Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).

Practical Scientific Visualization Examples

Russell M. Taylor IIDepartment of Computer Science

University of North Carolina, Chapel Hilltaylorr@cs.unc.edu http://www.cs.unc.edu/~taylorr/

Abstract

Scientific visualization is not yet a disciplinefounded on well-understood principles. In somecases we have rules of thumb, and there are stud-ies that probe the capabilities and limitations ofspecific techniques. For the most part, however, itis a collection ofad hoc techniques and lovelyexamples. This article collects examples wherevisualization was found to be useful for particularinsights or where it enabled new and fruitful typesof experiments.

Introduction

Many examples listed here are drawn from Kellerand Keller’s book Visual Cues, a valuable collec-tion of visualization techniques along with de-scriptions of which techniques apply each visuali-zation technique. (Keller and Keller 1993) Manyof the examples are from recent Case Studies pub-lished in the proceedings of the IEEE Visualiza-tion conference (a wealth of examples of actualvisualization systems in use). The rest of the ex-amples come from colleagues or the author’s ownresearch.

This work does not attempt to cover the range ofavailable techniques. Rather, it seeks to collectanecdotal and verified examples of when particu-lar benefits have come from particular visualiza-tion techniques. As possible, related examples aregrouped into categories. After each category,there is aNet section that summarizes the type ofresult the technique is best at producing.

Examples are restricted to visualizations for whichthere is an inherent spatial distribution to the data.This includes medical visualization but leaves outthe broad fields of algorithm visualization and in-

formation visualization that visualize abstract (of-ten many-dimensional) spaces.

Many excellent systems are unlisted because nopublished account was found of particular insightsgained using them, even when the visualizationtechniques used are obviously powerful. Spaceconstraints further limit the number of results pre-sented.

Visual Display

Viewing spatial data as spatial data

Lanzagorta and others at NRL looked at the in-ternal micro-structure ofsteel by polish-ing down onelayer at a timeand scanningeach with anSEM.(Lanzagorta,

Kral et al. 1998)They packed theslices back into 3Dand visualized themusing an immersiveenvironment (in theGROTTO). The goalwas to understandthe morphology anddistribution of grains

and precipitates within the steel. The 3D visuali-zation revealed new features: cementite precipi-tates that had appeared in 2D to be isolated werein fact in contact with austenite grain boundaries,and there was often a hole in the cementite grain

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).boundary film near the base of cementite precipi-tates that had gone unnoticed in 2D.

Ross and others at NASA Ames reconstructed animage of the 3D structure of mammalian gravity-sensing organs from multiple electron micro-graphs. The 3D image revealed for the first timethat the elements are morphologically organizedfor weighted, parallel-distributed processing ofinformation. (Keller and Keller 1993), p 155.

Volume rendering enabled the proper planning ofpartial, living-donor lung transplants. In this pro-cedure, one of the lobes of the donor’s lung istransplanted to the recipient. In order to avoiddamaging the neighboring lobe in the donor, sur-geons need to know just where to cut the bron-chial tubes and blood vessels. When this planningwas done based on CT slices, there was oftendamage to the neighboring lobe. Volume render-ing allowed the surgeries to be planned so thatthis could usually be avoided. (Aylward 1999)

Rupert and others at the Lawrence Livermore na-tional Laboratory used color coding to study theaccuracy of a material mix dynamic simulation;displaying the simulation results in the area of in-terest. Comparisons with expected behavior andexperimental results indicated areas where bugs inthe program caused unexpected behavior. (Kellerand Keller 1993), p. 44. Another example, whereCrowley and others found programming errors byvisual presentation of the field of study, is foundon page 71.

Klimenko at Freiburg University and others built asystem for the visualization of the sheets sweptout by the strings in subatomic string theory.(Klimenko, Nikitin et al.1999) They adjusted pa-rameters in their visualization system to explorepossible configurations, looking for singularities.This allowed them to formulate new hypotheses,which they were then able to prove analytically.

Students of Pizer and Coggins at UNC notedstriking similarities between the output of aCORE-based medial axis algorithm (Fritsch,Eberly et al. 1994) and that of code that performsstatistical classifications in feature space. "Theobservation showed us that the outputs of two

computations arising from disparate methodolo-gies with different histories were essentially thesame." This result has been replicated, with theresult that we know that the medial axes can befound by a specific set of operations in featurespaces. (Coggins 1999)

Net: Whereas projection and measurement tech-niques can provide better quantitative results toparticular known questions, viewing data in itsnatural spatial extent can provide insight and un-derstanding.

Viewing transformedspatial data as spatial data

The UNCRspaceprogram aids in the planning ofx-ray crystallography data collection. (Brooks,Thorvaldsdottir et al. 1988) For this task, a detec-tor that collects diffracted electrons must be posi-tioned for multiple collections to obtain the cor-rect amount of overlap between the sampled ar-eas. The geometry of detector attachment limitsthe possible positions and orientations. Whenplanning data collection, working around anEwald spherein reciprocal (diffraction) space al-lows collection over a fraction of the area; the restfollows by symmetry. This task is extremely diffi-cult when done in any straightforward way in 3DEuclidean space. Rspace allows the user to do theplanning directly in reciprocal space, making thetask easier to do without gaps and with enoughoverlap. This application allowed faster and betterplanning of collection strategies, and therefore hasbeen widely used. (Brooks 1999)

Parker and Samtaney at Princeton and Rutgersvisualized the results from a simulation of plasmaturbulence inside a fusion generator. (Parker andSamtaney 1994) They report that visualizationplayed a critical role in going from the "raw"nonlinear solution of these complex equations intoa simplified theoretical model explaining the es-sential underlying physics. "Through the 3D visu-alization we observed that the radial structure be-comes more elongated during nonlinear satura-tion, then quickly rips apart." To concentrate thesimulation resolution in areas that require it, thecalculations are performed on grids that followmagnetic field lines. They report that these “field-

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).line-following coordinates are non-orthogonaland get quite twisted … hence 3D visualizationbecomes essential in interpreting the results in realspace."

Seeger at UNC was developing a suite of hapticvisualization techniques (adhesion, friction,bumps, vibration, stiffness) for data display. Inorder to understand what the forces were doing,he displayed them graphically in a surface coordi-nate system. This showed the response of the sys-tem to forces normal to and tangent to the localsurface, and decoupled the surface geometry fromthe haptic computation. This visualization showedhim when the visualization code was misbehaving.(Seeger 1999)

Net: Mapping data from a simulation or experi-ment into the coordinate system that is most natu-ral for viewing it can make it easier to interpret.

Interaction with the model: natural view changes

Zuiderveld and others at University Hospital inUtrecht studied the effectiveness of a stereo vol-ume-visualization system in clinical use over a pe-riod of about three months. (Zuiderveld, Ooijen etal. 1996) They discuss three examples of clinicalresults found using the system. In the first, an un-usual piece of splenic artery was understood inthe interactive stereo view, where conventionalviews had not provided insight. In the second, asmall aneurysm was located in the interactive 3Dview that had not been seen in 2D and 3D viewson film. In the third, stereo viewing providedneeded insight into the complex anatomy sur-rounding an aneurysm of the aortic arch.

Andrzey Wojtczak, a biochemist visiting the UNCMolecular Graphics resource, worked for overtwo hours in a head-mounted display system witha wide-area tracker looking at his molecule “forthe first time”, though had seen it many times be-fore on a 2D screen. Due to his improved under-standing compared to previous views, he discov-ered that he had fit part of the molecule incor-rectly. (Brooks 1999)

Williams at UCLA had examined one data set us-ing pseudo-color images and cross sections formonths. Using the UNC nanoManipulator system

(a microscopy visualization system), he discov-ered an important feature of the data (graphitesheets coming out of the surface) within secondsof viewing a real-time fly-over of the data set.These sheets were visible in only a few frames ofthe fly-over, when the lighting and eye point werein just the right position to highlight them. (TaylorII, Robinett et al. 1993)

Using the UNCSculptsteerable molecular simula-tion system, Duke biochemist Jane Richardsonexamined theFelix custom-designed protein tounderstand what could make it stable. The keylies in the cross-bonding between four alpha heli-ces (the bonding of the two pairs making up eachsubunit was clear; understanding the bonding be-tween the two subunits was not). She couldn’t getit to work in the configuration that she initiallytried, so she used the online optimization to keepthe subunits together while she flipped one wholesubunit over to the other side. No automaticminimizer or simulated annealing method wouldhave tried this extreme configuration change, andshe discovered a better (lower-energy) solution inthe new configuration. The stability is due to ahandful of critical disulfide bonds between thesubunits. (Brooks 1999)

While building Sitterson Hall, the UNC ComputerScience department members were arguing withthe architects over whether a gathering space atthe receptionist window in the front lobby waswide enough to allow someone to pass a group ofpeople who were gathered there. To better under-stand the space, a virtual walkthrough was de-signed based on the blueprints. The flat-shadedwalkthrough, running at 1 frame every 4 seconds,convinced the department members that the areahad to be widened. This has proven to be a wisedecision. (Brooks 1999)

The Scientific Computing and Imaging group atthe University of Utah has developed a scientificproblem-solving environment,SCIRun, that candisplay (among other things) volume renderingsof medical data sets. (Parker, Weinstein et al.1997) A nearby company named Voxel has cre-ated a machine that makes 3D holograms frommedical images. The holograms can be overlaid

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).on the patient’s actual anatomy by viewing thepatient through the holographic plate. Natalie, thedaughter of Voxel’s owner, had a tumor rightabove her brainstem, a type that is nearly alwaysfatal due to the difficulty of removal. After thesurgeons had prepared their surgical plan usingtraditional slice-based visualization, they wereasked to revisit the plan using both holographicand immersive stereo volume-rendered display ofthe data. After an hour of review, the surgeonschanged their surgical plan. The operation was asuccess. (Johnson 1999)

Net: You can think of each frame in a real-timedisplay of the surface as a new filter applied to thedata set, with the user in control of the filter pa-rameters (viewpoint, lighting direction) throughnatural motions of head and hand. People areadept at understanding the structure of 3D sur-faces this way, having learned this skill over a life-time. Intuitive exploration can produce insight.

Combining multiple (measured or calculated) datasets

Figure XXX: Visualization of multiple parameters of measuredflow past an airfoil. Velocity, speed, vorticity, rate of strain anddivergence shear are all shown. Divergence (shown by ellipsearea) should be everywhere the same in an incompressible fluid;variations in ellipse radii are in the lower (zoomed) image indi-cate out-of-plane flow.

Kirby, Marmanis and Laidlaw at Brown Univer-sity show a layered visualization, where color lies

under ellipses, which lie under arrows. (R. M.Kirby and Laidlaw 1999) They have used this sys-tem to look at rate of strain (second-order ten-sor), turbulent charge (vector) and turbulentcharge (scalar) looking for correlations betweenthese and velocity and vorticity. A change of areain the ellipses within the visualization shown infigure XXX revealed that there was out-of-planeflow. They also report that the visualization posednew questions.

Nuuja and Lorig at the Pittsburgh Supercom-puting Center visualized the results of a time-dependent simulation of ozone over Los Angeles.The simulation was tied to ozone measurementsmade throughout the day and night. At night, themeasurements were dropping, so it appeared thatthe ozone concentration was dropping at nightand increasing again during the daytime. Thevisualization revealed that at night, ozone merelyfloats higher in the atmosphere rather than disap-pearing. (Keller and Keller 1993) p. 88.

The first protein whose structure was solved fromcrystallography data without first building aphysical model was Bovine Cu,Zn-superoxidedismutase in 1974 by Dave and Jane Richardsonusing the UNC GRIP system (a density fitting sys-tem). (Brooks 1999) Now, this is how proteinfitting is routinely done. People were using non-computer-based scientific visualization beforethen; they used regularly spaced sheets of Plexi-glas in a 3D cube, with the electron density sur-faces drawn on the sheets. A half-silvered mirrorsuperimposed a brass model of the protein (whoselinkages are known but whose exact shape in 3-space is not) on the volume. This apparatus wasknown as a Richards’ box, after its inventor FredRichards. The advantages to the electronic fittingsystem are that the density data can be recom-puted and redrawn automatically when the userhas improved the fit and that the coordinates forthe atoms are available upon completion (ratherthan having to be teased out with a plumb boband meter stick). The 3D embedding of the modelwithin the measured density is what allows thefitting to work at all.

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).

Kim and Bejczy at JPL overlaid a locally-computed simulation of the motion of a remote-controlled robot arm, where the visual from theremote channel had a 2-3 second delay in re-sponse to user action. The simulated arm position(which was updatedimmediately) was displayedin wire-frame on top of the delayed remote videoimage; this helped the user deal with the extremelatency in the remote feed by seeing a predictionof what the motion would be. (Keller and Keller1993), p. 171.

Smith and Littlefield at Lowell combined the in-formation in two MRI images to form a new im-age consisting ofglyphs(objects whose shape andcolor is determined by multiple data sets). Theshape of these glyphs formed a discernable texturepattern in the combination image that indicated ahot spot (metabolic center) in a tumor. This spotwas not visible in either of the original MRI im-ages. (Keller and Keller 1993), p. 177.

Pandit and others at the University of Utah simu-lated and visualized the energy absorption withina human head (from MRI scans) in response to acell-phone placed near the ear. (Pandit, McDer-mott et al. 1996) Seeing the relative location ofthe phone, antenna, head and energy distributionrevealed that the major contribution to powerdeposition was from the feed-point of the an-tenna. It also showed that peak absorption ratewas where the ear is in contact with the head(rather than the outside of the ear).

Montgomery and others at Stanford used a vir-tual-environment system to aid in the planning ofreconstructive surgery for a 17-year old boy whohad a tumor on the left side of his face.(Montgomery, Stephanides et al. 1998) Their sys-tem enabled the surgeons to compare the dam-aged side with mirror-image geometry of the in-tact side to determine the best fit and to produce atemplate for reconstruction. They also displayed asection of his hipbone for comparing curvatureswith the template and to produce a printed papertemplate to aid with harvesting of the bone. Thisimproved planning enabled the surgery to be per-formed in much less time than earlier surgeries,

and allowed the complete reconstruction to beperformed in one operation. (Reconstructions of-ten require several operations to make adjust-ments.)

Multimodal registration of MRI data with MRangiogram data allows the simultaneous viewingof the extent of a tumor within the brain and thepatient’s arteries and blood vessels. This allowssurgeons to determine which arteries should beblocked to cause tumor death and which shouldbe left open to avoid stroke. (Bullitt, Liu et al.1999)

During surgery, visual overlay of the surgical in-strument within a preoperative CT/MRI scan al-lows the surgeon to understand what is going oninside the real skull. Of course, the more modifi-cation (drain fluid or cut), the less registeredthings are because the brain is distorted comparedto the preoperative image. One visualizationshowed remaining tumor in the map image thathad been missed by visual inspection during sur-gery due to brain swelling. Another case involveda child who had a trapped fluid pocket on the sideof his head, hidden behind a jungle of blood ves-sels. The surgeon used a frameless system tonavigate her way through the blood vessels.(Bullit 1999)

Net: The simultaneous, registered display of mul-tiple data sets can provide improved understand-ing of the relationship between them. This hasbeen especially effective in medical visualization,for the comparison of radiographic data and ac-tual anatomy.

Other Visual Techniques

Highlighting critical values

Mapping a critical value to a color that contrastswith the display of other values can reveal smallerrors and unexpected structure. Watterberg thusdiscovered an unexpected Moire-like pattern in avisualization of a 2D field; the pattern indicatedthat regions which were supposed to be identi-cally zero in fact had areas with values very closeto zero. (Keller and Keller 1993), p. 107.

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).

Viewing a time series as atransformedtime series

Besenbacher and colleagues at the University ofAarhus studied the dynamic properties of surfacechanges by scanning with a Scanning TunnelingMicroscope (STM) at high speed during thechanges and then replaying the information as anadjustable-time movie. (Besenbacher, Jensen et al.1991) The movies in this case are computed off-line, and replayed at the original or other rates.They report that “Such information, which cannotbe obtained by any other means, is very decisivefor a full understanding of both the growth modeof reconstructed phases and the resulting staticstructure.” They provide several examples of par-ticular insights that were enabled by watching thetime series as a movie: They were able to deter-mine that the formation of O-Cu chains proceedsby adding rows to existing areas, using Cu atomsfrom step edges combining with O atoms diffusingon the surface; they discovered that the outermostrow in an island of chains is not stable; they alsofound that in Cu(110)-c(6x2)O growth, thec(6x2) phase grows isotropically.

Making the Invisible Visible

David Banks and colleagues at Mississippi Statedeveloped a system that visualizes the results ofapplying various lenses, slits and apertures to anoptical system. (Banks, Brown et al. 1998) Aninstructor who used the system reported that itwas superior for instruction to hands-on opticalbenches because it made the invisible (light travel-ing through the system) visible and it abstractedout the behavior of the system. (McNeil 1999)

Haptic Display

The haptic display section is grouped by project,then by technique. A more in-depth view of thefirst two projects (and other UNC projects notlisted here) can be found in (Brooks, Ouh-Younget al. 1990).

GROPE-1: An early haptic feedback applicationat UNC allowed the user to feel the effects of a2D force field on a simulated probe, and was usedto teach students in an introductory Physicscourse. (Batter and F.P. Brooks ) These experi-

ments used a 2D sliding-carriage device with ser-vomotors for force presentation. Experimentalresults showed that haptic feedback improved theunderstanding of field characteristics vs. a visual-only implementation (for students who were in-terested in the material). Students reported thatusing the haptic display dispelled previous mis-conceptions. They had thought that the field of a(cylindrical) diode would be greater near the platethan near the cathode, and they thought the gravi-tation vector in a 3-body field would always bedirected at one of the bodies.

Figure XXX: The Docker application simulates the forcesbetween a drug and its receptor site in a protein as the userguides it to the minimum-energy configuration.

Ming Ouh-young at UNC designed and built theDocker haptic feedback system to simulate theinteraction of a drug molecule with its receptorsite in a protein. Docker computes the force andtorque between the drug and protein due to elec-trostatic charges and inter-atomic collisions, pre-senting the force and torque vectors both visuallyand using haptic feedback. Experiment showedthat chemists could perform the rigid-body posi-tioning task required to determine the lowest-energy configuration of the drug up to twice asquickly with haptic feedback turned on comparedto using the visual-only representations. (Ouh-young, Beard et al. 1989; Ouh-young 1990) Themost valuable result from using this system isprobably the radically improved situation aware-

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).ness that serious users report. Chemists reportedthat when they felt the forces, they better graspedthe details of the receptor site and of why a par-ticular drug fit well or poorly.

Figure XXX: The nanoManipulatorapplication adds a virtual-reality interface to a scanned-probe microscope. Haptic feedbackallows the user to feel and modify the microscopic surface.

The nanoManipulator: The nanoManipulator(nM) provides an intuitive interface to scanned-probe microscopes, allowing scientists from a va-riety of disciplines to examine and manipulatenanometer-scale structures. (Taylor II, Robinett etal. 1993) The nM displays a 3D rendering of thedata as it arrives in real time. (Figure XXX) Usinghaptic display and control, the scientist can feelthe surface representation to enhance understand-ing of surface properties and can modify the sur-face directly. Studies have shown that the nMgreatly increases productivity. (Finch, Chi et al.1995)

The haptic feedback component of our system hasalways been exciting to the scientists on the team;they love being able to feel the surface they areinvestigating. However, it is during modificationthat haptic feedback has proved itself most useful,allowing finer control and enabling whole newtypes of experiments. Haptic feedback has provedessential to finding the right spot to start a modi-fication, finding the path along which to modify,and enabled greater precision than permitted bythe standard scan-modify-scan experiment cycle.(Taylor II, Chen et al. 1997)

Finding the right spot

Haptic feedback allows the user to locate objectsand features on the surface by feel when theymove the tip around near the starting point for amodification. Surface features marking a desiredregion can be located without relying only on vis-ual feedback from the previous scan (which is of-ten misaligned due to drift and other nonlinearitiesin the positioners). This allowed a collaborator toposition the tip directly over an adenovirus parti-cle, then increase the force to cause the particle todimple directly in the center. It also allowed thetip to be placed between two touching carbonfilaments in order to tease them apart. (Taylor II,Chen et al. 1997)

Finding the right path

The scanned image shows only the surface as itwas before a modification began. There is onlyone tip on an SPM: it can either be scanning thesurface or modifying it, but not both at the sametime. Haptic feedback during modification allowsone to guide changes along a desired path.

Figure XXX: In this sequence of images from left to right, a15nm gold ball (circled) is moved into a test rig. Haptic feedbackis used to feel as the ball is pushed and to determine when it hasslipped off the tip.

Figure XXX shows haptic feedback being used tomaneuver a gold colloid particle across a micasurface into a gap that has been etched into a goldwire. The colloid is fragile enough that it wouldbe destroyed by getting the tip on top of it withmodification force or by many pushes. This pre-

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).vents attempts to move it by repeated pro-grammed “kicks”. Haptic feedback allowed theuser to tune the modification parameters so thatthe tip barely rode up the side of the ball whilepushing it. This allowed the guidance of the ballduring pushing so only about a dozen pushes wererequired. (Taylor II, Chen et al. 1997)

Haptic feedback was also used to form a thin ringin a gold film. A circle was scraped to form theinside of the ring, leaving two “snow plow” ridgesto either side. By feeling when the tip bumped upagainst the outside of the outer ridge, anotherslightly larger circle was formed. This formed athin gold ring on the surface. (Taylor II, Chen etal. 1997)

A light touch: observation modifies the system

When deposited on the surface, carbon nanotubesare held in place by residue from the solution inwhich they are dispersed. On some surfaces, thetubes slide freely once detached from the residueuntil they contact another patch of residue or an-other tube. Even the light touch of scanningcauses them to move. Guiding the tip by hand andswitching between imaging and modificationforce, we have been able to move and re-orientone carbon tube across a surface and into position

alongside anothertube. Once set-tled against theother tube, it was

stable again andwe could resumescanning to imagethe surface. Hap-tic feedback andslow, deliberate hand motion allowed us to findthe tube at intermediate points when we could notscan. The fact that the surface cannot be imagedat intermediate stages prevents this type of ex-periment from being performed using the standard

scan-modify-scan cycle. (Taylor II, Chen et al.1997)

Net: We have found that haptic feedback givesimproved situational awareness in the applicationswe have built. It can give students a better under-standing of force fields and build a scientist’s in-tuition about a problem. Force feedback is espe-cially good for manipulation tasks, where the useris moving a real or simulated object as part of thetask.

Acknowledgments

Clearly, others did most of the work presentedhere. Additional examples (published or not)should be sent to vis_techniques@cs.unc.edu forinclusion in future versions of this work.

Major support for the haptic work at UNC hasbeen provided by grant RR02170 from the NIHNational Center for Research Resources, Freder-ick P. Brooks, Jr., PI. The other UNC work hasbeen supported by a wide range of agencies, no-tably the NSF and DARPA.

ReferencesAylward, S. (1999). Personal Communication.

Banks, D. C., J. M. Brown, et al. (1998). “Interactive 3D Visu-alization of Optical Phenomena.” IEEE Computer Graphics &Applications18(4): pp. 66-69.

Batter, J. J. and J. F.P. Brooks (1971). GROPE-I: A computerdisplay to the sense of feel. Information Processing, Proc. IFIPCongress 71. pp. 759-763.

Besenbacher, F., F. Jensen, et al. (1991). “Visualization of theDynamics in Surface Reconstructions.”Journal of Vacuum Sci-ence TechnologyB 9(2): pp. 874-877.

Brooks, F. P., Jr. (1999). Fourteen Years of Interactive Walk-throughs. ACM SIGGRAPH '99 Course Notes, course #20, LosAngeles, CA, ACM SIGGRAPH. pp. H1-H22.

Brooks, F. P., Jr. (1999). Personal communication.

Brooks, F. P., Jr., M. Ouh-Young, et al. (1990). Project GROPE -Haptic displays for scientific visualization. Computer Graphics:Proceedings of SIGGRAPH '90, Dallas, Texas. pp. 177-185.

Brooks, F. P., Jr., H. Thorvaldsdottir, et al. (1988). FourteenthAnnual Report Interactive Graphics for Molecular Studies.Chapel Hill, University of North Carolina Department of Com-puter Science.

Bullit, E. (1999). Personal Communication.

Published inComputer Graphics, ACM SIGGRAPH publications, Vol. 34, No. 1, February 2000.pp. 74-79 (and cover).

Bullitt, E., A. Liu, et al. (1999). “Registration of 3D CerebralVessels with 2D Digital Angiograms: Clinical Evaluation.” Aca-demic Radiology6: pp. 539-546.

Coggins, J. (1999). Personal Communication.

Finch, M., V. Chi, et al. (1995). Surface Modification Tools in aVirtual Environment Interface to a Scanning Probe Microscope.Computer Graphics: Proceedings of the ACM Symposium onInteractive 3D Graphics, Monterey, CA, ACM SIGGRAPH. pp.13-18.

Fritsch, D. S., D. Eberly, et al. (1994). Simulated cores and theirapplications in medical imaging. Information Processing inMedical Imaging, Proc. IPMI '95, Kluwer, Dordrecht, the Nether-lands. pp. 365-368.

Johnson, C. (1999). Personal Communication.

Keller, P. and M. Keller (1993). Visual Cues: Practical datavisualization. Los Alamitos, CA, IEEE Computer Society Press.

Klimenko, S., I. Nikitin, et al. (1999). Visualization in StringTheory. Hot Topics Proceedings of IEEE Visualization 1999: 29-32.

Lanzagorta, M., M. V. Kral, et al. (1998). Three-DimensionalVisualization of Microstructures. IEEE Visualization, ResearchTriangle Park, NC. pp. 487-490,574.

McNeil, L. E. (1999). Personal Communication.

Montgomery, K., M. Stephanides, et al. (1998). A Case StudyUsing the Virtual Environment for Reconstructive Surgery. IEEEVisualization, Research Triangle Park, NC. pp. 431-434.

Ouh-young, M. (1990). Force Display In Molecular Docking.Computer Science. Chapel Hill, University of North Carolina:Tech Report #90-004.

Ouh-young, M., D. V. Beard, et al. (1989). Force Display Per-forms Better Than Visual Display in a Simple 3-D DockingTask. Proceeding of IEEE Robotics and Automation Conference,Scottsdale, AZ. pp. 1462-1466.

Pandit, V., R. McDermott, et al. (1996). Electrical Energy Ab-sorption in the Human Head From a Cellular Telephone. IEEEVisualization, San Fransisco, CA. pp. 371-374.

Parker, S. E. and R. Samtaney (1994). Tokamak Plasma Turbu-lence Visualization. Visualization '94, Washington, D.C., IEEEComputer Society Press. pp. 337-340.

Parker, S. G., D. M. Weinstein, et al. (1997). The {SCIR}uncomputational steering software system. Modern Software Toolsin Scientific Computing. E. Arge, A. M. Bruaset and H. P. Lang-tangen, Birkhauser Press: 1-44.

R. M. Kirby, M. M. and D. H. Laidlaw (1999). Visualizing Mul-tivalued Data from 2D Incompressible Flows Using Conceptsfrom Painting. IEEE Visualization 99, San Fransisco, CA, IEEEPress. pp. 333-340.

Seeger, A. (1999). Personal Communication.

Taylor II, R. M., J. Chen, et al. (1997). Pearls Found on the wayto the Ideal Interface for Scanned-probe Microscopes. Visualiza-tion '97, Phoenix, AZ, IEEE Computer Society Press. pp. 467-470.

Taylor II, R. M., W. Robinett, et al. (1993). The Nanomanipula-tor: A Virtual-Reality Interface for a Scanning Tunneling Micro-scope. SIGGRAPH 93, Anaheim, California, ACM SIGGRAPH.pp. 127-134.

Zuiderveld, K. J., P. M. A. v. Ooijen, et al. (1996). ClinicalEvaluation of Interactive Volume Visualization. IEEE Visualiza-tion, San Fransisco, CA. pp. 367-370.