OMPUTER image guidance for surgical navigation hasentered standard neurosurgical practice in recentyears with the widespread availability of image

guidance systems. Both early and recent studies regardingthe accuracy of such systems usually relied on laboratoryphantom studies for the assessment of potential intraopera-tive accuracy.15,23,28 Authors of most clinical reports relatethe calculated accuracies of the systems, and when an esti-mate of true intraoperative surgical accuracy is given, it de-pends on subjective factors such as the impression of thesurgeon or on targets that are either too large for quantifi-able measures (such as a tumor)17 or inherently unreliable innature and not situated within the cranium (for example, ad-hesive scalp fiducial markers).20,25 Placement of depth elec-trodes for the investigation of intractable epilepsy has beenused in routine practice in epilepsy surgery centers, espe-cially for cases in which noninvasive studies are insufficientfor localizing a consistent pattern of seizure onset thatwould be amenable to surgical treatment.2,9,19,21,24 There hasbeen variability between centers about the indications forinvasive monitoring, although in recent years some degreeof agreement has existed. It has been my policy to consider

invasive monitoring in patients in whom bitemporal inde-pendent seizures have been recorded from scalp and sphe-noidal electrodes, in those in whom there is a discrepan-cy between the imaging and electrophysiological data, andin potential surgical candidates in whom seizure onset hasbeen poorly localized electrically. Patients in whom inva-sive monitoring is used represent a small subset of the pa-tients considered for surgery at the medical facility at whichI am employed.

The classic method of evaluating patients with suspectedepilepsy by using stereotactically placed depth electrodes(also known as stereo-EEG) is dependent on the placementof a stereotactic frame and implantation of electrodes. Ini-tially this was performed by reference to atlas coordinates;in subsequent years it was performed according to data ob-tained using stereoangiography, then CT, then MR imaging,and in some centers, a combination of some or most ofthese.2,9,19,21,24 After a transitional period of image guidancein parallel with frame-based stereotaxy, purely framelessplacement of depth electrodes was reported in 1995 (CRMascott, et al., unpublished data). There have been subse-quent series of image-guided (frameless) placement ofdepth electrodes with claims of good accuracy.16,18 In mypractice, I have taken advantage of the routine acquisition ofvolumetric MR images following depth electrode place-ment to fuse these data sets with preoperative data sets con-taining planned electrode trajectories. By measuring the

J. Neurosurg. / Volume 105 / October, 2006

J Neurosurg 105:561–567, 2006

In vivo accuracy of image guidance performed using opticaltracking and optimized registration

CHRISTOPHER R. MASCOTT, M.D., F.R.C.S.(C)

The Epilepsy Institute, Beacon Neurosciences, Beacon Hospital, Dublin, Ireland

Object. Image guidance systems involving the use of frameless referencing of surgical space to compile volumetricimaging data sets recently have come into widespread use. Few studies have addressed the true intraoperative surgicalaccuracy (that is, the application accuracy) of these systems except in a subjective manner. Calculated accuracies givenby the systems do not necessarily reflect true intraoperative accuracy.

Methods. To objectively assess the stereotactic accuracy of a frameless image guidance system using optical spatialreferencing, the author analyzed postoperative magnetic resonance (MR) images after placement of depth electrodesfor the investigation of epilepsy. Preoperative planning for the treatment of seven patients included implanting skullfiducial screws and obtaining computed tomography/MR fusion images by using ImMerge image fusion software onthe StealthStation (Medtronic, Inc.). A total of 42 electrodes were placed. Postoperative volumetric MR images werefused with preoperative study images. The difference between the planned electrode trajectories and targets and thevisualized electrodes was measured in stereotactic space.

Conclusions. The mean distance between the distal electrode contact and the distal end of the planned trajectory forthe 42 targets was 3 6 1.5 mm. The most common error was in depth. The author’s technique did not involve rigidskull fixation of electrodes because they were subsequently tunneled subcutaneously and later removed at the bedsideof the patient. Errors in depth were known to be due to traction at the time of tunneling and not due to stereotactic fac-tors. Correcting for depth along the electrode trajectory, the mean accuracy was found to be 2.4 6 1 mm.

KEY WORDS • imaging-guided surgery • electrode implantation • depth electrode •frameless stereotaxy

C

561

Abbreviations used in this paper: CSF = cerebrospinal fluid; CT = computed tomography; EEG = electroencephalography; MR =magnetic resonance.

distance between the visualized electrode and the plannedelectrode, it has been possible to estimate the in vivo accu-racy of the image guidance system.

Clinical Material and Methods

Depth electrodes were implanted in seven patients (sixelectrodes per patient) who had pharmacologically intract-able epilepsy. Indications included independent bitemporalseizures recorded using scalp and sphenoidal electrodes inthree patients, discrepancy between imaging findings andEEG or amobarbital test data in two patients, and poor lo-calization and lateralization due to muscle artifact in twopatients. Based on the stereotactic and image guidance lit-erature and my past experience, optimization of parametersinfluencing accuracy was performed. The methods used in-cluded the following: 1) the acquisition of both CT and MRimaging volumetric data sets for planning and guidance,which was done because CT usually has greater accuracythan MR imaging in stereotaxy given a lack of distortion;1,4,

5,26 2) implantation of skull-based fiducial markers prior toimage acquisition and surgery;11,14,27 and 3) performance ofsurgical registration by using implanted fiducial markers aswell as using the bridge of the nose as an anatomical land-mark to lower the center of the sphere of optimal accuracy.

Four fiducial screws (Stryker Leibinger Micro Implants,

Kalamazoo, MI) were implanted in the skull per patient asan outpatient procedure after scalp infiltration with a localanesthetic containing epinephrine. This was performed onthe same day as image acquisition and 3 to 5 days before theimplantation of electrodes. Both MR imaging and CT mark-ers were attached to the implanted screws before imagingand were removed subsequently. The implanted screws re-mained in place until the time of surgery with minimal orno adverse effects. Acquisition of CT scans was performedaccording to a routine contrast-enhanced volumetric im-age guidance protocol by using a 1-mm slice thickness ona Lightspeed scanner (General Electric Medical Systems,Waukesha, WI). Volumetric acquisition of MR images wasperformed using a spoiled gradient–recalled acquisitioncontrast-enhanced sequence with a 1.5-mm slice thicknesson a Signa machine (256 3 256 matrix; General ElectricMedical Systems). Data sets were transferred to a Stealth-Station (Medtronic, Inc., Louisville, CO) in the usual man-ner and image fusion between the CT and MR images wasperformed using ImMerge software (Medtronic, Inc.) withthe CT sequence as the primary data set. All planning wasverified on both CT and MR imaging data sets for accuracyand correlation. Planning was performed before the day ofsurgery, with targets selected and trajectories saved on theimage guidance system (Fig. 1).

On the day of surgery, the patients received general en-

C. R. Mascott

562 J. Neurosurg. / Volume 105 / October, 2006

FIG. 1. Three-dimensional planning model for electrode placement with three electrodes per side. Electrodes wereinserted through pediatric bur holes only. Targets were the amygdala (A) and hippocampus (B and C). Additional subduralstrips for lateral coverage were used but not illustrated on the planning model. The implanted skull fiducial markers canbe seen on the model (arrows).

dotracheal anesthesia, were immobilized using a Mayfieldheadholder, and were placed supine. The optical neuronav-igational system was installed in the usual fashion. Registra-tion was performed by referring to markers attached to thefour fiducial screws as well as the bridge of the nose. Inci-sion locations for the planned electrode trajectories weremarked, and a minimal shave was performed. A larger areaon the vertex of the head was shaved so that all electrodeleads could be brought out in this area after subgaleal tun-neling (Fig. 2). A 10-minute preparation of the whole headand hair was then performed before sterile drapes wereplaced. After draping, the sterile optical reference arc wasattached, as were sterile marker attachments to the implant-ed fiducial screws. In this manner, it was possible to reverifyregistration accuracy and reestablish registration if neces-sary.

After scalp and muscle incisions had been made, pedi-atric bur holes were drilled for the planned electrode trajec-tories. The dura mater was opened, and if any small corticalveins were noted, the entry point of the electrode trajectorywas modified slightly to avoid them. Because of these slightintraoperative modifications, I only considered the deep dis-tal targets of the planned electrode trajectories for my accu-racy assessments.

A rigid guide cannula (Ad-Tech, Racine, WI) was usedfor depth electrode placement. This guide was referencedand tracked using the image guidance system with SureTrakaccessories (Medtronic, Inc.). During placement, trajectoryand target views were displayed to maintain the cannula on

the planned trajectory. Because there was no commercial-ly available system to immobilize the cannula at the timeof this study, a variety of adaptors were tried, ranging froma Yasargil–Leyla retractor arm (Aesculap, San Francisco,CA) to prototypes of the Vertek system (Medtronic, Inc.),which was subsequently released commercially. Becausethe cannula was effectively immobilized in its final positionbefore introduction of the electrode, the mode of immobi-lization did not have any bearing on the stereotactic accura-cy. As previously noted, some trajectory entry points weremodified to avoid a cortical vein. Targets always remainedunchanged. When each electrode had reached its target, itwas immobilized manually with smooth forceps at the levelof the bur hole, the rigid cannula was withdrawn, and ashunt passer was used to tunnel to the shaved area on thevertex, where the electrode tails subsequently were broughtout. This had the advantage of eliminating CSF leaks, re-ducing risks for infection, and also facilitating bedside re-moval of the electrodes. The disadvantage of this strategywith regard to this study was that there was some outwardmovement of the electrodes during the tunneling becausethey were not rigidly immobilized by a skull bolt (Fig. 2).The depth of the original placement was noted with regardto the electrode contacts, and electrodes were advancedmanually to the original depth or slightly less after the leadshad been tunneled. For safety reasons, the tendency was tointentionally leave the electrodes at a site a few millimeterssuperficial rather than placing them too deep. As discussedlater, this variation in electrode placement had repercus-

J. Neurosurg. / Volume 105 / October, 2006

Optimized frameless application accuracy

563

FIG. 2. A: Photograph showing frameless electrode implantation performed in 1994 that involved rigid skull fixation(arrow). The electrode leads are seen secured to the scalp. B: Photograph obtained during the current study, in which theelectrodes were tunneled to the vertex (black arrow) by using a shunt passer. The entry point for the electrodes is at thelevel of the incisions above the ear (white arrows). Although this technique was more comfortable for patients and allowedelectrode removal at the bedside, the lack of skull fixation resulted in some variability in electrode depth. This variabilityhas since been corrected using temporary rigid fixation until wound closure.

sions for this study. On the scalp, the electrode tails wereimmobilized using a custom silicone clamp or sheet at-tached to one of the skull fiducials on the vertex.

Postimplantation imaging was performed either on thesame day as implantation or on the next morning by usingthe same MR imaging protocol that was used preoperative-ly. Image fusion of the preoperative data sets and the post-implantation volumetric MR images was performed on theStealthStation using ImMerge image fusion software (Fig.3). Because image fusion of the preoperative planning im-ages and postoperative electrode localization images wasused as a tool to measure image guidance accuracy, themethod of fusion and the assessment of accuracy of fusionwill be discussed in detail.

Preoperative CT and MR imaging data sets were fusedmanually by using the StealthStation and ImMerge packagewith bone-implanted fiducial markers visualized on bothCT and MR imaging data sets. At the time of this study, theautomated fusion component of the software was testedusing manual fusion with skull-implanted fiducials as astandard. The accuracy of automatic fusion was found to bepractically identical to manual fusion. Fusion of postopera-tive MR images was performed using the automatic fusionprogram to correlate the preoperative CT data set with the

postoperative MR imaging data set. Because implantedskull fiducials were removed at surgery, they were not avail-able as reference points for the fusion process on postoper-ative data sets. A computer-estimated fusion accuracy of 1.5mm or less was considered acceptable. As with any com-puter-based estimation of accuracy, true accuracy may notbe reflected by the estimated value. To control for this, fiveto 10 anatomical landmarks (for example, the pituitary stalkand vascular bifurcations) were assessed for fusion accura-cy before any measures were made, and any measurabledifference in landmarks prompted researchers to perform anew fusion of the data sets to obtain a better correlation.These landmarks were visualized and compared on the pre-operative and postoperative MR imaging data sets, afterboth these data sets had been fused to the preoperative CTdata set (triple fusion).

After verifying the accuracy of the image fusion, the sec-ondary fused postoperative data set showing the electrodeswas displayed so that the trajectories and targets defined inthe primary data sets were superimposed on the imagedelectrodes. Subsequently, the theoretical target was definedas the entry point of a new trajectory and the visualizedelectrode tip was defined as the target in this new trajecto-ry. In this manner, the newly defined trajectory represented

C. R. Mascott

564 J. Neurosurg. / Volume 105 / October, 2006

FIG. 3. Fused data sets of preoperative CT and MR images and postoperative MR images with electrodes in place. The designated targets(the amygdala and hippocampus) were defined preoperatively within a small (2-mm) sphere (Arrows A). The tips of the imaged electrodeswere then compared with the designated targets. For the two amygdalar electrodes illustrated, the distal electrode contact on the right corre-sponds to the planned target (Arrow B). The left amygdalar target (Arrows A’) is illustrated on axial, coronal, sagittal, and three-dimensional(3D) views. The actual electrode position (Arrows A’) is demonstrated on the same views. The distance from A to A’ represents the projectederror of localization in three planes, and the absolute error in three dimensions.

the absolute distance in euclidean space (vector) betweenthe planned target and the electrode tip (Fig. 4).

Because most electrodes were pulled out slightly alongtheir trajectories when compared with their preplanned pos-itions, a second measurement was performed to allow forcorrection of this. The correction in electrode depth was de-termined to better assess the true accuracy of placement be-cause the “pull-out” effect was not dependent on stereotac-tic accuracy but on mechanical factors, and was in partintentional as previously discussed. For each target, thelength of the planned trajectory was measured between theinner table of the skull and the target. This distance was thentransposed to a new trajectory along the path of the visual-ized electrode on the postoperative MR images so as to ex-tend the path of the visualized electrode to a target point atthe planned depth, as measured from the inner table. Thedistance between this depth-corrected point (Fig. 5, Point C)and the planned target was then measured in absolute (eu-clidean) distance to obtain the corrected localization error(Distance A–C).

Results

All electrodes were in an appropriate position to provideclinically useful EEG recordings. The absolute euclideantarget localization error for 42 targets as measured betweenthe preoperatively defined targets and the visualized elec-trode tips was 3 6 1.5 mm (95% confidence interval was2.4–3.6 mm). The maximal error was 6 mm. After correc-tion for depth, the localization error was 2.4 6 1 mm (95%confidence interval was 2.1–2.7 mm). The maximal errorwas 3.4 mm.

Discussion

Quantification of the accuracy of image guidance sys-tems poses some challenges in that it is difficult to find mil-limetric targets against which to measure accuracy. For thisreason, the majority of studies in which the accuracy of im-age guidance systems are quantified are laboratory studiesperformed on phantoms in which targets are easy to defineand measure. For similar reasons, there has been a paucityof studies in which the accuracy of stereotactic frames hasbeen quantified, except for studies performed on phan-toms.3,8,12

The four best studies in which attempts have been madeto quantify in vivo application accuracy relied on the use ofwell-defined, reliable targets. In one, a biopsy site visual-ized on postoperative volumetric imaging was fused withthe corresponding site on the preoperative plan and the lo-calization error was measured on the fused data set.7 Regis-tration of images was performed using adhesive scalp fidu-cial markers in that study, and the resultant accuracy wouldbe expected to be less than that obtained when registrationis optimized using implanted skull fiducials. The euclideanlocalization error found in that study was 4.8 6 2 mm in 15patients. In the second study, implanted skull fiducialscrews were used in addition to a stereotactic frame.27 Ad-ditional skull-implanted fiducial screws served as targets.A target localization error of 1.72 6 0.42 mm was reportedin 19 patients. In a recent article, Mehta, et al.,16 describedtheir strategy for frameless placement of depth electrodes.

Their study was similar to this one in that they used aStealthStation with similar instrumentation for the place-ment of 41 electrodes. The researchers measured localiza-tion accuracy on a fused postoperative MR imaging data setand reported a measured error of 3.1 6 0.5 mm (range 0–7mm). Their study differs from mine in a few significantareas as follows: 1) The majority (31 of 41) of their elec-trodes were placed through craniotomies at the same time asgrid placement. 2) Their image guidance method was basedon MR images (1.5 mm) alone as opposed to CT scans (1mm) fused with MR images (1.5 mm). 3) Image guidanceregistration in their study was based on the placement ofadhesive fiducial markers, whereas in this study, skull-im-planted fiducials were used. As such, the study by Mehtaand colleagues provides further evidence for the feasibilityof frameless placement of depth electrodes with or withouta craniotomy but is less of a study of image guidance appli-

J. Neurosurg. / Volume 105 / October, 2006

Optimized frameless application accuracy

565

FIG. 4. Navigational views. The absolute (euclidean) target errorwas measured in the trajectory view as the length of the trajectorybetween the preoperatively designated target (Arrows A) and thevisualized electrode tip (Arrows A’) on the postoperative MR im-age. The error after correction for depth was the measure of a per-pendicular between the designated target and an extension along thevisualized electrode trajectory. As discussed in the text, electrodesoften pulled out slightly and were always too superficial and nevertoo deep.

cation accuracy than this report. There is one report of free-hand, orthogonal placement of depth electrodes performedwithout image guidance during craniotomy for epilepsy sur-gery that achieved reasonable accuracy.6 Interestingly, therate of error was worse when occipitoparietal approacheswere used without craniotomy (4.4 6 3.2 mm) than whenorthogonal approaches were used with craniotomy (2.8 61.3 mm). The researchers attributed this to greater posteriortrajectory distances. I would argue that another possibility isthat image guidance involving adhesive fiducial-based reg-istration has higher rates of inaccuracies, especially whenposterior approaches are used.13 The means of registrationof an image guidance system at surgery has a huge influ-ence on subsequent accuracy. A crucial component of thepresent study is that all patients had skull-implanted fiducialmarkers, which is the best means to reduce the variability ofaccuracy introduced at the time of surgical registration.11,14,27

Any study in which implanted fiducials are not used for im-age guidance has the potential for considerable variability inregistration introduced at the time of surgery.

In another recent study, Holloway, et al.,10 measured theaccuracy of long-term placement of depth electrodes forfunctional neurosurgery, in which the surgeon referred topostoperative images fused to images of preoperative

planned targets. As in the current study, implanted fiducialmarkers were used for image guidance registration in all pa-tients. Holloway and colleagues used fine-cut CT for plan-ning (with fused MR imaging) and postoperative CT forelectrode position assessment, whereas postoperative MRimaging was used in the current study. In their study, Holl-oway and colleagues compared the accuracy of framelesselectrode placement with that of frame placement used in aprior study22 and found a euclidean (vector) error of 3.2 61.4 mm in both frameless (with bone fiducials) and frame-based studies. In the current study I found a measured local-ization error of 3 6 1.5 mm, but many electrodes wereknown to be withdrawn a few millimeters. There was nocomparable effect in the study by Holloway and colleaguesbecause the electrodes were rigidly secured to the skull.

Other investigators have produced figures for intraopera-tive accuracy by using targets that were too large to be reli-able.17 Several also have used additional scalp adhesive fi-ducials (not included for registration) as targets.20,25 Theseare mobile, unreliable targets for localization quantification.In a different study, I tried to define parameters for semi-quantitative assessment of intraoperative accuracy.13 Thiswas predicated on a more systematic approach than thatoften used by most surgeons to assess accuracy during sur-gery—sequentially localizing discrete targets (for example,vascular bifurcations), linear structures (such as veins or ar-teries and the tentorial edge), and surfaces (such as the skullbase and tumor surfaces). That approach is semiquantitativeat best.

The advantage of assessing electrode placement by usingvolumetric postoperative imaging is that the electrode tip isa clearly definable point that can be compared with theplanned target. Because placement does not involve resec-tion, there is minimal brain shift, and even with a slight CSFloss the electrode placement in space should not vary withCSF shift. There is a small degree of variability in definingthe electrode tip due to the MR imaging artifact. Therewould be less variability in depth if the electrodes were tobe secured to the skull with a screw or bolt (Fig. 2A), a tech-nique that has been traditional and routine in many centers.Because the tunneled electrode protocol used in the patientsin the current study had distinct advantages, it did not seemreasonable to use an older fixation technique just for thesake of this study.

Another source of potential inaccuracy in the study is thepossible error introduced by the image fusion protocol. Forthis study, image fusion of preoperative target plans withpostoperative images showing the electrode position is anintegral part of the methodology for quantifying accuracy ofelectrode placement. As such, the accuracy of the image fu-sion influences the reliability of error measurement of elec-trode placement. This study is predicated on the absence ofbrain shift, and because no craniotomies were performed,minimal CSF was lost if any air was introduced during elec-trode placement. To assess the accuracy of automated andmanual image fusion, anatomical landmarks were selected.These landmarks should not shift and were therefore used toassess and correct the accuracy of image fusion before mea-surement of electrode placement error.

Conclusions

It remains of concern to anyone using image guidance

C. R. Mascott

566 J. Neurosurg. / Volume 105 / October, 2006

FIG. 5. Axial projection from Fig. 3 demonstrating the depth-cor-rected measures. Using the image guidance workstation, the mea-sured planned depth from the inner table of the skull to the targetwas transposed to the visualized electrode trajectory (arrowhead).The end of this new trajectory defines a depth-corrected point thatcompensates for the pullout of electrodes in a manner that is notdependent on stereotactic factors. The distance between the plannedtarget (Arrow A) and the depth-corrected tip of the electrode (ArrowC) when using a 3D measurement represents the true localizationerror to the target (3D Distance A–C). The actual electrode positionis also demonstrated (Arrow A’).

frequently that brain shift can lead to errors during surgery.This can be addressed using intraoperative imaging and up-dates to image guidance. The overall accuracy of imageguidance prior to any shift is also important and is the objectof this study. This is predicated on the contention that imageguidance was optimized in these patients. Most important-ly, the use of skull-implanted fiducials results in the mostconsistently accurate frameless registrations reported.11,27

The use of fine-cut CT combined with MR imaging furtherdecreases variables that can lead to errors. In this study, aframeless system was used to place depth electrodes for theinvestigation of epilepsy. This has traditionally been per-formed using a stereotactic frame. Several institutions havemoved toward performing these procedures without aframe. The verification of electrode placement by using MRimaging for clinical reasons has allowed us to quantify theelectrode placement accuracy for this study. The estimatedoverall accuracy of 2.4 6 1 mm compares favorably to thereported accuracies of frame placement.

Disclaimer

The author has no commercial or other interest in Medtronic, Inc.,Stryker Leibinger Micro Implants, or Ad-Tech.

References

1. Alexander E III, Kooy HM, van Herk M, Schwartz M, Barnes PD,Tarbell N, et al: Magnetic resonance image-directed stereotacticneurosurgery: use of image fusion with computerized tomographyto enhance spatial accuracy. J Neurosurg 83:271–276, 1995

2. Benabid AL: [History of stereotaxis.] Rev Neurol (Paris) 155:869–877, 1999 (Fr)

3. Bourgeois G, Magnin M, Morel A, Sartoretti S, Huisman T, Tun-cdogan E, et al: Accuracy of MRI-guided stereotactic thalamicfunctional neurosurgery. Neuroradiology 41:636–645, 1999

4. Burchiel KJ, Nguyen TT, Coombs BD, Szumoski J: MRI distor-tion and stereotactic neurosurgery using the Cosman-Roberts-Wells and Leksell frames. Stereotact Funct Neurosurg 66:123–136, 1996

5. Cohen DS, Lustgarten JH, Miller E, Khandji AG, Goodman RR:Effects of coregistration of MR to CT images on MR stereotacticaccuracy. J Neurosurg 82:772–779, 1995

6. Davies KG, Phillips BL, Hermann BP: MRI confirmation of accu-racy of freehand placement of mesial temporal lobe depth elec-trodes in the investigation of intractable epilepsy. Br J Neurosurg10:175–178, 1996

7. Dorward NL, Alberti O, Palmer JD, Kitchen ND, Thomas DG:Accuracy of true frameless stereotaxy: in vivo measurement andlaboratory phantom studies. Technical note. J Neurosurg 90:160–168, 1999

8. Galloway RL Jr, Maciunas RJ, Latimer JW: The accuracies of fourstereotactic frame systems: an independent assessment. BiomedInstrum Technol 25:457–460, 1991

9. Guenot M, Isnard J, Ryvlin P, Fischer C, Ostrowsky K, MauguiereF, et al: Neurophysiological monitoring for epilepsy surgery: theTalairach SEEG method. StereoElectroEncephaloGraphy. Indica-tions, results, complications and therapeutic applications in aseries of 100 consecutive cases. Stereotact Funct Neurosurg 77:29–32, 2001

10. Holloway KL, Gaede SE, Starr PA, Rosenow JM, RamakrishnanV, Henderson JM: Frameless stereotaxy using bone fiducialmarkers for deep brain stimulation. J Neurosurg 103:404–413,2005

11. Maciunas RJ, Fitzpatrick JM, Galloway RL, Allen GS: Beyondstereotaxy: extreme levels of application accuracy are provided by

implantable fiducial markers for interactive image-guided neuro-surgery, in Maciunas RJ (ed): Interactive Image-Guided Neuro-surgery. Park Ridge, IL: AANS, 1993, pp 259–270

12. Maciunas RJ, Galloway RL Jr, Latimer JW: The application accu-racy of stereotactic frames. Neurosurgery 35:682–695, 1994

13. Mascott CR: Comparison of magnetic tracking and optical track-ing by simultaneous use of two independent frameless stereotac-tic systems. Neurosurgery 57 (Suppl 4):295–301, 2005

14. Maurer CR Jr, Fitzpatrick JM, Wang MY, Galloway RL Jr, Ma-ciunas RJ, Allen GS: Registration of head volume images usingimplantable fiducial markers. IEEE Trans Med Imaging 16:447–462, 1997

15. Maurer CRJ, Rohlfing T, Dean D, West JB, Rueckert D, Mori K,et al: Sources of error in image registration for cranial image-guid-ed neurosurgery, in Germano IM (ed): Advanced Techniques inImage-Guided Brain and Spine Surgery. New York: Thieme,2002, pp 10–36

16. Mehta AD, Labar D, Dean A, Harden C, Hosain S, Pak J, et al:Frameless stereotactic placement of depth electrodes in epilepsysurgery. J Neurosurg 102:1040–1045, 2005

17. Muacevic A, Uhl E, Steiger HJ, Reulen HJ: Accuracy and clinicalapplicability of a passive marker based frameless neuronavigationsystem. J Clin Neurosci 7:414–418, 2000

18. Murphy MA, O’Brien TJ, Cook MJ: Insertion of depth electrodeswith or without subdural grids using frameless stereotactic guid-ance systems—-technique and outcome. Br J Neurosurg 16:119–125, 2002

19. Olivier A, Bertrand G, Peters T: Stereotactic systems and proce-dures for depth electrode placement: technical aspects. Appl Neu-rophysiol 46:37–40, 1983

20. Sipos EP, Tebo SA, Zinreich SJ, Long DM, Brem H: In vivo accu-racy testing and clinical experience with the ISG Viewing Wand.Neurosurgery 39:194–204, 1996

21. So N, Gloor P, Quesney LF, Jones-Gotman M, Olivier A,Andermann F: Depth electrode investigations in patients with bi-temporal epileptiform abnormalities. Ann Neurol 25:423–431,1989

22. Starr PA, Christine CW, Theodosopoulos PV, Lindsey N, Byrd D,Mosley A, et al: Implantation of deep brain stimulators into thesubthalamic nucleus: technical approach and magnetic resonanceimaging-verified lead locations. J Neurosurg 97:370–387, 2002

23. Steinmeier R, Rachinger J, Kaus M, Ganslandt O, Huk W,Fahlbusch R: Factors influencing the application accuracy of neu-ronavigation systems. Stereotact Funct Neurosurg 75:188–202,2000

24. Talairach J, Szikla G: Application of stereotactic concepts to thesurgery of epilepsy. Acta Neurochir Suppl 30:35–54, 1980

25. Villalobos H, Germano IM: Clinical evaluation of multimodalityregistration in frameless stereotaxy. Comput Aided Surg 4:45–49, 1999

26. Villemure JG, Marchand E, Peters T, Leroux G, Olivier A:Magnetic resonance imaging stereotaxy: recognition and utiliza-tion of the commissures. Appl Neurophysiol 50:57–62, 1987

27. Vinas FC, Zamorano L, Buciuc R, Li QH, Shamsa F, Jiang Z, etal: Application accuracy study of a semipermanent fiducial sys-tem for frameless stereotaxis. Comput Aided Surg 2:257–263,1997

28. West JB, Fitzpatrick JM, Toms SA, Maurer CR Jr, Maciunas RJ:Fiducial point placement and the accuracy of point-based, rigidbody registration. Neurosurgery 48:810–817, 2001

Manuscript received August 10, 2005.Accepted in final form March 2, 2006.Address reprint requests to: Christopher R. Mascott, M.D.,

F.R.C.S.(C), The Epilepsy Institute, Beacon Neurosciences, BeaconHospital, Dublin 18, Ireland. email: mascottdonnelly@netscape.net.

J. Neurosurg. / Volume 105 / October, 2006

Optimized frameless application accuracy

567