nonlinear registration of diffusion-weighted images

FULL PAPERS

Nonlinear Registration of Diffusion-Weighted ImagesImproves Clinical Sensitivity of Functional Diffusion Mapsin Recurrent Glioblastoma Treated With Bevacizumab

Benjamin M. Ellingson,1* Timothy F. Cloughesy,2 Albert Lai,2

Phioanh L. Nghiemphu,2 and Whitney B. Pope1

Diffusion-weighted imaging estimates of apparent diffusioncoefficient (ADC) have shown sensitivity to brain tumor cellu-larity as well as response to therapy. Functional diffusionmaps (fDMs) exploit these principles by examining voxelwisechanges in ADC within the same patient over time. Currently,the fDM technique involves linear image registration of ADCmaps from subsequent follow-up times to pretreatment ADCmaps; however, misregistration of ADC maps due to geomet-ric distortions as well as mass effect from growing tumor canconfound fDM measurements. In this study, we compare theuse of a nonlinear registration scheme to the current linearfDM technique in 70 patients with recurrent glioblastomamultiforme treated with bevacizumab. Results suggest thatnonlinear registration of pretreatment ADC maps to post-treatment ADC maps improves the clinical predictability, sen-sitivity, and specificity of fDMs for both progression-free andoverall survival. Magn Reson Med 000:000–000, 2011. VC 2011Wiley-Liss, Inc.

Key words: functional diffusion map; fDM; glioblastoma;bevacizumab; diffusion MRI

Glioblastoma multiforme (GBM), a particularly infiltra-tive and aggressive type of glioma, makes up nearly 17%of all primary brain tumors and approximately 54% ofall primary brain and central nervous system gliomas (1).GBMs are trademarked by a very poor patient prognosis,with a mean survival of only around 14.6 months follow-ing standard therapy including radiation and chemother-apy (2). Because of this poor prognosis, there is a greatneed for early imaging biomarkers to evaluate effective-ness of new therapies and predict patient survival. Diffu-sion-weighted magnetic resonance imaging (DWI) techni-ques have shown value in evaluating regions suspectedof tumor invasion and proliferation (3), as well as the effi-cacy of specific treatment paradigms. Through calculationof an apparent diffusion coefficient (ADC) from DWIs,multiple studies have reported an inverse relationship

between ADC and tumor cell density (4–6), suggestingthat DWIs can potentially be used as imaging biomarkersto evaluate growing tumor and treatment effectiveness.

The functional diffusion map (fDM) was first proposedin 2005 as a method to detect changes in ADC known toaccompany successful cytotoxic therapies (7,8). In con-trast to other techniques that average the ADC measure-ments for an entire region, fDMs do not assume homo-geneity within tumors. Instead, multiple ADC mapscollected on different days are coregistered with a base-line image at an initial time point, and then regional(voxelwise) changes in ADC are isolated and labeledaccording to the magnitude of their change. This tech-nique has shown to be one of the most sensitive earlybiomarkers for tumor response to chemotherapeutics andradiotherapy and has shown to be highly specific to pro-gression of high-grade gliomas (9,10). Additionally,investigators have shown that fDMs can be used as a toolfor long-term clinical assessment of recurrent braintumors (11), are sensitive to infiltrating tumors such asgliomatosis cerebri (12), and allow visualization andquantification of regional changes in cellularity thatcorrelate with tumor status and response to various treat-ments (11,13).

Despite promising results from early fDM studies,proper image registration between DWI datasets remainsa major limitation to this technique. Slight misregistra-tion between DWI datasets due to image distortion aswell as mass effect from edema or growing tumor canpotentially confound the interpretation and quantifica-tion of fDM-classified tumor regions. To date, implemen-tation of fDMs for brain tumor evaluation has consistedof only linear image registration techniques to matchbaseline DWIs to subsequent DWI datasets. We hypothe-size that an additional step of nonlinear (elastic) registra-tion after a linear registration step may improve the clin-ical sensitivity of fDMs when applied to the brain,reducing the effects of misregistration and misclassifica-tion. In this study, we explore the effects of nonlinearregistration of DWIs on fDMs in patients with recurrentglioblastoma before and after treatment with the antian-giogenic agent bevacizumab.

MATERIALS AND METHODS

Patients

All patients participating in this study signed institu-tional review board-approved informed consent to have

1Department of Radiological Sciences, David Geffen School of Medicine,University of California Los Angeles, Los Angeles, California, USA.2Department of Neurology, David Geffen School of Medicine, University ofCalifornia Los Angeles, Los Angeles, California, USA.

Grant sponsors: Brain Tumor Funders Collaborative, Art of the Brain,Ziering Family Foundation in memory of Sigi Ziering, Singleton FamilyFoundation, Clarence Klein Fund for Neuro-Oncology.

*Correspondence to: Benjamin M. Ellingson, Ph.D., Department ofRadiological Sciences, David Geffen School of Medicine, University ofCalifornia Los Angeles, 924 Westwood Boulevard, Suite 615, Los Angeles,CA 90024. E-mail: [email protected]

Received 25 January 2011; revised 21 April 2011; accepted 24 April 2011.

DOI 10.1002/mrm.23003Published online in Wiley Online Library (wileyonlinelibrary.com).

Magnetic Resonance in Medicine 000:000–000 (2011)

VC 2011 Wiley-Liss, Inc. 1

their data collected and stored in our institution’s neuro-oncology database. Data acquisition was performed incompliance with all applicable Health Insurance Portabil-ity and Accountability Act regulations. The studyspanned November 15, 2005 to August 31, 2010. Patientswere retrospectively selected from our institution’s neuro-oncology database. A total of n ¼ 252 patients who metthe following criteria were initially selected: (1) pathologyconfirmed GBM with recurrence based on MRI and clini-cal data, (2) regularly treated every 2 weeks per cyclewith bevacizumab (Avastin, Genentech, South San Fran-cisco, CA; 5 or 10 mg/kg body weight), and (3) baseline(pre-bevacizumab treatment) and minimum of one follow-up MRI scans. Of these patients, n ¼ 70 patients hadgood quality DWIs before and after initiation of bevacizu-mab treatment and had adequate image registration (linearand nonlinear) without gross misalignment. Baselinescans were obtained approximately 1.5 weeks before treat-ment, and follow-up scans were obtained at approxi-mately 6 weeks after the initiation of bevacizumab. At thetime of last assessment (August, 2010), 57 of the 70patients were deceased. All patients were treated withradiation therapy (typically 6000 cGy) and maximal tumorresection at time of initial tumor presentation. Patientsdid not receive additional radiation or chemotherapybetween pretreatment and post-treatment MRI scans.

MRI

Data were collected on a 1.5-T MR system (General Elec-tric Medical Systems, Waukesha, WI) using pulse sequen-ces supplied by the scanner manufacturer. Standardanatomical MRI sequences included axial T1-weighted(echo time [TE]/repetition time [TR] ¼15 ms/400 ms, slicethickness ¼ 5 mm with 1 mm interslice distance, numberof excitations (NEX) ¼ 2, matrix size ¼ 256 � 256, andfield-of-view [FOV] ¼ 24 cm), T2-weighted fast spin-echo(TE/TR ¼ 126–130 ms/4000 ms, slice thickness ¼ 5mmwith 1 mm interslice distance, NEX ¼ 2, matrix size ¼256 � 256, and FOV ¼ 24 cm), and fluid-attenuated inver-sion recovery (FLAIR) images (inversion time ¼ 2200 ms,TE/TR ¼ 120 ms/4000 ms, slice thickness ¼ 5 mm with1 mm interslice distance, NEX ¼ 2, matrix size ¼ 256 �256, and FOV ¼ 24 cm). DWIs were collected with TE/TR¼ 102.2 ms/8000 ms, NEX ¼ 1, slice thickness ¼ 5 mmwith 1 mm interslice distance, matrix size ¼ 128 � 128(reconstructed images were zero-padded and interpolatedto 256 � 256), and a FOV ¼ 24 cm using a twice-refocused spin echo echo planar preparation (14). ADCimages were calculated from acquired DWIs with b ¼1000 s/mm2 and b ¼ 0 s/mm2 images. Additionally, gado-pentetate dimeglumine-enhanced (Magnevist; Berlex,Wayne, NJ; 0.1 mmol/kg) axial and coronal T1-weightedimages (coronal: TE/TR ¼ 15 ms/400 ms, slice thickness3 mm with 1mm interslice distance, NEX ¼ 2, a matrixsize of 256 � 256, and FOV ¼ 24 cm) were acquiredimmediately after contrast injection.

Linear Registration

All images for each patient were registered to baselineanatomical T1-weighted images using a mutual informa-

tion algorithm and a 12-degree of freedom transformationusing FSL (FMRIB, Oxford, UK; http://www.fmrib.ox.ac.uk/fsl/). Fine registration (1–2 degrees and 1–2 voxels)was then performed using a Fourier transform-based, 6-degree of freedom, rigid body registration algorithm (15)followed by visual inspection to ensure adequate align-ment. Similar registration techniques were used in previ-ous fDM studies (5,7,12,16).

Nonlinear Registration

After linear registration was complete, nonlinear registra-tion was implemented using the FMRIB non-linear regis-tration tool (FNIRT), a free-form deformation algorithmusing a B-spline basis function (17–19). FNIRT uses aLevenberg-Marquardt modification of the Gauss-Newtonmethod for optimizing the minimum of a sum-of-squarescost function to perform the nonlinear registration.Default values for registration parameters supplied byFMRIB were used in this study (see http://www.fmrib.ox.ac.uk/fsl/fnirt/index.html for more details). Two dif-ferent nonlinear registrations were performed (Fig. 1band c). First, we performed nonlinear registration on pre-treatment ADC maps to register to post-treatment ADCmaps (‘‘pre-to-post’’). This registration scheme wasthought to reduce the amount of mass effect and edemapresent in the pretreatment ADC maps by warping to thepost-treatment ADC maps. Second, we performed nonlin-ear registration on post-treatment ADC maps to alignwith pretreatment ADC maps (‘‘post-to-pre’’). Weexpected this registration scheme to produce similarfDM measurements to the linear case, as much of theeffects are likely to be due to reduction in edema.

As FNIRT uses a sum-of-squares cost function, whichmay lead to problems due to dependence on imageintensities, we also tested a nonlinear registration algo-rithm using the image registration toolkit (IRTK), whichuses cubic B-splines basis function and a normalizedmutual information cost function (20,21). Default valuesfor registration parameters supplied by IRTK were used(see http://www.doc.ic.ac.uk/�dr/software/usage.htmlfor more details). Additionally, we explored the use ofnonlinear registration between T1-weighted anatomicaldatasets to determine whether results differed comparedto registering ADC maps directly. Specifically, nonlinearregistration was used to align pretreatment and post-treatment T1-weighted anatomical images, and then thistransformation was subsequently applied to link pretreat-ment and post-treatment ADC maps.

Regions of Interest

In this study, we chose to examine both regions of FLAIRsignal abnormality on pretreatment FLAIR images as wellas regions of contrast enhancement on pretreatment, post-contrast T1-weighted images to provide a comprehensivecomparison. FLAIR regions of interest (ROIs) (5,11–13,22)and contrast-enhancing ROIs (7–9,16) have both previ-ously been used in interpreting fDM results. For isolatingROIs, we used a semiautomated process consisting of: (1)manually defining the relative region of tumor occurrence,then (2) thresholding either the FLAIR or postcontrast

2 Ellingson et al.

images using an empirical threshold combined with aregion-growing algorithm, and then (3) manually editingthe resulting masks to exclude obvious errors.

Functional Diffusion Map Calculation

fDM calculation was performed as recommended in aprevious publication (5). After proper registration wasvisually verified, voxelwise subtraction was performedbetween acquired, post-treatment ADC maps and base-line, pretreatment ADC maps. Individual voxels werestratified into three categories based on the change inADC relative to the baseline ADC map. Voxels weredefined as ‘‘significantly increasing’’ or ‘‘significantlydecreasing’’ if DADC values were beyond 0.40 mm2/ms,which has been suggested to be the best threshold for abalance between sensitivity and specificity to progress-ing disease and corresponds to an estimated change in

cell density of more than 63960 nuclei/mm2 inuntreated gliomas (5). The percent fractional volume ofincreasing (%VADC > 0.4 mm2/ms; red) and decreasing(%VADC < 0.4 mm2/ms; blue) voxels within FLAIR andcontrast-enhancing regions were used as biomarkers forcomparing linear to nonlinear fDMs and for predictingpatient outcomes.

Hypothesis Testing

We hypothesized that fDMs generated from ADC mapsusing nonlinear registration (‘‘nonlinear fDMs’’) wouldhave significantly lower volume fraction of changingvoxels (increasing or decreasing) compared to fDMsgenerated using only linear registration (‘‘linear fDMs’’).To test this hypothesis, we performed multiplerepeated measures, two-way analysis of variance foreach of the fDM measures (%VADC > 0.4 mm2/ms and

FIG. 1. Linear and nonlinear fDM calculations. a: Linear (traditional) fDMs consisting of using a linear registration algorithm to alignpost-treatment ADC maps to pretreatment ADC maps. b: Pre-to-post nonlinear fDMs consisting of nonlinear registration of pretreatment

ADC maps to post-treatment ADC maps. c: Post-to-pre nonlinear fDMs consisting of nonlinear registration of post-treatment ADC mapsto pretreatment ADC maps. For fDMs, blue voxels represent a significant decrease in ADC (beyond 0.4 mm2/ms), red voxels represent asignificant increase in ADC (beyond 0.4 mm2/ms), and green voxels are those with no significant change in ADC.

Nonlinear fDMs in Recurrent GBM 3

%VADC < 0.4 mm2/ms), ROIs (FLAIR and contrastenhancing), and registration algorithms (FNIRT-ADC,FNIRT-Anatomy, IRTK-ADC, and IRTK-Anatomy) acrossthe different registration schemes (linear, pre-to-post non-linear, and post-to-pre nonlinear). Tukey’s test for multi-ple comparisons was used for post hoc comparisons.Additionally, we tested whether the volume fraction oftissue showing a significant decrease in ADC using eitherthe linear or nonlinear techniques was a significant pre-dictor of time-to-progression (TTP) or overall survival (OS)with respect to the post-treatment MRI scan date. Thevolume fraction of decreasing ADC (blue regions) wasused as a predictor of patient outcomes because it isthought to reflect both reduction in edema as well as achange in tumor cellularity after bevacizumab treatment(13,22). Disease progression was defined by a modifiedMacdonald criteria (23,24), which included either a 25%increase in enhancing tumor or an increase in nonenhanc-ing tumor evident by increased mass effect and/or archi-tectural distortions such as the blurring of the gray–whitematter interface. We hypothesized that nonlinear fDMswould be a better predictor of these outcomes comparedto linear fDMs. Log-rank statistical analyses on Kaplan–Meier data were performed using GraphPad PrismVR v4.0statistical software to test this hypothesis. Additionally,the sensitivity, specificity, positive predictive value (PPV),and negative predictive value (NPV) were calculated for3-month progression-free survival (PFS), 6-month PFS, 6-month OS, and 12-month OS using each of the fDM met-rics, ROIs, and registration schemes.

RESULTS

In general, linear fDMs showed a substantial volume oftissue with decreased ADC after treatment with bevaci-zumab, presumably due to reduction in edema or grow-ing tumor, both of which are thought to reduce ADCwithin FLAIR abnormal regions (Fig. 1a). When examin-ing pre-to-post nonlinear fDMs, the quantity of tissuewith high ADC was typically lower as the pretreatmentADC maps were warped to the post-treatment, edema-reduced, ADC maps (Fig. 1b). This resulted in a generaldecrease in the number of fDM-classified voxels. Qualita-tively, post-to-pre nonlinear fDMs appeared similar tolinear fDMs, showing a large volume of tissue with a sig-nificant reduction in ADC (Fig. 1c) likely due to eitherreduction in edema or increased tumor cellularity.

Quantitative analysis confirmed these observations,suggesting a significant difference in all fDM metrics andall ROIs due to the different registration schemes (Fig. 2;repeated measures, two-way analysis of variance, P <0.001 for all fDM metrics and ROIs); however, no differ-ences were observed between nonlinear registration algo-rithms or due to registration of ADC versus anatomicalimages (repeated measures, two-way analysis of variance,P > 0.8 for all combinations and interactions). Withincontrast-enhancing regions, the fractional volume of tis-sue having a significantly lower ADC (blue regions) wasreduced in pre-to-post nonlinear fDMs compared to bothlinear fDMs in Fig. 2a (Tukey, P < 0.05 for all registra-tion algorithms) and post-to-pre nonlinear fDMs (Tukey,P < 0.05 for all registration algorithms). The fractional

volume of tissue having a significantly higher ADC (redregions) within pretreatment contrast-enhancing regionswas significantly different between linear fDMs andpost-to-pre nonlinear fDMs (Fig. 2b; Tukey, P < 0.01 forall registration algorithms). Similar to contrast-enhancingROIs, the fractional volume of tissue having significantlylower ADC within FLAIR ROI was significantly lower inpre-to-post nonlinear fDMs compared to linear fDMs(Fig. 2c; Tukey, P < 0.01 for all registration algorithms).The volume fraction of tissue having significantlyincreasing ADC within FLAIR ROIs was significantly dif-ferent between linear fDMs and nonlinear fDMs (Fig. 2d;Tukey, P < 0.001 for linear fDMs vs. both pre-to-postand post-to-pre nonlinear fDMs).

Next, we tested the hypothesis that a fractional volumeof tissue having a decrease in ADC larger than the me-dian change in all patients results in a shorter TTP andshorter OS, and that nonlinear fDMs provide better strati-fication of survival compared to linear fDMs. As no sta-tistical difference was observed between volume frac-tions calculated with FNIRT versus IRTK, or betweenADC registration and registration performed between T1-weighted anatomical datasets (Fig. 2), we chose to testsurvival differences using FNIRT applied to ADC maps.Within contrast-enhancing ROIs, the median percentageof tumor volume exhibiting a significant decrease inADC for linear and pre-to-post nonlinear fDMs was notable to stratify TTP (Fig. 3a; log-rank, linear fDMs, P ¼0.1060, pre-to-post nonlinear fDMs, P ¼ 0.0666) or OS(Fig. 3b; log-rank, linear fDMs, P ¼ 0.1229, pre-to-postnonlinear fDMs, P ¼ 0.0539). Interestingly, the fractionalvolume of tissue with decreasing ADC within contrast-enhancing ROIs in post-to-pre nonlinear fDMs were ableto stratify long- and short-term TTP (log-rank, P ¼0.0167) and OS (log-rank, P ¼ 0.0260), despite not hav-ing statistically different fractional volumes to linearfDMs after pairing (Fig. 2a; linear fDMs vs. nonlinearfDMs, P > 0.05). Within FLAIR ROIs, however, both lin-ear and nonlinear fDMs were able to significantly stratifyshort- and long-term TTP and OS (Fig. 3c and d). Specif-ically, a volume fraction of tissue having a significantdecrease in ADC within FLAIR ROIs larger than themedian using linear fDMs resulted in a statisticallyshorter TTP (Fig. 3c; Log-rank, P ¼ 0.0016) and OS (Fig.3d; Log-rank, P ¼ 0.0385). Post-to-pre nonlinear fDMsshowed similar trends to linear fDMs with respect toTTP (Log-rank, P ¼ 0.0171) and OS (Log-rank, P ¼0.0137); however, pre-to-post nonlinear fDMs showeda substantial reduction in the level of significance(P value) compared to both linear and post-to-pre nonlin-ear fDMs for TTP (Log-rank, P ¼ 0.0009) and OS (Log-rank, P ¼ 0.0004). Pre-to-post nonlinear registrationapplied to FLAIR regions appeared to provide the beststratification between short- and long-term PFS and OS,as well as had the highest hazard ratio; however, statisti-cal comparison of hazard ratios suggests that only linearfDMs applied to contrast-enhancing regions had a signifi-cantly different hazard ratio for both PFS and OScompared to other biomarkers. Results further suggestpre-to-post nonlinear fDMs evaluated in FLAIR abnormalregions had a higher sensitivity and specificity to3-month and 6-month PFS compared to other measures

4 Ellingson et al.

(Table 1). Similarly, results also suggest that pre-to-postnonlinear fDMs had a higher sensitivity and specificityto 6-month and 12-month OS compared to other meas-ures (Table 2). Together, these results indicate that pre-to-post nonlinear fDMs applied to FLAIR abnormalregions provides the best clinical predictability, sensitiv-ity, and specificity to both TTP and OS compared to lin-ear fDMs or post-to-pre nonlinear fDMs applied to FLAIRor contrast-enhancing regions.

DISCUSSION

Voxelwise analysis of imaging parameters is a novelapproach for both quantifying and visualizing potentially

heterogeneous tumor response to various cancer treat-ments. By examining voxelwise changes in ADC, thefDM technique has shown promising sensitivity to bothcytotoxic (7,8) and antiangiogenic therapies (22). Despitethese initial results, no significant improvements to thefDM technique have been made since the application offDMs in regions of FLAIR abnormality (compared toregion of contrast enhancement, exclusively). Since thedevelopment of fDMs in 2005, a linear registrationscheme has been used for aligning ADC maps within thesame patient over time. This technique poses severalchallenges when geometric distortions arise in ADCmaps, or when misalignment of ADC maps occurs due tomass effect from growing tumor. These limitations to

FIG. 2. Fractional volumes of fDM-classified tissues within contrast-enhancing and FLAIR abnormal regions for linear fDMs, pre-to-post

nonlinear fDMs, and post-to-pre nonlinear fDMs and different registration algorithms/schemes (FNIRT applied to ADC, FNIRT applied toprecontrast T1-weighted images, IRTK applied to ADC, IRTK applied to precontrast T1-weighted images). a: Fractional volume of tissuewith a significant decrease in ADC (blue voxels) within contrast-enhancing regions. b: Fractional volume of tissue with a significant

increase in ADC (red voxels) within contrast-enhancing regions. c: Fractional volume of tissue with a significant decrease in ADC (bluevoxels) within FLAIR abnormal regions. d: Fractional volume of tissue with a significant increase in ADC (red voxels) within FLAIR abnor-

mal regions. *P < 0.05; **P < 0.01; ***P < 0.001 for Tukey’s test for multiple comparisons.


linear fDMs have been identified for some time; however,it has not been clear how best to implement an alternativeregistration scheme. For example, should baseline ADCmaps be nonlinearly registered to each subsequent ADCmap during follow-up (the ‘‘pre-to-post’’ scheme), orshould each follow-up ADC map be nonlinearly registeredto the baseline ADC map (the ‘‘post-to-pre’’ scheme).Alternatively, one might propose registering all ADCmaps to a template space for subsequent analysis. As afirst attempt to improve the fDM technique, we haveimplemented a simple nonlinear, elastic registrationscheme. Results from this study suggest that nonlinearregistration of ADC maps significantly changes the calcu-lated fDMs and measured parameters, and these changesin fDM parameters tend to improve clinical predictabilityof fDMs with respect to bevacizumab treatment.

Although our results suggest that nonlinear registrationimproves the ability for fDMs to predict survival, it is stillnot clear whether voxels classified as significantly

increasing or decreasing in ADC accurately depict physi-cal changes in tumor microstructure or whether this is anartifact of image registration. To address this concern, theauthors of a previous study have proposed using a simi-larity-guided correspondence correction algorithm forimproving alignment of ADC maps for fDM analysis (25).Results from this pilot study suggested that fDM analysiswas more accurate after application of this correctionalgorithm; however, no validation with respect to tissueor patient outcomes was presented. Therefore, tissue vali-dation of nonlinear fDM-classified voxels is still required.

Comparison to Other Biomarkers

This study demonstrates a significant improvement inbiomarker technology for use in predicting response ofGBM to antiangiogenic therapies. Many imaging bio-markers have been explored for use in assessing theresponse to antiangiogenic therapies (26), including

FIG. 3. Survival analyses for linear and nonlinear fDMs. a: Effects of linear and nonlinear fDM analysis within contrast-enhancing regions

on time-to-progression (TTP). b: Effects of linear and nonlinear fDM analysis within contrast-enhancing regions on overall survival (OS).c: Effects of linear and nonlinear fDM analyses in FLAIR abnormal regions on TTP. d: Effects of linear and nonlinear fDM analyses in

FLAIR abnormal regions on OS. *P < 0.05; **P < 0.01; ***P < 0.001 for Log-rank statistical analyses.

6 Ellingson et al.

Table

1Sensitivity,

Specificity,

PositivePredictiveValue(PPV),andNegativePredictiveValue(NPV)forLinearandNonlinearfDM

MetricsWithRespectto

3-M

onth

and6-M

onth

Progression-FreeSurvival

Median%

volume

3-m

onth

progression-freesurvival

6-m

onth

progression-freesurvival

Sensitivity

(%)

Specificity

(%)

PPV

(%)

NPV

(%)

Sensitivity

(%)

Specificity

(%)

PPV

(%)

NPV

(%)

Contrast-enhancing

regions

VADC

<0.4

mm

2/m

sLinearfDMs

24.80%

60.00

54.00

34.29

77.14

52.27

53.85

65.71

40.00

Pre-to-postnonlinearfDMs

16.03%

55.00

52.00

31.43

74.29

54.55

57.69

68.57

42.86

Post-to-pre

nonlinearfDMs

17.28%

60.00

54.00

34.29

77.14

54.55

57.69

68.57

42.86

VADC

>0.4

mm

2/m

sLinearfDMs

6.73%

50.00

50.00

28.57

71.43

45.45

42.31

57.14

31.43


3.44%

50.00

50.00

28.57

71.43

52.27

53.85

65.71

40.00

Post-to-pre

nonlinearfDMs

2.09%

45.00

48.00

25.71

68.57

52.27

53.85

65.71

40.00

FLAIR

abnorm

al

regions

VADC

<0.4

mm

2/m

sLinearfDMs

19.64%

55.00

52.00

31.43

74.29

59.09

65.38

74.29

48.57


11.38%

65.00

56.00

37.14

80.00

59.09

65.38

74.29

48.57

Post-to-pre

nonlinearfDMs

17.14%

55.00

52.00

31.43

74.29

54.55

57.69

68.57

42.86

VADC

>0.4

mm

2/m

sLinearfDMs

8.67%

60.00

54.00

34.29

77.14

59.09

65.38

74.29

48.57


5.20%

55.00

52.00

31.43

74.29

63.64

73.08

80.00

54.29

Post-to-pre

nonlinearfDMs

2.54%

50.00

50.00

28.57

71.43

54.55

57.69

68.57

42.86

Table

2Sensitivity,

Specificity,

PositivePredictiveValue(PPV),andNegativePredictiveValue(NPV)forLinearandNonlinearfDM

MetricsWithResp

ectto

6-M

onth

and12-M

onth

OverallSurvival

Median%

volume

6-m

onth

Overallsurvival

12-m

onth

Overallsurvival

Sensitivity

(%)

Specificity

(%)

PPV

(%)

NPV

(%)

Sensitivity

(%)

Specificity

(%)

PPV

(%)

NPV

(%)

Contrast-enhancing

regions

VADC

<0.4

mm

2/m

sLinearfDMs

24.80%

44.44

48.08

22.86

71.43

52.17

54.17

68.57

37.14


16.03%

50.00

50.00

25.71

74.29

56.52

62.50

74.29

42.86

Post-to-pre

nonlinearfDMs

17.28%

38.89

46.15

20.00

68.57

58.70

66.67

77.14

45.71

VADC

>0.4

mm

2/m

sLinearfDMs

6.73%

50.00

50.00

25.71

74.29

47.83

45.83

62.86

31.43


3.44%

66.67

55.77

34.29

82.86

52.17

54.17

68.57

37.14

Post-to-pre

nonlinearfDMs

2.09%

61.11

53.85

31.43

80.00

45.65

41.67

60.00

28.57

FLAIR

abnorm

al

regions

VADC

<0.4

mm

2/m

sLinearfDMs

19.64%

50.00

50.00

25.71

74.29

54.35

58.33

71.43

40.00


11.38%

55.56

51.92

28.57

77.14

58.70

66.67

77.14

45.71

Post-to-pre

nonlinearfDMs

17.14

44.44

48.08

22.86

71.43

58.70

66.67

77.14

45.71

VADC

>0.4

mm

2/m

sLinearfDMs

8.67%

61.11

53.85

31.43

80.00

54.35

58.33

71.43

40.00


5.20%

61.11

53.85

31.43

80.00

58.70

66.67

77.14

45.71

Post-to-pre

nonlinearfDMs

2.54%

55.56

51.92

28.57

77.14

47.83

45.83

62.86

31.43


positron emission tomography tracers (27), perfusionMRI (28), and other diffusion MR analyses (29); how-ever, this is the first study to demonstrate the ability topredict both TTP and OS in recurrent GBM patientstreated with bevacizumab. Additionally, linear fDManalyses have been successfully performed on patientstreated with bevacizumab (22) by focusing on temporalchanges in fDM-classified tissue volumes over time.Unlike this previous study, this study focused on usingthe initial changes after treatment to predict patient out-comes. Consistent with previous results, however, wefound that patients having a larger volume of tissuewith a lower ADC after treatment were more likely toprogress sooner and have a shorter survival thanpatients having a lower volume of tissue with signifi-cantly lower ADC after treatment.

Technical Limitations and Considerations

In the current retrospective study, we have chosen to useADC maps calculated from diffusion MR data that werecollected with only b ¼ 0 s/mm2 and b ¼ 1000 s/mm2.By performing ADC calculations on only these DWIs, weensure that our techniques can be retroactively appliedacross multicenter clinical trials. Despite this potentialadvantage, the use of a standard, clinical diffusion MRsequence parameters with only two b values posespotential limitations as to the accuracy of ADC measure-ments. The National Cancer Institute has recommendedthat ADC calculations should be performed using threeor more DWIs at different b values (0, >100, and >500 s/mm2) for perfusion-insensitive estimates of ADC. Addi-tionally, the use of only two b values for calculation ofADC maps in fDM analyses can be confounded by otherpathologies that may have been present, including cere-bral ischemia.

Another potential limitation is the nonlinear algorithmused for image registration. Many nonlinear image regis-tration algorithms have been proposed and compared(30), and each has advantages and disadvantages. Wechose to use FNIRT as part of the FSL software packagefor ease of use in our fDM postprocessing pipeline, aswell as previous studies suggesting adequate perform-ance (30).

CONCLUSIONS

In summary, nonlinear fDMs appear to provide a signifi-cant improvement in clinical predictability, sensitivity,and specificity of fDMs compared to the traditional, lin-ear fDM approach. Additionally, pre-to-post nonlinearfDMs applied in FLAIR abnormal regions may providethe best sensitivity and specificity for progressing dis-ease and survival compared to post-to-pre nonlinearfDMs.

ACKNOWLEDGMENTS

This research was supported by Brain Tumor FundersCollaborative (to W.B.P.), Art of the Brain (to T.F.C.),Ziering Family Foundation in memory of Sigi Ziering (toT.F.C.), Singleton Family Foundation (to T.F.C.), andClarence Klein Fund for Neuro-Oncology (to T.F.C.).

REFERENCES

1. CBTRUS. Primary brain tumors in the United States 2004–2006.

Chicago, IL: Central Brain Tumor Registry of the United States; 2010.

2. Stupp R, Mason WP, van den Bent MJ, Weller M, Fisher B, Taphoorn

MJ, Belanger K, Brandes AA, Morosi C, Bogdahn U, Curschmann J,

Janzer RC, Ludwin SK, Gorlia T, Allgeier A, Lacombe D, Cairncross

JG, Eisenhauer E, Mirimanoff RO. Radiotherapy plus concomitant

and adjuvent temozolomide for glioblastoma. N Engl J Med 2005;352:

987–996.

3. Ellingson BM, Laviolette PS, Rand SD, Malkin MG, Connelly JM,

Mueller WM, Prost RW, Schmainda KM. Spatially quantifying micro-

scopic tumor invasion and proliferation using a voxel-wise solution

to a glioma growth model and serial diffusion MRI. Magn Reson Med

2011;65:1131–1143.

4. Sugahara T, Korogi Y, Kochi M, Ikushima I, Shigematu Y, Hirai T,

Okudo T, Liang L, Ge Y, Komohara Y, Ushio Y, Takahashi M. Use-

fulness of diffusion-weighted MRI with echo-planar technique in the

evaluation of cellularity in gliomas. J Magn Reson Imaging 1999;9:

53–60.

5. Ellingson BM, Malkin MG, Rand SD, Connelly JM, Quinsey C, LaVio-

lette PS, Bedekar DP, Schmainda KM. Validation of functional diffu-

sion maps (fDMs) as a biomarker for human glioma cellularity.

J Magn Reson Imaging 2010;31:538–548.

6. Chenevert TL, Stegman LD, Taylor JM, Robertson PL, Greenberg HS,

Rehemtulla A, Ross BD. Diffusion magnetic resonance imaging: an

early surrogate marker for therapeutic efficacy in brain tumors. J Natl

Cancer Inst 2000;92:2029–2036.

7. Hamstra DA, Chenevert TL, Moffat BA, Johnson TD, Meyer CR,

Mukherji S, Quint DJ, Gebarski SS, Fan X, Tsien CI, Lawrence TS,

Junck L, Rehemtulla A, Ross BD. Evaluation of the functional diffu-

sion map as an early biomarker of time-to-progression and overall

survival in high-grade glioma. Proc Natl Acad Sci USA 2005;102:

16759–16764.

8. Moffat BA, Chenevert TL, Lawrence TS, Meyer CR, Johnson TD,

Dong Q, Tsien CI, Mukherji S, Quint DJ, Gebarski SS, Robertson PL,

Junck L, Rehemtulla A, Ross BD. Functional diffusion map: a nonin-

vasive MRI biomarker for early stratification of clinical brain tumor

response. Proc Natl Acad Sci USA 2005;102:5524–5529.

9. Moffat BA, Chenevert TL, Meyer CR, McKeever PE, Hall DE, Hoff

BA, Johnson TD, Rehemtulla A, Ross BD. The functional diffusion

map: an imaging biomarker for the early prediction of cancer treat-

ment outcome. Neoplasia 2006;8:259–267.

10. Hamstra DA, Galban CJ, Meyer CR, Johnson TD, Sundgren PC, Tsien

CI, Lawrence TS, Junck L, Ross DJ, Rehemtulla A, Ross BD, Chene-

vert TL. Functional diffusion map as an early imaging biomarker for

high-grade glioma: correlation with convensional radiologic response

and overall survival. J Clin Oncol 2008;26:3387–3394.

11. Ellingson BM, Malkin MG, Rand SD, Bedekar DP, Schmainda KM.

Functional diffusion maps applied to FLAIR abnormal areas are

valuable for the clinical monitoring of recurrent brain tumors. In:

Proceedings of the 17th Annual Meeting of ISMRM, Honolulu,

Hawaii, USA, 2009. p 285.

12. Ellingson BM, Rand SD, Malkin MG, Schmainda KM. Utility of func-

tional diffusion maps to monitor a patient diagnosed with gliomato-

sis cerebri. J Neurooncol 2010;97:419–423.

13. Ellingson BM, Malkin MG, Rand SD, Hoyt A, Connelly J, Bedekar

DP, Kurpad SN, Schmainda KM. Comparison of cytotoxic and anti-

angiogenic treatment responses using functional diffusion maps

in FLAIR abnormal regions. In: Proceedings of the 17th Annual

Meeting of ISMRM, Honolulu, Hawaii, USA, 2009. p 1010.

14. Reese TG, Heid O, Weisskoff RM, Wedeen VJ. Reduction of eddy-cur-

rent-induced distortion in diffusion MRI using a twice-refocused

spin echo. Magn Reson Med 2003;49:177–182.

15. Cox RW, Jesmanowicz A. Real-time 3D image registration for func-

tional MRI. Magn Reson Med 1999;42:1014–1018.

16. Hamstra DA, Galban CJ, Meyer CR, Johnson TD, Sundgren PC, Tsien

CI, Lawrence TS, Junck L, Ross DJ, Rehemtulla A, Ross BD, Chene-

vert TL. Functional diffusion map as an early imaging biomarker for

high-grade glioma: correlation with conventional radiologic response

and overall survival. J Clin Oncol 2008;26:3387–3394.

17. Andersson JL, Smith SM, Jenkinson M. FNIRT–FMRIB’s non-linear

image registration tool. Annual Meeting of the Organization Human

Brain Mapping, Melborne, Australia; 2008.

18. Andersson JL, Jenkinson M, Smith SM. Non-linear optimisation,

FMRIB technical report TR07JA1; 2007.

8 Ellingson et al.

19. Andersson JL, Jenkinson M, Smith SM. Non-linear registration, aka

spatial normalisation, FMRIB technical report TR07JA2; 2007.

20. Rueckert D, Sonoda LI, Hayes C, Hill DL, Leach MO, Hawkes DJ.

Nonrigid registration using free-form deformations: application to

breast MR images. IEEE Trans Med Imaging 1999;18:712–721.

21. Denton ER, Sonoda LI, Rueckert D, Rankin SC, Hayes C, Leach MO,

Hill DL, Hawkes DJ. Comparison and evaluation of rigid, affine, and

nonrigid registration of breast MR images. J Comput Assist Tomogr

1999;23:800–805.

22. Ellingson BM, Malkin MG, Rand SD, Laviolette PS, Connelly JM,

Mueller WM, Schmainda KM. Volumetric analysis of functional dif-

fusion maps is a predictive imaging biomarker for cytotoxic and anti-

angiogenic treatments in malignant gliomas. J Neurooncol 2011;102:

95–103.

23. Macdonald DR, Cascino TL, Schold Jr SC, Cairncross JG. Response

criteria for phase II studies of supratentorial malignant glioma. J Clin

Oncol 1990;8:1277–1280.

24. Pope WB, Sayre J, Perlina A, Villablanca JP, Mischel PS, Cloughesy

TF. MR imaging correlates of survival in patients with high-grade

gliomas. Am J Neuroradiol 2005;26:2466–2474.

25. Hoisak JD, Koh E-S, Yu E, Kassner A, Laperriere NJ, Menard C,

Jaffray DA. An image similarity-guided correspondence correction for

voxel-wise analysis applied to MR imaging of glioblastoma multi-

forme acquired pre- and post-chemotherapy. In: Proceedings of the

18th Annual Meeting of ISMRM, Stockholm, Sweden, 2010. p 616.

26. Pope WB, Young JR, Ellingson BM. Advances in MRI assessment of

gliomas and response to anti-VEGF therapy. Curr Neurol Neurosci

Rep 2011;11:336–343.

27. Chen W, Delaloye S, Silverman DH, Geist C, Czernin J, Sayre J,

Satyamurthy N, Pope W, Lai A, Phelps ME, Cloughesy T. Predicting

treatment response of malignant gliomas to bevacizumab and irinote-

can by imaging proliferation with [18F] fluorothymidine positron

emission tomography: a pilot study. J Clin Oncol 2007;25:4714–4721.

28. Sorensen AG, Batchelor TT, Zhang WT, Chen PJ, Yeo P, Wang M,

Jennings D, Wen PY, Lahdenranta J, Ancukiewicz M, di Tomaso E,

Duda DG, Jain RK. A ‘‘vascular normalization index" as potential

mechanistic biomarker to predict survival after a single dose of cedir-

anib in recurrent glioblastoma patients. Cancer Res 2009;69:

5296–5300.

29. Pope WB, Kim HJ, Huo J, Alger J, Brown MS, Gjertson D, Sai V,

Young JR, Tekchandani L, Cloughesy TF, Mischel PS, Lai A,

Nghiemphu PL, Rahmanuddin S, Goldin J. Recurrent glioblastoma

multiforme: ADC histogram analysis predicts response to bevacizu-

mab treatment. Radiology 2009;252:182–189.

30. Klein A, Andersson J, Ardekani BA, Ashburner J, Avants B, Chiang

MC, Christensen GE, Collins DL, Gee J, Hellier P, Song JH, Jenkinson

M, Lepage C, Rueckert D, Thompson P, Vercauteren T, Woods RP,

Mann JJ, Parsey RV. Evaluation of 14 nonlinear deformation algo-

rithms applied to human brain MRI registration. Neuroimage 2009;

46:786–802.


nonlinear registration of diffusion-weighted images

Documents