interactive diffusion tensor tractography visualization for neurosurgical planning

Interactive Diffusion Tensor TractographyVisualization for Neurosurgical Planning

BACKGROUND: Diffusion tensor imaging (DTI) infers the trajectory and location of largewhite matter tracts by measuring the anisotropic diffusion of water. DTI data may thenbe analyzed and presented as tractography for visualization of the tracts in 3 dimensions.Despite the important information contained in tractography images, usefulness forneurosurgical planning has been limited by the inability to define which are criticalstructures within the mass of demonstrated fibers and to clarify their relationship to thetumor.

OBJECTIVE: To develop a method to allow the interactive querying of tractography datasets for surgical planning and to provide a working software package for the researchcommunity.

METHODS: The tool was implemented within an open source software project. Echo-planar DTI at 3 T was performed on 5 patients, followed by tensor calculation. Softwarewas developed that allowed the placement of a dynamic seed point for local selection offibers and for fiber display around a segmented structure, both with tunable parameters.A neurosurgeon was trained in the use of software in , 1 hour and used it to reviewcases.

RESULTS: Tracts near tumor and critical structures were interactively visualized in3 dimensions to determine spatial relationships to lesion. Tracts were selected using3 methods: anatomical and functional magnetic resonance imaging-defined regions ofinterest, distance from the segmented tumor volume, and dynamic seed-point spheres.

CONCLUSION: Interactive tractography successfully enabled inspection of white matterstructures that were in proximity to lesions, critical structures, and functional corticalareas, allowing the surgeon to explore the relationships between them.

KEY WORDS: Diffusion tensor imaging, Magnetic resonance imaging, Neurosurgery, Surgical planning,

Tractography

Neurosurgery 68:496–505, 2011 DOI: 10.1227/NEU.0b013e3182061ebb www.neurosurgery-online.com

For most patients with brain tumors, grosstotal resection increases time to pro-gression, lengthens survival, reduces mass

effect and associated neurologic deficits, providessignificant cytoreduction, and may be curativefor some tumor types.1-4 Tumors located nearcritical cortical regions or functionally significant

white matter (WM) fiber tracts are difficult toresect maximally while avoiding postoperativeneurological deficits.5 Defining the limits of thetumor relative to key cortical areas and associatedWM tracts can be difficult because of the in-ability to distinguish tumor from brain and toidentify those key structures intraoperatively.Functional brain mapping with a variety oftechniques (functional magnetic resonance im-aging [fMRI], positron emission tomography,magnetic source imaging) now allows pre-operative and noninvasive demonstration ofcritical cortical areas.6-8 Defining critical WManatomy and its relationship to cortical areas hasbeen more problematic because WM tractscannot be visualized on conventional imaging

Alexandra J. Golby, MD*†

Gordon Kindlmann, PhD†

Isaiah Norton, BS*

Alexander Yarmarkovich, PhD†

Steven Pieper, PhD†

Ron Kikinis, MD†

Brigham and Women’s Hospital, Depart-

ments of *Neurosurgery and †Radiology,

Harvard Medical School, Boston, Massa-

chusetts

Correspondence:

Alexandra J. Golby, MD,

Department of Neurosurgery,

Brigham and Women’s Hospital,

75 Francis St,

Boston, MA 02115.

E-mail: [email protected]

Received, January 20, 2010.

Accepted, March 5, 2010.

Copyright ª 2011 by the

Congress of Neurological Surgeons

ABBREVIATION: DTI, diffusion tensor imaging; FA,

fractional anisotropy; fMRI, functional magnetic

resonance imaging; WM, white matter

Supplemental digital content is available for this article.

Direct URL citations appear in the printed text and are

provided in the HTML and PDF versions of this article on

the journal’s Web site (www.neurosurgery-online.com).

496 | VOLUME 68 | NUMBER 2 | FEBRUARY 2011 www.neurosurgery-online.com

CONCEPTS, INNOVATIONS, AND TECHNIQUESTOPIC Concepts, Innovations, and Techniques

Copyright © Congress of Neurological Surgeons. Unauthorized reproduction of this article is prohibited.


and there is no reliable way to test for their presence. Moreover, inpatients with brain tumors, variable displacement, disruption, orinfiltration of WM tracts may occur,5,9 and functional WM tractshave even been found within tumor boundaries.10 If injury tocritical WM tracts occurs, the patient will incur a new neuro-logical deficit even if the eloquent cortex has been respected.11

Subcortical stimulation mapping of the WM is the clinical goldstandard and the only functional method for identifying themotor12 and language13 pathways. However, this technique doesnot reveal the full 3-dimensional (3D) extent of the tracts,13-15 isnot in general use, and is not available preoperatively. If thesurgeon could have a preoperative understanding of the patient’sWM anatomy, he or she would be able to assess the risks ofsurgery and the likelihood of complete resection and couldcounsel patients accordingly. Such preoperative informationcould also guide the deployment of intraoperative subcorticalmapping when available, thereby making it more efficient andpossibly validating diffusion tensor imaging (DTI) and intra-operative subcortical WM stimulation against one another.

DTI is an emerging MRI-based technique that can demon-strate WM anatomy by measuring the directional anisotropy ofwater. Diffusion MRI is the first noninvasive technique formeasuring WM fiber structure in vivo.16 In DTI analysis, a tensormodel is used to represent the orientation of WM fibers. In voxelsin which one WM fiber population is predominant, the principaldiffusion direction is aligned with the WM fiber tract direction.17

By following the principal directions of diffusion, a process calledtractography18,19 estimates the trajectories of WM fiber tracts.These reconstructions may then be displayed in 3D, providinga detailed map of the configuration of the tracts and their re-lationship to other structures. Numerous studies have used DTItractography to define WM anatomy in healthy subjects and inpatients with neurological and neurosurgical diseases. In partic-ular, in patients with brain tumors, DTI can demonstrate dis-placement, interruption, or infiltration of WM tracts by thetumor.9,20-29 Tractography to demonstrate WM anatomy inneurosurgical patients has been incorporated into neuro-navigation systems30,31 and even acquired intraoperatively.32

Despite the important information contained in these images,usefulness for neurosurgical planning has been limited by theinability to define which are critical structures within the mass offibers and to clearly demonstrate the relationship of tracts to thetumor. Our goal was to develop a method to allow the interactivequerying of tractography data sets for surgical planning to defineclinically important WM tracts and the relationship to the sur-gical lesion to provide the neurosurgeon with an optimal un-derstanding of the functional brain anatomy for individualpatients. This effort is aided by recent progress in computationalpower that has made such an interactive approach feasible to runon a laptop computer.

To isolate WM structures of interest, the standard method is toseed tractography trajectories (‘‘fibers’’) using manually identifiedregions of interest (ROIs) such as edited contours, spheres, orboxes.33,34 To further select fibers of interest, additional ROIs are

used to limit the tractography to a specific structure.20,35

However, space-occupying lesions are known to have manypathological effects on WM tracts, including disruption, de-struction, infiltration, displacement, and production ofedema.9,36 Because of these and other pathological changesthroughout the brain, the manual identification of ROIs based onknown neuroanatomy is often not straightforward in tumorpatients. To define the relevant anatomy, it would be useful toidentify those fibers that pass within a certain (variable) distanceof the tumor or that run through the tumor, as well as thoseassociated with particular, patient-specific cortical areas such asfMRI activations or magnetoencephalography findings. This canprovide a preoperative functional brain map defining both thecritical cortical functional areas and the WM tracts leading to andfrom these areas. By interactively querying the data, the surgeonshould be able to build a mental representation of the relevantanatomy. Moreover, such an approach could be used intra-operatively to select and display critical WM tracts based onintraoperative findings, particularly the determination of elo-quent cortex in the region of the lesion using electrocorticalstimulation testing.36a

With these goals in mind, we set out to develop a softwaremodule that could displayWM tracts selectively and interactively.We wanted a tool that would show all the tracts within a certaindistance of the tumor boundary and that could interactivelychange that distance to progressively view the layers of tractsaround the tumor. The second tool was designed to allow theplacement of a dynamic seed point of variable size that would seedall the tracts passing through that area and could be movedaround the 3D data set to demonstrate tracts ‘‘on the fly.’’ Wealso wished to be able to use certain regions as seed regions, eg,fMRI activations or anatomic structures. The tools were developedwith the goal of being disseminated widely for investigational useby clinicians. The tools were developed in 3D Slicer, version 3(www.slicer.org), a freely available anatomic visualization softwareproject based at Brigham and Women’s Hospital.37

METHODS

Subjects

All subjects were recruited in accordance with the policies of thePartners Healthcare Institutional Review Board for human subjects.Written informed consent was obtained from all subjects. All patientsunderwent fMRI and DTI preoperatively. As directed by clinical con-siderations and tumor location, fMRI was used to define motor cortex,somatosensory cortex, visual cortex, and/or language areas. DTI wasacquired over the whole brain during the same scanning session. Forpurposes of developing the tool, we applied the method in patients withnewly diagnosed or recurrent primary glial tumors or metastatic braintumors.

Patient 1

Patient 1 is a 19-year-old man who presented with personality changesand new onset of seizures who was found to have a large nonenhancing

INTERACTIVE DTI VISUALIZATION NEUROSURGERY PLANNING

NEUROSURGERY VOLUME 68 | NUMBER 2 | FEBRUARY 2011 | 497



lesion in the left parietal lobe. DTI was acquired with 1 baseline and 31gradient directions. In addition, the patient underwent fMRI duringmotor, sensory, and visual tasks. Using the tools described, tractographywas performed to delineate the principal motor and somatosensory tractsby seeding from fMRI activations and dynamically from known ana-tomic landmarks. The patient underwent surgery for subtotal resectionof the tumor. Pathology demonstrated World Health Organization(WHO) grade III astrocytoma.

Patient 2

Patient 2 is 31-year-old right-handed woman with a large right-hemisphere lesion who noted visual changes a month before pre-sentation. MRI revealed a very large nonenhancing mass involving theoccipital, temporal, and parietal lobes. She underwent stereotactic biopsyat an outside institution that demonstrated infiltrating oligoastrocytoma,WHO grade II. DTI was acquired with 1 baseline and 55 gradientdirections. fMRI was performed for left motor mapping (foot, hand, andface) and visual field mapping. She underwent subtotal resection underlocal anesthesia with intraoperative electrocortical determination of so-matosensory and visual cortical areas. The patient had slight worsening ofher right-sided visual field cut postoperatively.

Patient 3

Patient 3 is a 30-year-old ambidextrous bilingual (English and Por-tuguese) woman with a very large T2-intense non–contrast-enhancinglesion involving the left frontal and temporal lobes. She underwentpreoperative fMRI during language tasks, face and right-hand–clenchingmotor tasks, and DTI with 5 baselines and 31 gradient directions.Surgical resection of the lesion was performed under local anesthesia withlanguage mapping in both English and Portuguese. Electrocorticalstimulation testing demonstrated 2 regions in which stimulation at lowcurrent reliably interfered with language task performance in bothlanguages: 1 directly superficial to the tumor in the posterior inferiorfrontal lobe and 1 directly superficial to the tumor in the superiortemporal gyrus. Intraoperative findings were congruent with fMRI-demonstrated language-associated activations. The resection was performedwith continuous monitoring of language function, and preoperativelyvisualized areas of preservedWM tracts, as well as areas subjacent to positivecortical stimulation points, were spared during the resection. There were nopostoperative language changes.

Patient 4

Patient 4 is a 28-year-old right-handed woman with right posteriortemporal WHO grade II mixed glioma (oligoastrocytoma) presentingwith a probable seizure. Presurgical MRI scans were taken for 3 fMRItasks (left-hand clenching, lip pursing, and visual field stimulation), aswell as DTI (31 directions, 1 baseline). She underwent resection of thelesion without new neurologic deficit.

Patient 5

Patient 5 is a 55-year-old right-handed female with metastatic ade-nocarcinoma of the lung who presented with an asymptomatic contrast-enhancing lesion in the superior left temporal lobe. She underwent fMRI(3 language tasks, right-hand clenching, lip pursing) and DTI (31 di-rections, 5 baselines) before surgical resection of the lesion performedunder local anesthesia with language mapping. The patient did not haveany language changes associated with the procedure.

Image Acquisition

MRIs were obtained with a 3.0-T scanner (EXCITE Signa scanner,GE Medical System, Milwaukee, Wisconsin) with Excite 14.0 using an8-channel head coil and ASSET. As a first step, a high-resolution whole-brain T1-weighted axial 3D spoiled gradient recall (repetition time, 7500milliseconds; echo time, 30 milliseconds; matrix, 256 3 256; field ofview, 25.6 cm; flip angle, 20�; imaging 120–182 slices of 1-mmthickness) was acquired. Next, diffusion-weighted imaging was acquiredwith a multislice single-shot diffusion-weighted echo-planar imagingsequence (repetition time, 14 000 milliseconds; echo time, 30 milli-seconds) consisting of 55 or 31 gradient directions with a b value of 1000s/mm2 and 1 or 5 baseline T2 images. The field of view was 25.6 cm; theimaging matrix was 128 3 128 with a slice thickness of 2.6 mm.A single-shot gradient-echo echo-planar imaging was used to acquireblood oxygen level–dependent functional images (repetition time, 2000milliseconds; echo time, 40 milliseconds; flip angle, 90�; field of view,24 cm; acquisition matrix, 64 3 64; slice gap, 0 mm; voxel size, 3.7533.75 3 5 mm3). In each image volume, 28 axial slices were acquiredusing ascending interleaved scanning sequence. Behavioral paradigmswere selected on the basis of clinical considerations and included visualalternating whole-field checkerboard; self-paced motor tasks for hand,foot, and mouth; somatosensory stimulation; and visually presentedantonym generation and noun categorization language tasks.High-resolution T2-weighted gradient-echo MRIs (repetition time,

8000 milliseconds; echo time, 98 milliseconds; flip angle, 90�; matrix,5123 512; slices, 93; voxel size, 0.53 0.53 1.5 mm3) were acquired todemonstrate the surgical pathology.

Image Analyses

For each subject, a diffusion tensor volume derived from the diffusion-weighted image volume was calculated with the Teem (http://teem.sf.net)library through 3D Slicer. As implemented in 3D Slicer, the WM tracttrajectories are estimated with a single tensor model. The standard streamlinetractography method (as in the work by Basser et al38) repeatedly steps in theprincipal diffusion direction defined by the tensor at each location.fMRI data were realigned, motion corrected, and analyzed with SPM2

(Wellcome Department of Cognitive Neurology, London, UK). Becauseimages may be acquired in multiple sessions and because patients oftenmove during the course of an imaging session, image-based registrationtechniques are used to align the structural, DTI, and fMRI volumes. Oursoftware relies on the Insight Toolkit (ITK, http://www.itk.org39) toprovide image registration using a wide range of linear and nonlinearalgorithms. Care must be taken when reviewing the alignment of theseregistrations because MRI scans in general, and echo-planar imagingsequences used for fMRI and DTI in particular, often trade improvedsignal at the cost of increased geometric distortion.40 A full character-ization of these distortions is beyond the scope of the present discussionbecause the complexity of these nonlinear distortions is difficult tocapture with current tools. In practice, we currently rely on a process ofmanual registration to rough fit, followed by automatic affine registration(linear-only compensation for variations in scale, translation, rotationand shear; ITK-implemented algorithm39), followed by manual in-spection and correction as necessary.

Hardware Requirements

Data analyses and interactive tractography were performed on a Dellworkstation with Intel Core 2 Duo 6300 processor at 1.86 GHz, 2 Gb ofRAM, 7200 RPM SATA hard-drive, nVidia Quadro FX3450 graphics

GOLBY ET AL




accelerator with 256-Mb dedicated memory, and Linux operating systemkernel running proprietary nVidia drivers. For case 3, with a 2563 256,38–axial-slice diffusion-weighted volume with 5 baseline scans and 31gradient directions, estimation of diffusion tensors in 3D Slicer requiredapproximately 1 minute. Interactive seed-point tractography runs atclose to real time on this system using default parameters and seedingregions up to 6-mm radius, with increasing update delays as the seedingregion is enlarged.

Software Environment

The software discussed in this report has been implemented in the 3DSlicer (version 3) open source software platform (www.slicer.org). 3DSlicer is an open source research software package, facilitating rapidimprovement of image analysis techniques. A suite of software toolsknown collectively as the NA-MIC Kit support the core functionalityneeded in the implementation of the custom tools described here.41 Allcomponents of the NA-MIC Kit have been developed according to theopen source model of software development (http://opensource.org),meaning that they are available for download, free of charge and re-strictions, by any groups wishing to replicate or extend our results.A complete inventory of the software is available at the NA-MICWeb Site;for the work described here, modifications were made primarily to theTeem library (http://teem.sf.net) and the 3D Slicer application software.3D Slicer includes modules for the selection of structures of interest,including tumor segmentation, ROI analysis, and interactive DTI.

Tract Selection Tools

A primary goal of this research is to provide the surgeon with a detailedmental model of the preoperative condition of the tumor and sur-rounding anatomy. Our techniques rely on interactive 3D graphics andallow exploration of the local WM anatomy using structural or func-tionally defined ROIs as input. In addition, novel techniques allowa controlled exploration of clinically relevant portions of the diffusiondata set. Tract selection and display was performed 3 ways.

ROI From fMRI

Using seed points in the WM adjacent to cortical areas associated withfunctional activation can suggest the pathways associated with the elo-quent cortex that should be preserved during tumor removal. Trac-tography is performed using the coregistered (to DTI) functionalactivation as the seeding input voxels. Seed points are selected bysoftware at either fixed-distance intervals or a randomly distributed gridwithin the activation volume.

Distance Seeding From Tumor Segmentation

Tracts generated from seed points at the boundary of the tumorprovide an indication of the WM pathways in the immediate vicinity ofthe tumor. Segmentation of tumors was performed using a combinationof subvolume thresholding, threshold level set selection, and manualadjustments using label map drawing tools. The software allows gen-eration of tracts from an isocontour shell inward or outward from thesegmentation boundary, allowing exploration of the local WM anatomy.If seeding is done near the tumor, the presence of preserved tracts withinor near an infiltrating tumor can be demonstrated (see Figure 4D andvideo supplement, described later). This technique can also be applied toany structure in the scan, including normal or pathological tissue that canbe delineated and segmented. The method precomputes a distancetransform around a segmented model of the tumor boundary (or other

tissue). Isocontour surfaces of the distance transform represent all pointsat specific distances from the tumor margin. Tractography is updatedwith each distance change, interactively demonstrating fiber tracts thatpass within the specified distance of the tumor (within the limits of thetractography method). To allow interactive updates as the distance isvaried, the number of seeds per update is set to a user-defined maximumto minimize latency (varies with data set and computer configuration). Inpractice, 200 to 500 seed points provide a sufficient overview of regionaltractography. This method is usually followed by the use of dynamic seedpoints for specific tract selection.

Dynamic Seed Points

Our software allows the clinician to select arbitrary 3D points (dy-namic seed points) to define a cluster of seed points within an adjustablesphere around the selected center point. The location of the seed pointcan be interactively moved while the tractography is updated in real timesuperimposed with the multimodal image display. The user specifies thespherical seeding region size (millimeter radius around dynamic seedpoint) and intrasphere seeding step size (in millimeters). This mode ofexploration allows precise movement of the seeds with respect to theanatomy and allows the surgeon to build an intuitive understanding ofareas in the peritumoral volume that may contain critical WM tracts (seeFigures 1A through 1C and 3D). Because only a small volume is used asa seed, this tool reduces the visual clutter generated by larger seed regionsand can be used for precise probing of specific areas of interest.

Parameters

The methods detailed above provide a number of parameters tocontrol the tractography algorithm. Fractional anisotropy (FA) is a de-rived measure of the orientation of water diffusion on a per-voxel basis.FA ranges from 0 (isotropic) to 1 (fully anisotropic), with the minimumfor tractography computation generally set around 0.2 (patient, scanner,and anatomy dependent). Stopping tract curvature (degrees per milli-meter) controls the maximum radius of curvature of a tractography path;the tractography algorithm will not return paths that exceed this value(anatomy and algorithm dependent). Path length range restricts thedisplay of fibers to those within the given range and is useful to reducethe visual clutter associated with very short tracts or to specifically selectonly certain tracts. Seed-point spacing determines the minimum distancebetween starts of tractography computation. Integration step length setsthe distance between successive calculations of principal diffusion di-rection. For a detailed explanation of these parameters and the generalprinciples of DTI, please see elsewhere.16,17,19,21,34,36,38

RESULTS

The structural, fMRI, and DTI data were loaded into 3DSlicer for interactive tractography. The clinician (A.G.) was ableto use the tool immediately with assistance from the softwaredevelopers regarding workflow. The use of the software tool with5 patients whose images were retrospectively reviewed is dis-cussed, with illustrations for patients 1 through 4 and a supple-mental video (described later) for patient 5.Figure 1 shows images from the investigation of patient 1. In

this case, the tumor (green model) is adjacent to the centralsulcus, and particular interest was given to the location of the





corticospinal tract and its relationship to the tumor. Seeding wasperformed with dynamic seed points to explore the corticospinaltract (Figure 1A with cortex, Figure 1B without) and parietaltracts peripheral to the tumor (Figure 1C, displayed with the FAmap). For comparison, ROI seeding from right-hand motorfMRI activations (magenta, anterior and lateral to the tumor)shows a similar set of tracts demonstrated adjacent to the tumor(Figure 1D).

Figure 2 shows images from patient 2 showing tracts bothdisplaced by and within the tumor. Figure 2A shows segmen-tation of high-intensity tumor regions displayed with corticaloutline. ROI-based tractography was used for exploratory seedingwithin the brightest region of T2 (Figure 2B), showing antero-superior tracts consistent with displacement rather than infil-tration. Dynamically seeded tracts wrapping around the tumorare displayed against the FA map (Figure 2C), demonstrating inparticular a tract running through the T2-bright area of the lesion.ROI-based seeding from the cerebral peduncle was used to dem-onstrate the corticospinal tract (Figure 2D). Intraoperative electro-cortical stimulation points were recorded and used for retrospectivetract seeding (Figure 2E and 2F), giving results consistent with theintegrity of the wrapped tract in Figure 2C and supporting thelocation of the corticospinal tract in Figure 2D.

Figure 3 shows tractography including fMRI seeding results forpatient 3. Dynamically seeded tracts are observed (Figure 3A)

wrapped or displaced posterior to the tumor, with display againstthe FA map. Tractography was seeded from the Portugueseantonym activation (Figure 3B). Seeding from a tumor-proximatesubset of the English noun categorization task demonstratedtractography connections to several unseeded activation areasfrom the same task (Figure 3C). Figure 3D demonstrates in-tratumor seeding with manual dynamic seed points to exploretracts in the superior portion of the tumor near the margin (basedon T2 intensity), which could be at risk during resection.Figure 4 shows images from patient 4 using seeding from fMRI,

dynamic seed points, and a shell around the lesion. In Figure 4A,ROI seeding was performed on the basis of visual task fMRI(orange), which resulted in many tracts (red tracts) in addition tothe optic radiation being displayed, thus limiting the direct utilityof this approach in this case. The optic radiations (blue tracts) weremanually delineated using dynamic seed points from the lateralgeniculate nucleus, showing correlation with fMRI findings. Usingthe dynamic seed points was helpful in selecting limited tracts ofinterest from among those obtained by seeding from the fMRIanalogous to dissection in cadaveric study.42 Dynamic seed-pointexploration was used (Figure 4B) to locate tracts wrappingthe medial and superior margins of the tumor. The dynamicseeding shell method was used (Figure 4C) to show tracts at themedial tumor margin. A short video (see Video, SupplementalDigital Content 1, http://links.lww.com/NEU/A355) shows the

FIGURE 1. Images from patient 1 with left frontoparietal tumor. A and B, segmented tumor (green) with the relative location ofcorticospinal tract (red ellipse), with and without cortical surface rendering (pink). The tract was defined by manual seedingexploration near the tumor. C, posterior dynamic seeding shows parietal tracts in the tumor vicinity against the fractionalanisotropy map. D, region of interest seeding from right-hand functional magnetic resonance imaging activation area (magenta)shows diverse fiber projections.

GOLBY ET AL




interactive tractography being performed in patient 5. The videodemonstrates rotation of the volume and use of the dynamic seed-point method to develop appreciation for the inherently 3Dqualities of these tools. The initial clip introduces the scene, withcortical surface and lesion segmentations displayed prominently.For context, the next scene demonstrates the result of whole-brainseeding combined with spectral clustering of tracts using a distance-affinity measure.43 Next, the distance map function is used todemonstrate the creation of a seeding shell around the tumor toexplore the bulk organization of tracts near the tumor. Finally, thecorticospinal tract is located using manually placed dynamic seedpoints, and the tumor periphery is explored interactively to locatethe putative arcuate fasciculus.

DISCUSSION

We have demonstrated the use of a novel software tool thatallows the interactive querying of DTI data to define the locationand trajectory of WM tracts around the tumor and critical areasof the brain. Knowledge of the anatomy of the WM and therelationship to the tumor can be very helpful in planning andcarrying out surgical resections in the region of key cortical areasand associated WM tracts31 The presence of eloquent WM tractsin the vicinity of the tumor is a strong predictor of residualtumor.44,45 Clinically, it is critically important to distinguishbetween cases in which the lesion infiltrates the WM and cases inwhich the fibers are displaced by the lesion, because this

FIGURE 2. Images from patient 2 with large right frontoparietooccipital tumor. A, tumor segmentation (green) with brainoutline rendering (pink). B, seeding within high-intensity T2-bright area demonstrates tracts within tumor region. C, manuallyseeded infiltrating and displaced tracts against the diffusion tensor imaging (DTI) fractional anisotropy map.D, region of interestseeding from cerebral peduncle to identify corticospinal tract. Seeding region was outlined on several T2 slices, and tractographywas seeded within this volume. E and F, offline, postoperative tract seeding from intraoperative cortical stimulation locations(red spheres, positive). The yellow spheres represent enlarged seeding area around positive sites to seed DTI tracts that do not reachthe cortical surface.





determines the extent of resectability of the lesion. To the extentthat DTI is able to resolve tracts given the limitations imposed byresolution and anisotropy thresholds, this tool allows the clinicianto explore areas of potential infiltration with precise seed-pointplacement. However, this tool requires some understanding ofthe underlying DTI methods and inherent limitations. For basicvalidation, we recommend frequent comparison of the tractsdisplayed by our selection tools against both the FA map (eg,Figure 2C) and the tensor direction maps available in Slicer.Moreover, as with any imaging method, DTI tractography mustbe used to supplement fundamental clinical understanding andconsiderations.

Whereas the information provided by DTI is a great advanceover previous understanding of the WM that was based on roughextrapolations from normal anatomical knowledge, making senseof the results is challenging even for neurosurgeons accustomed tothinking in 3D and knowledgeable about WM anatomy ingeneral. We have developed this method in response to severalquestions that have arisen repeatedly when looking at DTI data.First, does a tract run in the lesion or just beyond it? Whenviewing 3D tractography data sets, this can be difficult to ap-preciate. Second, how far is a given tract from the edge of thelesion? The user can define the edge however he or she feels ismost clinically relevant. Third, which tracts are associated with

a particular area of cortex? Cortical areas may be defined on thebasis of a priori knowledge or any functional mapping techniquesuch as fMRI or magnetoencephalography. Fourth, at the mar-gins of the lesion, what are the various WM structures all around?The user can interactively move a dynamic seed point around toquery any region in detail.Our method of distance transform isosurfacing is conceptually

complementary to previous work based on hulls around fiberpathways by Enders et al.46 To better visualize the relationshipbetween fibers and tumor, that method creates a set of tractog-raphy paths that are visually simplified to a single hull of ellipticalcross section, with a radial size that is varied to represent a user-specified percentile of enclosed paths. In our method, rather thanvarying the size of the geometric representation of the fiberpathway, we vary the size of the seeding geometry around thetumor to capture fibers that traverse the region. A related exampleof interactive DTI analysis is given by Sherbondy et al.47 Thatmethod precomputes DTI tractography paths that may then beselected using interactively defined volumes of interest. Ourmethod computes tractography on request, allowing fine-tunedtract selection using dynamic seed points and facilitating thedistance-based tract selection method.This interactive visualization approach can be used with any

whole-brain tractography acquisition and analysis methods, thus

FIGURE 3. Images from patient 3 with left frontotemporal tumor. A, dynamically seeded tracts wrapping along the posteriortumor margin displayed against a diffusion tensor imaging fractional anisotropy map. B, seeding from the Portuguese antonymfunctional magnetic resonance imaging (fMRI) task activation. C, seeding from a near-tumor subset of fMRI activation for theEnglish noun task. The seeded area (pink) yields tracts connecting to unseeded areas within the same task activation (yellow). D,dynamic seed-point exploration demonstrates tracts of concern in the superior margin of the tumor against T2 with segmented3-dimensional tumor model (green semitransparent).

GOLBY ET AL




taking advantage of technical and theoretical improvements asthey are developed. Two-tensor models of DTI have been foundto resolve crossing fibers better than the single tensor model.48

This can be particularly relevant for demonstrating the lateralprojections of the corticospinal tract to the hand and face regionsof the motor cortex.49 Thus, as advances in tractography makeDTI increasingly accurate, robust, and useful, those results can beformulated and displayed optimally for surgical planning pur-poses by using this type of approach.

The software allows the use of fMRI to select functionallyrelevant tracts.50 Intact and sometimes functional WM tracts maybe found within some infiltrative low-grade tumors.10 fMRI-

derived ROIs have been shown to perform better than ana-tomically defined ROIs to delineate pyramidal tracts and thesuperior longitudinal fasciculus.51 Care must be taken wheninterpreting the output of this style of analysis because the dif-fusion signal becomes more isotropic as the WMmerges into graymatter, and there may be little or no real geometric overlapbetween detectable WM tracts and cortical activations. Twopossible approaches are to seed from a region representing adilated fMRI activation or to place and iteratively move arounddynamic seed points in the WM directly deep to the fMRI.Interactive approaches and selective visualization has been used

in a variety of fields, including engineering, manufacturing,architecture, and gaming.52,53 Various approaches have beentaken to visualize complex 3D volumetric data such as peelingaway layers to view inside the volume.54,55 Some approachesdepend on presenting the viewer with a vantage point that is notfixed, thus allowing the inference of 3D relationships. Othercommon approaches include the demonstration of only selectedportions of the structure. Our tool takes advantage of both ofthese general approaches to allow the viewer to gradually form aninternal representation of the complex 3D structure. The printedimages included here are poorly suited to fully demonstrating theadvantages of such an approach because of their static nature.Thus, we have included a video capture of a session in whichclinical data are queried. The viewer is able to gradually form aninternal representation of the data, which then makes both themoving images and the static images more readily interpreted.Although this approach can serve as a useful adjunct to other

DTI images, there are some potential pitfalls of this approach.The approach is prone to subjective decisions made by the personinterrogating the data and may be influenced by existing biases.Thus, if the user has not considered the likely presence of a criticaltract in the region, he or she may not visualize it using the tools.Such a subjective bias is always true of surgeon decision making.It is reasonable, therefore, to complement this approach withother visualizations such as 2-dimensional cross-sectional imageswith DTI glyphs that show the data in a less processed form.Nevertheless, an interactive approach such as that presented hereputs some of the decision making in the hands of the persondoing the surgery rather than a third party who may not be awareof all the clinical facts.Although the software developed for this research is available

for download, it is not approved as a medical device by the Foodand Drug Administration or any other authority. Investigatorsand institutions wishing to use the tools must take appropriatecare to establish appropriate research protocols. In our experience,effective use of these techniques requires that the research teaminclude 1 or more members with image analysis experience whoare responsible for ensuring that site-specific variability isaccounted for in the processing workflow. In particular, fMRIand diffusion-weighted imaging data acquisition is often thesource of incompatibilities resulting from vendor-dependent fileformats, variations in processing techniques, difficulties in imageregistration, and a reliance on research scanning protocols.

FIGURE 4. Images from patient 4 with right posteriortemporal tumor. A, case overview with segmented corticalsurface (pink), segmented tumor (green), and whole-fieldvision functional magnetic resonance image (fMRI; posterioryellow area). B, identification using dynamic seed points ofoptic radiations (blue) as a subset of tracts (red) seeded fromwhole-field vision fMRI (orange). C, region of interestseeding demonstrates tracts passing through the medial andanterior portions of tumor. D, distance-map seeding shell(5 mm) demonstrates tracts at the tumor margins.





CONCLUSION

Allowing the clinician to interact with DTI-based WM trac-tography preoperatively has the potential to provide helpful an-atomic and functional information that can make neurosurgicalresection safer and more effective. We developed a tool to allowclinicians to interactively query DTI tractography data to definefunctionally important and anatomically relevant WM tracts.

Disclosure

This work was supported by the National Institutes of Health(1U41RR019703-01A2 and P01-CA67165), the Brain Science Foundation, andthe Klarman Family Foundation. The authors have no personal financial orinstitutional interest in any of the drugs, materials, or devices described in thisarticle.

REFERENCES

1. Berger MS, Deliganis AV, Dobbins J, Keles GE. The effect of extent of resectionon recurrence in patients with low grade cerebral hemisphere gliomas. Cancer.1994;74(6):1784-1791.

2. Bradley WG. Achieving gross total resection of brain tumors: intraoperative MRimaging can make a big difference. AJNR Am J Neuroradiol. 2002;23(3):348-349.

3. Claus EB, Horlacher A, Hsu L, et al. Survival rates in patients with low-gradeglioma after intraoperative magnetic resonance image guidance. Cancer.2005;103(6):1227-1233.

4. Sanai N, Berger MS. Glioma extent of resection and its impact on patient out-come. Neurosurgery. 2008;62(4):753-764.

5. Talos I-F, Zou KH, Ohno-Machado L, et al. Supratentorial low-grade gliomaresectability: statistical predictive analysis based on anatomic MR features andtumor characteristics. Radiology. 2006;239(2):506-513.

6. Alberstone CD, Skirboll SL, Benzel EC, et al. Magnetic source imaging and brainsurgery: presurgical and intraoperative planning in 26 patients. J Neurosurg.2000;92(1):79-90.

7. Fried I, Nenov VI, Ojemann SG, Woods RP. Functional MR and PET imaging ofrolandic and visual cortices for neurosurgical planning. J Neurosurg.1995;83(5):854-861.

8. Tharin S, Golby A. Functional brain mapping and its applications to neurosurgery.Neurosurgery. 2007;60(4)(Suppl 2):185-201.

9. Witwer BP, Moftakhar R, Hasan KM, et al. Diffusion-tensor imaging of whitematter tracts in patients with cerebral neoplasm. J Neurosurg. 2002;97(3):568-575.

10. Skirboll SS, Ojemann GA, Berger MS, Lettich E, Winn HR. Functional cortex andsubcortical white matter located within gliomas. Neurosurgery. 1996;38(4):678-684.

11. Kinoshita M, Yamada K, Hashimoto N, et al. Fiber-tracking does not accuratelyestimate size of fiber bundle in pathological condition: initial neurosurgical ex-perience using neuronavigation and subcortical white matter stimulation. Neu-roImage. 2005;25(2):424-429.

12. Keles GE, Lundin DA, Lamborn KR, Chang EF, Ojemann G, Berger MS.Intraoperative subcortical stimulation mapping for hemispherical perirolandicgliomas located within or adjacent to the descending motor pathways: evaluationof morbidity and assessment of functional outcome in 294 patients. J Neurosurg.2004;100(3):369-375.

13. Duffau H, Capelle L, Sichez N, et al. Intraoperative mapping of the subcorticallanguage pathways using direct stimulations: an anatomo-functional study. Brain.2002;125(pt 1):199-214.

14. Duffau H, Capelle L, Denvil D, et al. Usefulness of intraoperative electricalsubcortical mapping during surgery for low-grade gliomas located within eloquentbrain regions: functional results in a consecutive series of 103 patients. J Neurosurg.2003;98(4):764-778.

15. Duffau H, Peggy Gatignol ST, Mandonnet E, Capelle L, Taillandier L. Intra-operative subcortical stimulation mapping of language pathways in a consecutiveseries of 115 patients with grade II glioma in the left dominant hemisphere.J Neurosurg. 2008;109(3):461-471.

16. Basser PJ, Mattiello J, LeBihan D. Estimation of the effective self-diffusion tensorfrom the NMR spin echo. J Magn Reson B. 1994;103(3):247-254.

17. Beaulieu C. The basis of anisotropic water diffusion in the nervous system:a technical review. NMR Biomed. 2002;15(7-8):435-455.

18. Mori S, Barker PB. Diffusion magnetic resonance imaging: its principle andapplications. Anat Rec. 1999;257(3):102-109.

19. Mori S, van Zijl PC. Fiber tracking: principles and strategies: a technical review.NMR Biomed. 2002;15(7-8):468-480.

20. Bello L, Gambini A, Castellano A, et al. Motor and language DTI fiber trackingcombined with intraoperative subcortical mapping for surgical removal of gliomas.NeuroImage. 2008;39(1):369-382.

21. Beppu T, Inoue T, Shibata Y, et al. Measurement of fractional anisotropy usingdiffusion tensor MRI in supratentorial astrocytic tumors. J Neurooncol. 2003;63(2):109-116.

22. Clark CA, Barrick TR, Murphy MM, Bell BA. White matter fiber tracking inpatients with space-occupying lesions of the brain: a new technique for neuro-surgical planning? NeuroImage. 2003;20(3):1601-1608.

23. Hendler T, Pianka P, Sigal M, et al. Delineating gray and white matter in-volvement in brain lesions: three-dimensional alignment of functional magneticresonance and diffusion-tensor imaging. J Neurosurg. 2003;99(6):1018-1027.

24. Lu S, Ahn D, Johnson G, Cha S. Peritumoral diffusion tensor imaging ofhigh-grade gliomas and metastatic brain tumors. AJNR Am J Neuroradiol.2003;24(5):937-941.

25. Price SJ, Burnet NG, Donovan T, et al. Diffusion tensor imaging of brain tumoursat 3T: a potential tool for assessing white matter tract invasion? Clin Radiol.2003;58(6):455-462.

26. Smits M, Vernooij MW, Wielopolski PA, Vincent AJPE, Houston GC, van derLugt A. Incorporating functional MR imaging into diffusion tensor tractographyin the preoperative assessment of the corticospinal tract in patients with braintumors. AJNR Am J Neuroradiol. 2007;28(7):1354-1361.

27. Tummala RP, Chu RM, Liu H, Truwit CL, Hall WA. Application of diffusiontensor imaging to magnetic-resonance-guided brain tumor resection. PediatrNeurosurg. 2003;39(1):39-43.

28. Wieshmann UC, Symms MR, Parker GJ, et al. Diffusion tensor imaging dem-onstrates deviation of fibres in normal appearing white matter adjacent to a braintumour. J Neurol Neurosurg Psychiatry. 2000;68(4):501-503.

29. Yamada K, Kizu O, Mori S, et al. Brain fiber tracking with clinically feasiblediffusion-tensor MR imaging: initial experience. Radiology. 2003;227(1):295-301.

30. Nimsky C, Ganslandt O, Merhof D, Sorensen AG, Fahlbusch R. Intraoperativevisualization of the pyramidal tract by diffusion-tensor-imaging-based fibertracking. NeuroImage. 2006;30(4):1219-1229.

31. Wu JS, Zhou LF, Tang WJ, et al. Clinical evaluation and follow-up outcome ofdiffusion tensor imaging-based functional neuronavigation: a prospective, con-trolled study in patients with gliomas involving pyramidal tracts. Neurosurgery.2007;61(5):935-948.

32. Nimsky C, Ganslandt O, Hastreiter P, et al. Preoperative and intraoperativediffusion tensor imaging-based fiber tracking in glioma surgery. Neurosurgery.2007;61(1)(Suppl):178-185.

33. Catani M, Howard RJ, Pajevic S, Jones DK. Virtual in vivo interactive dissectionof white matter fasciculi in the human brain. NeuroImage. 2002;17(1):77-94.

34. Conturo TE, Lori NF, Cull TS, et al. Tracking neuronal fiber pathways in theliving human brain. Proc Natl Acad Sci U S A. 1999;96(18):10422-10427.

35. Wakana S, Caprihan A, Panzenboeck MM, et al. Reproducibility of quantitativetractography methods applied to cerebral white matter. NeuroImage.2007;36(3):630-644.

36. Jellison BJ, Field AS, Medow J, Lazar M, Salamat MS, Alexander AL. Diffusiontensor imaging of cerebral white matter: a pictorial review of physics, fiber tractanatomy, and tumor imaging patterns. AJNR Am J Neuroradiol. 2004;25(3):356-369.

36a.Elhawary H, Liu H, Patel P, et al. Intraoperative real-time querying of white mattertracts during frameless stereotactic neuronavigation. Neurosurgery. 2011;68(2):506-516.

37. Pieper S, Halle M, Kikinis R. 3D Slicer: Biomedical Imaging: Nano to Macro, 2004.IEEE International Symposium; 2004;631:632-635.

38. Basser PJ, Pajevic S, Pierpaoli C, Duda J, Aldroubi A. In vivo fiber tractographyusing DT-MRI data. Magn Reson Med. 2000;44(4):625-632.

39. Yoo TS. Insight Into Images: Principles and Practice for Segmentation, Registration,and Image Analysis. Natick, MA: AK Peters Ltd; 2004.

GOLBY ET AL




40. Jezzard P, Balaban RS. Correction for geometric distortion in echo planar imagesfrom B0 field variations. Magn Reson Med. 1995;34(1):65-73.

41. Pieper S, Lorensen B, Schroeder W, Kikinis R. The NA-MIC Kit: ITK, VTK,pipelines, grids and 3D slicer as an open platform for the medical image computingcommunity. Biomedical Imaging: Nano to Macro, 2006. 3rd IEEE InternationalSymposium, 2006:698-701.

42. Ture U, Yasargil MG, Friedman AH, Al-Mefty O. Fiber dissection technique:lateral aspect of the brain. Neurosurgery. 2000;47(2):417-426.

43. O’Donnell LJ, Westin C-F, Golby AJ. Tract-based morphometry. Medical ImageComputing and Computer-Assisted Intervention: MICCAI: International Conference onMedical Image Computing and Computer-Assisted Intervention. 2007;10(pt 2):161-168.

44. Mandonnet E, Jbabdi S, Taillandier L, et al. Preoperative estimation of residualvolume for WHO grade II glioma resected with intraoperative functional map-ping. Neurooncology. 2007;9(1):63-69.

45. Talos IF, Zou KH, Kikinis R, Jolesz FA. Volumetric assessment of tumor in-filtration of adjacent white matter based on anatomic MRI and diffusion tensortractography. Acad Radiol. 2007;14(4):431-436.

46. Enders F, Sauber N, Merhof D, Hastreiter P, Nimsky C, Stamminger MS.Visualization of White matter tracts with wrapped streamlines. VisualizationConference, IEEE. Los Alamitos, CA: IEEE Computer Society; 2005:7.

47. Sherbondy A, Akers D, Mackenzie R, Dougherty R, Wandell B. Exploringconnectivity of the brain’s white matter with dynamic queries. Visualization andComputer Graphics, IEEE Transactions on. 2005;11(4):419-430.

48. Peled S, Friman O, Jolesz F, Westin C-F. Geometrically constrained two-tensor model for crossing tracts in DWI. Magn Reson Imaging. 2006;24(9):1263-1270.

49. Qazi AA, Radmanesh A, O’Donnell L, et al. Resolving crossings in the cortico-spinal tract by two-tensor streamline tractography: method and clinical assessmentusing fMRI. NeuroImage. 2008.

50. Guye M, Parker GJM, Symms M, et al. Combined functional MRI and trac-tography to demonstrate the connectivity of the human primary motor cortex invivo. NeuroImage. 2003;19(4):1349-1360.

51. Schonberg T, Pianka P, Hendler T, Pasternak O, Assaf Y. Characterization ofdisplaced white matter by brain tumors using combined DTI and fMRI.NeuroImage. 2006;30(4):1100-1111.

52. Li W, Agrawala M, Curless B, Salesin D. Automated generation of interactive 3Dexploded view diagrams. ACM SIGGRAPH 2008 papers. Los Angeles, California:ACM; 2008:1-7.

53. Li W, Ritter L, Agrawala M, Curless B, Salesin D. Interactive cutaway illustrationsof complex 3D models. ACM SIGGRAPH 2007 papers. San Diego, California:ACM; 2007:31.

54. Correa CD, Silver D, Chen M. Feature Aligned Volume Manipulation for Il-lustration and Visualization. Visualization and Computer Graphics, IEEE Trans-actions on. 2006;12(5):1069-1076.

55. McGuffin MJ, Tancau L, Balakrishnan R. Using Deformations for BrowsingVolumetric Data. Proceedings of the 14th IEEE Visualization 2003 (VIS’03): IEEEComputer Society; 2003:53.

Supplemental digital content is available for this article. Direct URL citationsappear in the printed text and are provided in the HTML and PDF versions of thisarticle on the journal’s Web site (www.neurosurgery-online.com).





interactive diffusion tensor tractography visualization for neurosurgical planning

Documents