusion, computed DWI is a straightforward post-processing technique requiring no additional imaging time thatprovides improved suppression of benign prostate than directb-1,000 images combined with improved distortion and ghost-ing compared with direct b-1,500 images. These benefitsyielded improved diagnostic performance for tumour detectionrelative to radical prostatectomy findings for two independentradiologists. Thus, we suggest that computed DW images at bvalues greater than 1,000 s/mm2 might be routinely incorpo-rated into multi-parametric prostate MRI protocols.

Acknowledgements We would like to acknowledge the Joseph andDiane Steinberg Charitable Trust for support.

C. Geppert is an employee of Siemens (Erlangen, Germany). How-ever, C.G. did not provide any funding for this study, and the remainingauthors had control of all study data.

References

1. Barentsz JO, Richenberg J, Clements R et al (2012) ESUR prostateMR guidelines 2012. Eur Radiol 22:746–757

2. Hoeks CM, Barentsz JO, Hambrock T et al (2011) Prostate cancer:multiparametric MR imaging for detection, localization, and stag-ing. Radiology 261:46–66

3. Lim HK, Kim JK, Kim KA, Cho KS (2009) Prostate cancer:apparent diffusion coefficient map with T2-weighted images fordetection—a multireader study. Radiology 250:145–151

4. Haider MA, van der Kwast TH, Tanguay J et al (2007) CombinedT2-weighted and diffusion-weighted MRI for localization of pros-tate cancer. AJR Am J Roentgenol 189:323–328

5. Selnaes KM, Heerschap A, Jensen LR et al (2012) Peripheral zoneprostate cancer localization by multiparametric magnetic resonanceat 3 T: unbiased cancer identification by matching to histopatholo-gy. Invest Radiol 47:624–633

6. Hambrock T, Hoeks C, Hulsbergen-van de Kaa C et al (2012)Prospective assessment of prostate cancer aggressiveness using 3-T diffusion-weighted magnetic resonance imaging-guided biopsiesversus a systematic 10-core transrectal ultrasound prostate biopsycohort. Eur Urol 61:177–184

7. Somford DM, Hoeks CM, Hulsbergen-van de Kaa CA et al (2013)Evaluation of diffusion-weighted MR imaging at inclusion in anactive surveillance protocol for low-risk prostate cancer. InvestRadiol 48:152–157

8. Somford DM, Hambrock T, Hulsbergen-van de Kaa CA et al(2012) Initial experience with identifying high-grade prostate can-cer using diffusion-weighted MR imaging (DWI) in patients with aGleason score≤3+3=6 upon schematic TRUS-guided biopsy: aradical prostatectomy correlated series. Invest Radiol 47:153–158

9. Kim CK, Park BK, Kim B (2010) Diffusion-weighted MRI at 3 T forthe evaluation of prostate cancer. AJR Am J Roentgenol 194:1461–1469

Eur Radiol

10. Vilanova JC, Barcelo-Vidal C, Comet J et al (2011) Usefulness ofprebiopsy multifunctional and morphologic MRI combined withfree-to-total prostate-specific antigen ratio in the detection of pros-tate cancer. AJR Am J Roentgenol 196:W715–W722

11. Tanimoto A, Nakashima J, Kohno H, Shinmoto H, Kuribayashi S(2007) Prostate cancer screening: the clinical value of diffusion-weighted imaging and dynamic MR imaging in combination withT2-weighted imaging. J Magn Reson Imaging 25:146–152

12. Vargas HA, Akin O, Shukla-Dave A et al (2012) Performancecharacteristics of MR imaging in the evaluation of clinically low-risk prostate cancer: a prospective study. Radiology 265:478–487

13. Kobus T, Vos PC, Hambrock T et al (2012) Prostate cancer aggres-siveness: in vivo assessment of MR spectroscopy and diffusion-weighted imaging at 3 T. Radiology 265:457–467

14. Kitajima K, Kaji Y, Kuroda K, Sugimura K (2008) High b-valuediffusion-weighted imaging in normal and malignant peripheralzone tissue of the prostate: effect of signal-to-noise ratio. MagnReson Med Sci 7:93–99

15. Kim CK, Park BK, Kim B (2010) High-b-value diffusion-weightedimaging at 3 T to detect prostate cancer: comparisons between bvalues of 1,000 and 2,000 s/mm2. AJR Am J Roentgenol 194:W33–W37

16. Metens T, Miranda D, Absil J, Matos C (2012) What is the optimalb value in diffusion-weighted MR imaging to depict prostate cancerat 3T? Eur Radiol 22:703–709

17. Katahira K, Takahara T, Kwee TC et al (2011) Ultra-high-b-valuediffusion-weighted MR imaging for the detection of prostate can-cer: evaluation in 201 cases with histopathological correlation. EurRadiol 21:188–196

18. Ohgiya Y, Suyama J, Seino N et al (2012) Diagnostic accuracy ofultra-high-b-value 3.0-T diffusion-weighted MR imaging for de-tection of prostate cancer. Clin Imaging 36:526–531

19. Kitajima K, Takahashi S, Ueno Y et al (2012) Clinical utility ofapparent diffusion coefficient values obtained using high b-valuewhen diagnosing prostate cancer using 3 tesla MRI: comparisonbetween ultra-high b-value (2000 s/mm2) and standard high b-value(1000 s/mm2). J Magn Reson Imaging 36:198–205

20. Ben-Amitay S, Jones DK, Assaf Y (2012) Motion correction andregistration of high b-value diffusion weighted images. MagnReson Med 67:1694–1702

21. Blackledge MD, Leach MO, Collins DJ, Koh DM (2011)Computed diffusion-weighted MR imaging may improve tumordetection. Radiology 261:573–581

22. Ahmed HU (2009) The index lesion and the origin of prostatecancer. N Engl J Med 361:1704–1706

23. Bott SR, Ahmed HU, Hindley RG, Abdul-Rahman A, Freeman A,Emberton M (2010) The index lesion and focal therapy: an analysisof the pathological characteristics of prostate cancer. BJU Int106:1607–1611

24. Karavitakis M, Winkler M, Abel P, Livni N, Beckley I, Ahmed HU(2011) Histological characteristics of the index lesion in whole-mount radical prostatectomy specimens: implications for focaltherapy. Prostate Cancer Prostatic Dis 14:46–52

25. Cornfeld DM, Israel G, McCarthy SM, Weinreb JC (2008) Pelvicimaging using a T1W fat-suppressed three-dimensional dual echoDixon technique at 3T. J Magn Reson Imaging 28:121–127

26. Hori M, Kim T, Onishi H et al (2011) Uterine tumors: comparisonof 3D versus 2D T2-weighted turbo spin-echo MR imaging at 3.0T—initial experience. Radiology 258:154–163

27. Langer DL, van der Kwast TH, Evans AJ et al (2008) Intermixednormal tissue within prostate cancer: effect on MR imaging mea-surements of apparent diffusion coefficient and T2—sparse versusdense cancers. Radiology 249:900–908

28. Rosenkrantz AB, Kong X, Niver BE et al (2011) Prostate cancer:comparison of tumor visibility on trace diffusion-weighted imagesand the apparent diffusion coefficient map. AJR Am J Roentgenol196:123–129

29. Liu W, Laitinen S, Khan S et al (2009) Copy number analysisindicates monoclonal origin of lethal metastatic prostate cancer.Nat Med 15:559–565

30. Wise AM, Stamey TA,McNeal JE, Clayton JL (2002) Morphologicand clinical significance of multifocal prostate cancers in radicalprostatectomy specimens. Urology 60:264–269

31. Rosenkrantz AB, Sigmund EE, Johnson G et al (2012) Prostatecancer: feasibility and preliminary experience of a diffusional kur-tosis model for detection and assessment of aggressiveness ofperipheral zone cancer. Radiology 264:126–135

32. Quentin M, Blondin D, Klasen J et al (2012) Comparison ofdifferent mathematical models of diffusion-weighted prostate MRimaging. Magn Reson Imaging 30:1468–1474

Eur Radiol