three-dimensional modeling and visualization of the cochlea on the

144 IEEE TRANSACTIONS ON INFORMATION TECHNOLOGY IN BIOMEDICINE, VOL. 4, NO. 2, JUNE 2000

Three-Dimensional Modeling and Visualization ofthe Cochlea on the Internet

Sun K. Yoo, Ge Wang, Jay T. Rubinstein, Margaret W. Skinner, and Michael W. Vannier

Abstract—Three-dimensional (3-D) modeling and visualizationof the cochlea using the World Wide Web (WWW) is an effectiveway of sharing anatomic information for cochlear implantationover the Internet, particularly for morphometry-based researchand resident training in otolaryngology and neuroradiology. In thispaper, 3-D modeling, visualization, and animation techniques areintegrated in an interactive and platform-independent manner andimplemented over the WWW. Cohen’s template shape with meancross-sectional areas of the human cochlea is extended into a 3-Dgeometrical model. Also, spiral computer tomography data of apatient’s cochlea is digitally segmented and geometrically repre-sented. The cochlear electrode array is synthesized according to itsspecification. Then, cochlear implantation is animated with bothidealized and real cochlear models. Insertion length, angular posi-tion, and characteristic frequency of individual electrodes are esti-mated online during the virtual insertion. The optimization of theprocessing parameters is done to demonstrate the feasibility of thistechnology for clinical applications.

Index Terms—Cochlear implantation, modeling, spiral CT, vi-sualization, WWW.

I. INTRODUCTION

T HE WORLD Wide Web (WWW) facilitates informationsharing on the Internet. The WWW is an aggregate of

information resources stored on computers around the worldlinked via the Internet. In particular, the WWW allows a varietyof three-dimensional (3-D) visual resources to be retrieved in acommon manner through the standardized virtual reality mod-eling language (VRML) 2.0/97 [1], independent of the specificcomputer systems used to access the Internet. The dynamic andinteractive visualization capabilities of the VRML 2.0/97 attractincreasing attention in medical education and training [2], [3].

Cochlear implantation is widely used to manage profoundsensorineural hearing loss [4], [5]. Since early research onelectrical stimulation of the auditory nerve in the deaf in the1960’s, more than 20 000 patients have received various typesof cochlear implants in the USA alone. Many researchersput an emphasis on the importance of 3-D visualization andmeasurement of the cochlear features [6]–[12]. Develop-

Manuscript received May 28, 1999; revised October 15, 1999. This work wassupported in part by the National Institutes of Health under DC03590.

S. K. Yoo is with the Department of Radiology, the University of Iowa Schoolof Medicine, Iowa City, IA 52242 USA, and with the Department of MedicalEngineering, Yonsei University College of Medicine, Seoul, Korea.

G. Wang and M. W. Vannier are with the Department of Radiology, the Uni-versity of Iowa School of Medicine, Iowa City, IA 52242 USA.

J. T. Rubinstein is with the Department of Otolaryngology–Head and NeckSurgery, the University of Iowa School of Medicine, Iowa City, IA 52242 USA.

M. W. Skinner is with the Department of Otolaryngology–Head and NeckSurgery, Washington University School of Medicine, St. Louis, MO 63130USA.

Publisher Item Identifier S 1089-7771(00)03951-0.

ment and application of speech processing strategies in acochlear implant may benefit from the precise knowledge ofthe positions of the individual electrode bands [5]. Althoughcochlear implantation has been accepted as the most successfulimplanted prosthesis to interface directly with the humansensory system, a Web-based interactive visualization andquantification tool for the cochlear implantation has not beendeveloped yet. Interactive visualization and quantificationof the modeled and real human cochlea on the WWW cangive new insight into complex cochlear implantation and therelationship between spectral characteristics and electrodelocation within the cochlea.

In this paper, cochlear modeling and visualization methodsare proposed for research and training in cochlear implanta-tion. A geometric model of an “average” human cochlea anda real human cochlea with vestibule and lateral semicircularcanal (LSCC) are represented and visualized on the WWW. Themultichannel cochlear electrode bands are synthesized and ani-mated to simulate the insertion of a cochlear implant electrodearray. Also, the features such as insertion depth, angle, and char-acteristic frequency corresponding to each electrode are imme-diately computed and displayed. Finally, the optimization of theprocessing parameters is done to demonstrate the feasibility ofthis technology for clinical applications.

II. M ETHOD AND MATERIALS

A. Generalized Cohen’s Model of the Cochlea

The parameters used to generate a 3-D cochlear model in-clude the two-dimensional (2-D) spiral central path, the axialheight of the cochlea, and the cross-sectional area. The coordi-nate system for matching the 3-D cochlear geometry is shownin Fig. 1, based on the mid-modiolar axis projection [11] andthe reference line drawn through vestibule, LSCC, and roundwindow. The mid-modiolar axis of the cochlea is defined as the

axis. The – plane is orthogonal to the mid-modiolar axis.The axis is selected to be orthogonal to the reference line andparallel to the intersection point of the round window on the ref-erence line.

Cohen’s template [12] is used to model the central path of thecochlea in polar coordinates in the projected– plane:

(1)

where is the radial distance from the spiral center,is theangle in degrees, and , and are constants. The meanvalues suggested by Cohenet al. [12] are used as constants in

1089–7771/00$10.00 © 2000 IEEE

YOO et al.: 3-D MODELING AND VISUALIZATION OF THE COCHLEA ON THE INTERNET 145

Fig. 1. The projected central path, the cross section, and the axial height areused to parameterize the geometrical model of the cochlea.

(1). The vertices on the projected central path are synthesizedfrom (relating to the basal end of the organ of Corti) [12]to 910.3 (corresponding to the 2-1/2 turn of the cochlea) witha constant angle increment of 36.

The values of the vertices along the projected central pathare obtained by lifting the projected central path along theaxiswith a constant speed

(2)

where is the angle in degree,is constant, and is the startingangular degree, 10.3. The mean value of the axial height, 2.75mm, reported in Ketten’s paper [8], is used to determine theconstant .

The cross sections of the generalized Cohen’s cochlea at eachvertex on the central path are approximated by circular disks andinterpolated using Gulya’s average data [13] for Scala Tympani(ST) and combined Scala Vestibuli and Scala Media (SV-SM) atthree locations, 12, 18, and 24 mm from the round window. Thediameter of the disk is obtained from the summed area of ST andSV-SM. Finally, the 3-D geometry of the generalized Cohen’scochlea is obtained by putting the cross sections through eachcorresponding vertex and perpendicular to the central path andconverting the structure into meshed polygons.

B. Real Model of the Human Cochlea

A spiral computer tomography (CT) scanner1 with a researchspiral CT software package was used to scan the human cochleawith 12-bit gray level and 0.1-mm reconstruction interval [4],[14]. The region growing method in combination with thresh-olding is applied to segment the vestibule, LSCC, and cochlearcanal from the round window using the Analyze image anal-ysis and visualization software package2 [15]. The segmenteddata are also converted into meshed polygons and aligned tothe coordinate system of the 3-D cochlear geometry. The cen-tral path is manually drawn, then automatically refined. In therefinement, the manually selected vertices are uniformly redis-tributed along the manually drawn central path, then each of theredistributed vertices is moved into the mass center of the cor-responding cross section perpendicular to the central path. Ar-tificial rays from each vertex of the central path are cast to in-

1Siemens Somatom PLUS-S, Siemens Medical Systems, Iselin, NJ, USA.2Mayo Clinic, Rochester, MN, USA.

tersect the meshed cochlea surface and find the mass center ofthe cross section [16]. The total length of the cochlear canal iscalculated by summing up the incremental arc lengths along thecentral path. Bredburg’s data [17] are interpolated to find theangle from the normalized (percentage) length at each vertexalong the central path.

C. Virtual Electrode Array

The script language 3D Studio MAX3 [18] allows genera-tion and manipulation of various geometrical objects. The vir-tual cochlear electrode array with 32 platinum round rings (thedistal 22 are electrodes) with diameters ranging from 0.3 to 0.5mm are synthesized, based on the specification of the Nucleuscochlear implant system4 [4]. The electrodes are separated by0.75 mm (electrode length of 0.3 mm and interelectrode dis-tance of 0.45 mm). Finally, the virtual electrode array is madeby mounting the bands on a curvilinear carrier with a diameterof 0.4 mm.

The insertion of the cochlear electrode array is simulated byanimating the synthesized virtual electrode array in the contextof either the generalized Cohen’s or real cochlear model. Theanimation along the central path is based on the progressivetransformation of the virtual electrode array object into anotherusing the morphing function of the 3D Studio MAX 2.5 [18].Ten different virtual electrode array objects along the centralpath are generated for morphing, in which the starting point onthe central path is changed to depict the electrode array inser-tion.

Greenwood’s place-to-frequency equation [19] is used to es-timate the characteristic frequencies corresponding to electrodelocations over the animation loop:

(3)

where is the characteristic frequency,is the location (in-sertion depth) of the electrode array in millimeters, and ,and are constants. Constants for the average human derivedby Greenwood [19] are applied.

D. Visualization on the Web

VRML 2.0/97 with Java script is chosen to support the 3-Dvisualization and the platform-independent development of themodeled cochlea on the WWW. VRML 2.0/97 is an open stan-dard for describing 3-D objects in conjunction with the WWW.Both Netscape Navigator and Microsoft Internet Explorer sup-ports platform independence and 3-D information sharing afterinclusion of the VRML browser.

An important limiting factor for Web-based 3-D visualiza-tion is the bandwidth of the WWW. The file size dictates boththe speed and download performance of the VRML browser onthe WWW. The optimization modifier in 3D Studio Max 2.5[18] is applied to the meshed polygonal objects of the cochlea,vestibule, and LSCC to reduce the number of polygons and im-prove the navigation performance using the VRML browser.The number of polygons is iteratively reduced until a reasonable

33D Studio MAX R2.5, Kinetix, San Francisco, CA, USA.4Cochlear Corporation, Englewood, CO, USA.


Fig. 2. The central paths of the human cochlea found by the automaticadjustment method and the 3-D generalized Cohen’s model compared toKetten’s Archimedian spiral model.

tradeoff between the display quality and the number of polygonsis obtained.

Texture mapping is applied to each electrode of the synthe-sized array to reduce computational complexity, since each ofthe 32 platinum rings of the synthesized array has the same geo-metric pattern. All of the polygonal objects are converted intoan indexed face set [1] to be consistent to the VRML 2.0/97format. The coordinate sequence of the synthesized electrodearray generated by morphing is transformed into a coordinatorinterpolator set [1] to be animated on the WWW. Finally, thecrease angle field [1] in the indexed face set is added to gen-erate a smooth display of sharp edges.

E. Manipulation on the Web

User interface tools include four boxes with different colorsand one sphere with a sliding bar, which are embedded into theVRML 2.0/97 file to interactively manipulate on the WWW:

• Red box: electrode selection among the 22 implanted elec-trodes;

• Yellow box: starting one loop animation of virtual elec-trode array insertion;

• Magenta box: setting transparency of the cochlea;• Green sphere: step-wise movement of the virtual electrode

array along the cochlear central path.Java script routines [1] are embedded in the user interface to

set the transparency of the cochlea, control the animation of theelectrode array insertion, as well as compute the insertion depth,angle, and characteristic frequency of each electrode during theanimation. Also, text objects [1] are generated by the user’s in-teraction, which contain descriptions and feature values and areimmediately displayed on the VRML browser.

III. RESULTS

Fig. 2 shows the comparison of the central paths between the3-D generalized Cohen’s cochlear model and Ketten’s cochlearmodel, based on an Archimedian spiral, and the real humancochlea. To precisely represent the real human cochlea, the re-gion of interest and the range of density distribution for thecochlea, LSCC, and vestibule were inspected to constrain thesegmentation space before applying the method described in

Fig. 3. Rendered views of the 3-D generalized Cohen’s model, the humancochlea, and the 3-D Archimedian (Ketten) model at different view points.

Section II-B. The incorrect segmentation near the ambiguousborder due to partial volume artifacts is manually modified byan otolaryngologist. The mean spiral constants and mean axialheight of the cochlea [8] are used to synthesize Ketten’s cochlearmodel. The calculated total lengths of three central paths are36.2, 20.3, and 30.2 mm, respectively. The central path of thehuman cochlea depicts a practical geometrical pattern, but thelength of the real cochlear model is smaller than those of theidealized cochlear models. This is due to the fact that the masscenters of the cross section used to find the central path in thehuman cochlea are located inside the cochlear canal, while thecentral path connects pillar cells of the organ of corti in his-tologic measurements [6], [7]. Actually, the length of the realcochlear model is close to the mean length, 18.29 mm, of thescala tympani inner wall [6]. On the other hand, the length ofthe generalized Cohen’s model approximates the mean lengthof the human organ of Corti, 35.58 mm, measured by Kawano[6], and the average cochlear length, 35 mm [19]. The length ofKetten’s model is smaller than that of the generalized Cohen’smodel, because the hook region is excluded in Ketten’s model.

Fig. 3 compares the 3-D rendered features from two differentviewing points. The overall shape of the generalized Cohen’smodel fits better to the real human cochlea than Ketten’smodel, because Ketten’s model does not accurately follow thelogarithmic pattern of the cochlear geometry near the base.Therefore, the generalized Cohen’s model is more desirablethan Ketten’s model to realistically display the cochlea on theWWW and precisely estimate the characteristic frequenciesbased on Greenwood’s formula. The 2-D geometrical modelderived by Cohen [12] to estimate the angular positions ofthe electrode bands cannot be well applied to estimate lengthinformation because the 2-D length is smaller than the actual3-D length [7].

To estimate the characteristic frequency on the Web-baseddisplay, the angular position versus the normalized cochlearlength of both the generalized Cohen’s model and the realhuman cochlea is compared to Bredburg’s distance map data[17] in Fig. 4. Bredburg’s distance map shows the relationshipbetween the length and angular measurement from the basal


Fig. 4. The Angular positions of the 3-D generalized Cohen’s model and thehuman cochlea compared to the Bredburg map.

Fig. 5. The characteristic frequencies of the 3-D generalized Cohen’s modeland the human cochlea compared to the Bredburg map.

end of the cochlea. The portion from the basal turn to themiddle turn of the central path in the real human cochleais empirically chosen, because the apical turn region is notsufficiently clear to accurately define the cross section due tothe limited resolution of the spiral CT scanner. The normalizedlength versus angle curve of the real human cochlea is veryclose to that of Bredburg’s data.

In the 3-D generalized model, the characteristic frequency isestimated from the angular position instead of the length, be-cause the coordinate system to calculate the angular positionis well defined. Particularly, the measured angular position isreferenced to the average distance of the Bredburg map data toavoid the use of length required in Greenwood’s length-to-fre-quency equation. Use of the angular position provides a simpleway of estimating the characteristic frequency in the cochlea.Angular measurement is very sensitive, however, to the locationof the exact center of the cochlea; accurate coordinate alignmentis somewhat difficult due to the limited resolution of spiral CTand the complexity of temporal bone anatomy. The normalizedlength of the real cochleas found by the automatic central pathadjustment method is well suited to Bredburg data, as shownin Fig. 5. Because the normalized length is independent of thechoice of the coordinate system, it can be more effective for esti-mation of the characteristic frequency than the angular mappingmethod.

As far as the display on the WWW is concerned, the perfor-mance depends on the downloading time and the navigationspeed. The former is related to the file size, while the latter

TABLE INUMBERS OFVERTICES AND FACES, VRML FILE SIZES,AND DOWNLOADING TIME OF A HUMAN COCHLEA AS

A FUNCTION OF THENUMBER OF OPTIMIZATION ITERATIONS

(a) (b)

(c) (d)

Fig. 6. Rendering quality with the optimizing modifier, which is for reductionof the number of polygons (see also Table I). (a) Case 1. (b) Case 2. (c) Case 3.(d) Case 4.

Fig. 7. An animated sequence of the synthesized electrode array insertionalong the cochlear central path. The time interval is normalized to [0, 1].

to the model complexity. Complexity is also related to thefile size. Table I shows the numbers of vertices and faces,the VRML file sizes, and downloading times of a humancochlea with a vestibule and LSCC by iteratively applying theoptimizing modifier [18]. The rendered quality associated withTable I is demonstrated in Fig. 6. The staircase artifacts can beseen in rendered images in (a) and (b), while shape distortionis evident in (d). On the other hand, (c) shows a smoothrepresentation without significant loss of cochlear geometry,which realistically describes the 3-D cochlea in relation to itssurrounding vestibule and LSCC. This subjective assessment isperformed on the Pentium 260 MHz PC (Dell Precision 410,


Fig. 8. Simulation of implant array insertion into the 3-D generalized Cohen’s model. The insertion length, angular position, and associated characteristicfrequency are instantly displayed, while the cochlear electrode array is animated.

Round Rock, TX USA) with a 10-Mbit/s Ethernet connection.It is found that downloading and navigating a large VRML fileare inconveniently slow if the number of faces is more than20 000 (cases 1 and 2). This is particularly true when changingthe viewpoint and controlling the animation. The navigatingspeed in case 3 is quite satisfactory at a minimal expense ofdisplay quality.

Fig. 7 shows the animated sequence of the synthesizedelectrode array. Two turns of the total 2-1/2 turn cochlearcentral path from the round window of the human cochlea are

selected for demonstration of the electrode array insertion,because cochlear implantation is inserted only partially to thedepth of 22–30 mm within the cochlea [5]. As the number ofmorphing targets is increased, the electrode array can preciselytrack the central path, but the number of coordinates to beinterpolated at each vertex of the polygonal representation ofthe electrode array is accordingly increased, which degradesthe navigation performance.

Figs. 8 and 9 (http://dolphin.radiology.uiowa.edu/ge/) showrendered views of the 3-D generalized Cohen’s model and


Fig. 9. Simulation of implant array insertion into the human cochlea. The insertion length, angular position, and associated characteristic frequency are instantlydisplayed, while the cochlear electrode array is animated. Vestibule and lateral semicircular canal assist orientation.

the real cochlea using WorldView 2.05 by changing thetransparency and inserting the electrode array, respectively.Setting transparency of the shaded surface of the cochlea allowsthe user to observe the cochlear canal and the synthesizedelectrode array simultaneously. Tools for interaction are activeicons (yellow, red, and magenta boxes and a green sphere),which give the user convenience in navigating the cochlea onthe WWW using the open VRML browser. Text objects are

5Intervista, San Francisco, CA, USA.

included for displaying the length, angle, and characteristicfrequency.

IV. DISCUSSION

The geometrical cochlear modeling technique is valuable tocompensate for the individual variation of the human cochlea,especially the cochlear canal length [20]. The 2-D geomet-rical modeling technique developed by Cohen [12] describesthe shape of an implanted cochlea array in a radiographic


image with a template spiral curve. Cohen’s model provides asimplified method to measure angular positions of implantedelectrodes. However, Cohen’s 2-D method is limited when usedwith tomographic imaging modalities, such as spiral CT andMRI [9]. Ketten [8] devised a 3-D method with an Archimedianspiral approximation and a uniform-axis elevation, which canovercome the 2-D drawback and measure the cochlear canallength, but does not precisely depict the cochlear morphometrynear the basal end, as shown in Figs. 2 and 3. Therefore, theextension of Cohen’s 2-D spiral template to a 3-D templaterepresents a way to combine the strengths of both Cohen’sand Ketten’s methods and has the potential to be refined into aframework for accurate 3-D modeling.

In multichannel cochlear implants, an electrode array isinserted to stimulate various auditory nerve fibers with differentfrequencies depending on electrode locations. Greenwooddeveloped the location-to-frequency equation, which might beused to determine the characteristic frequency of the signalas a function of the electrode insertion depth [5]. Skinneretal. [11] suggested that accurate specification of the cochlearcanal length may provide a basis for determining the insertionlength of each electrode, because fitting speech processorsmight be optimized with such information. However, there isa substantial variation in the length and radius of the humancochlear duct [12]. Some investigators [10], [12] reportedthat the electrode position may be specified in terms of thespiral angular parameter instead of the spiral length. Theyestimate the characteristic frequency from angular coordinatesof the electrodes. Because Greenwood’s formula gives thecharacteristic frequency as a function of the length along theorgan of Corti, Bredburg’s map data [17] relating the angularparameter to the spiral length is utilized to map measured angleto length. However, a subjective choice of the feature points isprone to errors due to resolution limitation and interobservervariability, which may lead to a large variance in measuredangles. On the other hand, the normalized length used toestimate the characteristic frequency in the human cochleais highly consistent with Bredburg’s data, as shown in Figs.4 and 5. Use of the normalized length rather than either theactual length or the spiral angle is preferable, because neitherspecification of anatomical landmarks nor consideration ofindividual length variation is needed for estimation of thecharacteristic frequencies.

Our application of Web technology has several distinctfeatures. The 3-D visualization of synthesized and real cochlearmodels facilitates direct comparison between the averagehuman cochlea and a specific patient’s cochlea. Moreover,interactive animation of an implant array insertion and gradualchange of surface transparency may improve the capabilityfor surgeons in training to comprehend the spatial relationshipbetween an implant array and the cochlea. Particularly, in thereal human model, surrounding structures such as the vestibuleand lateral semicircular canal give not only the informationto establish the cochlear coordinate system but also surgicallandmarks. Quantitative information about the length, angle,and characteristic frequency of each electrode is particularlyvaluable to understand how the insertion depth might affectthe individual performance with a cochlear implant. Finally,

the computer interface is quite user-friendly; all visualizationoperations can by done by simple clicking and/or draggingwith a mouse, independent of the computer platform.

Our web-based modeling and visualization techniques can befurther developed in several aspects. First, spiral CT image res-olution should be refined to clearly show the apical turn region,as well as to distinguish ST, SV, and SM. It is recently reportedthat spiral CT image resolution can be significantly improved byiterative deblurring [14], [21]. Also, a fully automatic segmen-tation algorithm should be developed to accelerate the modelingprocess, because the segmentation with manual refinement re-quires more than 30 min. Then, the path planning strategy for in-sertion of an implant electrode array should be refined, becauseit is generally inserted into the scala tympani instead of alongthe central path of the cochlear canal [5]. Furthermore, aninvivo3-D distribution of implanted electrodes should be obtainedto compensate for variation in interelectrode distances due tobending of the array. We are developing an X-ray stereo-pho-togrammetric method to localize implanted electrodes in 3-D,and combine this knowledge into spiral CT data [22]. Finally,extensive work should be done to register CT and MRI imagevolumes to include more surrounding structures and better de-fine the scala tympani within the cochlea.

In conclusion, we have demonstrated the feasibility forapplying the WWW technology for simultaneous visualizationand analysis of the human cochlea. Efficient navigation andplatform independence can be achieved by VRML 2.0/97 withJava Script on a regular PC. Our central path-finding algorithmin combination with the 3-D generalization of Cohen’s 2-Dspiral template has the potential to accurately map implantedelectrodes to characteristic frequencies of the adjacent auditoryneurons. The Web-based capability to visualize the cochleaand estimate its features with a user-friendly interface isdesirable for both research and education in otolaryngologyand neuroradiology.

REFERENCES

[1] The virtual reality modeling language, in The VRML Consortium, 1997.[2] D. C̀osic̀, “An open medical imaging workstation architecture for

platform-independent 3D medical image processing and visualization,”IEEE Trans. Inform. Technol. Biomed., vol. 1, pp. 279–283, Dec. 1997.

[3] P. A. Warrick and W. R. J. Funnell, “A VRML-based anatomical vi-sualization tool for medical education,”IEEE Trans. Infrom. Technol.Biomed., vol. 2, pp. 55–61, June 1998.

[4] G. Wang, M. W. Vannier, M. W. Skinner, W. A. Kalender, A. Polacin,and D. R. Ketten, “Unwrapping cochlear implants by spiral CT,”IEEETrans. Biomed. Eng., vol. 43, pp. 891–900, Sept. 1996.

[5] P. C. Loizou, “Introduction to cochlear implants,”IEEE Eng. Med. Biol.Mag., pp. 32–42, Jan./Feb. 1999.

[6] A. Kawano, H. L. Seldon, and G. M. Clark, “Computer-aided three-dimensional reconstruction in human cochlear maps: Measurement ofthe lengths of organ of Corti, outer wall, inner wall, and Rosenthal’scanal,”Ann. Otol. Rhinol. Laryngol., vol. 105, pp. 701–709, 1996.

[7] A. Takagi and I. Sando, “Computer-aided three-dimensional recon-struction: A method of measuring temporal bone structures includingthe length of the cochlea,”Ann. Otol. Rhinol. Laryngol., vol. 98, pp.515–522, 1989.

[8] D. R. Ketten, M. W. Skinner, G. Wang, M. W. Vannier, G. A. Gates, and J.G. Neely, “In vivomeasures of cochlear length and insertion depth of Nu-cleus cochlear implant electrode arrays,”Ann. Otol. Rhinol. Laryngol.,vol. 107, no. 12, pp. 1–16, 1998.

[9] H. Takahashi and I. Sando, “Computer-aided 3D temporal bone anatomyfor cochlear implant surgery,”Laryngoscope, vol. 100, pp. 417–421,1990.


[10] M. A. Marsh, J. Xu, P. J. Blamey, L. A. Whitford, S. A. Xu, J. M. Sil-verman, and G. M. Clark, “Radiological evaluation of multiple-channelintracochlear implant insertion depth,”Am. J. Otol., vol. 14, pp.386–391, 1993.

[11] M. W. Skinner, D. R. Ketten, M. W. Vannier, G. A. Gates, R. L. Yoffie,and W. A. Kalender, “Determination of the position of nucleus cochlearimplant electrodes in the inner ear,”Amer. J. Otol., vol. 15, no. 5, pp.644–651, 1994.

[12] L. T. Cohen, J. Xu, S. A. Xu, and G. M. Clark, “Improved and simplifiedmethods for specifiying positions of the electrode bands of a cochlearimplant array,”Amer. J. Otol., vol. 17, pp. 859–865, 1996.

[13] A. J. Gulya and R. L. Steenerson, “The scala vestibuli for cochlear im-plantation: An anatomic study,”Arch. Otolaryngol. Head Neck Surg.,vol. 122, pp. 130–133, 1996.

[14] G. Wang, M. W. Vannier, M. W. Skinner, M. G. P. Cavalcanti, and G. W.Harding, “Spiral CT image deblurring for cochlear implantation,”IEEETrans. Med. Imag., vol. 17, pp. 251–262, 1998.

[15] R. A. Robb and ANALYZE™et al., Reference Manual: BiomedicalImaging Resource, Mayo Foundation, 1993.

[16] S. K. Yoo, G. Wang, J. T. Rubinstein, M. W. Skinner, and M. W. Van-nier, “Cochlear modeling and visualization on the internet,” inProc.CARS’99, Computer Aided Radiology and Surgery, 1999.

[17] G. Bredburg, “Cellular pattern and nerve supply of the human organ ofcorti,” Acta Otolaryngol. (Stockh), pp. 1–138, 1968.

[18] Kinetix, 3D Studio MAX R2 User’s Guide. San Francisco, USA: Au-todesk Inc., 1997.

[19] D. D. Greenwood, “A cochlear frequency-position function for severalspecies—29 years later,”J. Acoust. Soc. Amer., vol. 87, no. 6, pp.2592–2605, 1990.

[20] Q. Fu, R. V. Shannon, and X. Wang, “Effects of noise and spectral res-olution on vowel and consonant recognition: Acoustic and electricalhearing,”J. Acoust. Soc. Amer., vol. 104, pp. 1–11, 1998.

[21] G. Wang, M. W. Skinner, and M. W. Vannier, “Temporal bone volumetricimage deblurring in spiral CT,”Academic Radiol., vol. 2, pp. 888–895,1995.

[22] G. Wang, M. W. Skinner, J. T. Rubinstein, M. A. Howard, and M. W.Vannier, “Digital X-ray stereophotogrammetry for cochlear implanta-tion,” IEEE Trans. Biomed. Eng., to be published.

Sun K. Yoo received the B.S., M.S., and Ph.D. de-grees in electrical engineering from Yonsei Univer-sity in 1981, 1985, and 1989, respectively.

He was an Assistant Professor from 1990 to1994 of electrical engineering, SoonchunhyungUniversity. He is now an Assistant Professor ofbiomedical engineering, Yonsei University, and aVisiting Scholar at the Department of Radiology,University of Iowa, Iowa City. His interests include3-D visualization, telemedicine, and real-time imageprocessing.

Ge Wangreceived the Ph.D. degree.He is an Associate Professor with the Department

of Radiology, University of Iowa, Iowa City. Hisinterests are computed tomography (CT) and imageanalysis, with emphasis on spiral/helical CT. Hehas written 64 journal papers and numerous otherpublications. He serves as an Associate Editor ofMedical Physicsand has served as Guest Editor fora special issue of IEEE TRANSACTIONS ONMEDICAL

IMAGING (devoted to multi-slice spiral CT).Dr. Wang received the 1996 Hounsfield Award

from the Society of Computed Body Tomography and Magnetic Resonance,the 1997 Giovanni DiChiro Award for outstanding Scientific Research fromthe Journal of Computer Assisted Tomography, and the 1999 Medical PhysicsTravel Award from the American Association of Physicists in Medicine(AAPM) and the Institute of Physics and Engineering in Medicine (IPEM).

Jay T. Rubinstein received the Sc.B. and Sc.M. de-grees in electrical engineering from Brown Univer-sity, Providence, RI, in 1981 and 1983, respectively,and the M.D. and Ph.D. degrees in bioengineeringfrom the Medical Scientist Training Program at theUniversity of Washington, Seattle, in 1988.

He completed a surgical internship at Beth IsraelHospital and otolaryngolgy residence at the Massa-chusetts Eye and Ear Infirmary with post-doctoral re-search training in the Harvard Medical School’s De-partment of Otology and Laryngology. In 1995, he

completed a clinical fellowship in otology/neurotology at the University of Iowahospitals and clinics. Since then, he has been an Assistant Professor of Otolaryn-gology and Physiology and Biophysics at the University of Iowa, Iowa City. Hisclinical interests encompass disease of the ears and temporal bone. His researchinterests include physiological modeling of the electrically stimulated auditorysystem, and determinants of speech perception with cochlear implants.

Margaret W. Skinner received the B.A. degree inchemistry from Wellesley College, Wellesley, MA,the M.A. degree in audiology from Case-Western Re-serve University, Cleveland, OH, and the Ph.D. de-gree in audiology from Washington University, St.Louis, MO.

She is an audiologist in the Department ofOtolaryngology–Head and Neck Surgery, Wash-ington University School of Medicine, St. Louis,MO, where she is a Professor and Director of theAdult Cochlear Implant Program. Her classic book,

Hearing Aid Evaluation, (Englewood Cliffs, NJ: Prentice-Hall, 1988) integratesher broad clinical experience with her research on what hearing aid character-istics provide hearing-impaired individuals with optimized speech recognition.In her current research, she examines what speech coding strategies and speechprocessing characteristics will provide profoundly deaf adults implanted withmultilectrode intracochlear implants with the best opportunity to recognizespeech. Knowledge of electrode position in the inner ear may prove importantin the selection of coding strategy and processing parameters. She has authored48 publications in peer-reviewed journals or books.

Michael W. Vannier received the M.D. degree fromthe University of Kentucky School of Medicineand completed a diagnostic radiology residency atMallinckrodt Institute of Radiology, WashingtonUniversity School of Medicine, St. Louis, MO.

He is a Professor and Chairman at the Depart-ment of Radiology, University of Iowa School ofMedicine, Iowa City. His primary research interestsare morphometry and anthropometry based onvolumetric imaging modaliters, especially CT, MR,and PET/SPECT. He has authored more than 300

scientific publications.Dr. Vannier was inducted into the U.S. Space Foundations’ Hall of Fame for

work on digital medical imaging in 1994 and serves as Editor-in-Chief of theIEEE TRANSACTIONS ONMEDICAL IMAGING.

three-dimensional modeling and visualization of the cochlea on the

Documents