te
con
nb,
E. Hecoxa, Vernon L. Towlea,b,c
, 584
o, 584
5841
11 Ma
accurate determination of the location of implanted subdural
any direction to achieve an optimal view of the electrode
Clinical Neurophysiology 1E-mail address: jdhunter@nitace.bsd.uchicago.edu (J.D. Hunter).Abstract
Objective: To determine the accuracy of locating subdural electrodes by means of 3-D surface rendering of CT scans.
Methods: Open source software has been developed and posted on the web which segments the electrodes into 3-D surfaces and allows their
3-D locations to be exported to other EEG analysis programs. The accuracy of the technique was determined by studying 410 subdural
electrodes implanted in four epilepsy patients. Accuracy was determined by comparing the locations from the rendering analysis to the
locations of the same electrodes determined by conventional analysis of their appearance on individual CT slices.
Results: The average accuracy of a study of 410 electrodes imaged in four patients repeated two times by three observers was 0.91
(G0.41) mm, with a maximum error of 3.3 mm, about half of the diameter of an electrode.
Conclusions: The location of subdural electrodes can easily and quickly be determined within high-resolution CT scans through the use of
3-D rendering.
Significance: This relatively fast and easy method for determining the location of subdural electrodes should facilitate their use in both
clinical and research investigations.
q 2005 Published by Elsevier Ireland Ltd. on behalf of International Federation of Clinical Neurophysiology.
Keywords: Subdural electrodes; 3-D localization; Source analysis; Epilepsy surgery; Surface rendering
1. Introduction
The widespread inclusion of chronic depth and subdural
recordings in the surgical work-up of intractable epilepsy
patients has contributed to the efficacy and reduced the
morbidity of modern epilepsy surgery (Bancaud et al., 1965;
Berger et al., 1989; Burchiel et al., 1989; Goldring and
Gregorie, 1984; Wyler et al., 1984; Wyllie et al., 1988).
New approaches to the analysis of intracerebral recordings
require knowledge of the 3-D locations of recording
electrodes (Cuffin et al., 1991; Merlet and Gotman, 1999;
Sutherling et al., 2001; Towle et al., 1999, 2003). To date,
clinical use of advanced quantitative EEG techniques for
patient care. We present here a procedure which segments
the electrodes as 3-D surfaces within CT images, allowing
the location and identification of hundreds of electrode
locations within a few minutes.
2. Methods
High-resolution CT scans (1 mm slice thickness) were
loaded into the program and viewed as slices or 3-D surfaces
(Fig. 1). Slices were displayed orthogonally or rotated inLocating chronically implan
surface re
John D. Huntera,*, Diana M. Hana
Katherine A. Brubakerb, Kurt
aDepartment of Pediatrics, The University of Chicago
bDepartment of Neurology, The University of Chicag
cDepartment of Surgery, The University of Chicago,
Accepted1 South Maryland Avenue, Chicago, IL 60637, USA
1 South Maryland Avenue, Chicago, IL 60637, USA
South Maryland Avenue, Chicago, IL 60637, USA
rch 2005d subdural electrodes using
struction
Bryan F. Singerb, Samir Shaikhb,
16 (2005) 1984–1987
www.elsevier.com/locate/clinphwere utilized. Images were rotated, translated, zoomed, and
brightness and contrast adjusted, under interactive mouse
control.
1388-2457/$30.00 q 2005 Published by Elsevier Ireland Ltd. on behalf of International Federation of Clinical Neurophysiology.
doi:10.1016/j.clinph.2005.03.027
* Corresponding author.electrodes is tedious and difficult, hindering the routine
arrays to facilitate identification of each electrode. Arbitrary
3-D viewports which contained three non-orthogonal slices

Fig. 1. A screenshot of the program revealing 3-D rendered arrays of subdural electrodes (top) and CT slices (below). The upper left panel is a 3D interactive
view which supports interaction, electrode labeling, etc. The upper right is a 3D rendering window showing the segmented electrodes and other metallic
ws sho
ol.
J.D. Hunter et al. / Clinical Neurophysiology 116 (2005) 1984–1987 1985Surfaces for the skin, skull and electrodes were super-
imposed with various colors and transparencies (Fig. 2),
using features to align the different viewports for these 3-D
surfaces with the CT image slices. To render the electrodes,
skin or skull as 3-D surfaces, an intensity value for the
hardware and identified electrodes (colored spheres). The three lower windo
arbitrarily and reposition the electrodes in them with 3-button mouse contrisosurface (e.g. electrodes) to be reconstructed was selected,
using an image sampling tool that collects averages over CT
Fig. 2. A detail of the rendering window illustrating how various surfaces can b
opaque displays. The electrodes and their tunneled leads are juxtaposed relativeregions selected with the mouse. To select these regions, the
cursor was dragged over one or more of the electrodes, and
the average intensity of the pixels sampled was computed.
This could be adjusted up or down via the keyboard, as
needed. This value was then used to render the surface of all
w 3 planar slices, with electrodes. One can pan and zoom all of these viewsof the electrodes and any other objects containing that
intensity value (See Fig. 1). The rendering engine uses
e individually rendered, colored, and superimposed using translucent and
to the craniotomy.

algorithm builds one or more isosurfaces along the specified
contour value. These isosurfaces are fed to a connectivity
would correspond to less accurate results. The electrodes
were 6 mm in diameter with 10 mm center-to-center
wands, skull films, and CT, MRI and ultrasound during
surgery (Barnett et al., 1993; Bootsveld et al., 1993;
Grzeszczuk et al., 1992; Winkler et al., 2000; Towle
et al., 2003). The 3-D location of scalp EEG electrodes has
also been determined using radiologic localizers, infrared
camera arrays, and CT (Homan et al., 1987; Myslobodsky
and Bar-Ziv, 1989; Towle et al., 2003). Winkler’s (2000)
analysis revealed that 39% of electrodes were difficult to
identify on CT slices, largely because their relevant context
rophyfilter, which extracts cells that share common points. The
outputs of this filter are regions, which include the
electrodes as well as other pieces of surgical metal that
are readily distinguishable by visual inspection, both
because they have different geometries and because the
electrodes lie on regular grids. The center of the undersur-
face of the electrodes (not the center of mass) which were in
contact with the brain were marked by clicking on them, and
this screen location was fed to a cell picker (vtkCellPicker)
that returned world coordinates of the point under the mouse
click position with a tolerance of 0.5% of the window width.
If multiple isosurfaces were in the viewport (e.g. skin, bone,
and electrodes), it was possible select which of the surfaces
the picker intersected with. This enabled the placement of
markers on the outer or inner surfaces of fiducial points on
multiple surfaces for inter-image registration.
When the electrodes were marked, spherical polygons
with configurable radii and colors were added to the images,
centered on the selected x, y, z, locations. These polygons
were then labeled in a semi-automatic process to indicate
the grid and electrode number. The user specified the label
and the number of the initial electrode, and the program
incremented the label for each successive electrode. This
marker information (label, position, radius and color) was
exported as an ASCII text CSV file which can be read by
most data base and word processing packages.
The application, which runs on Microsoft Windows,
Linux and OS X, was developed solely with open source
tools, and is available free at http://pbrain.sourceforge.net
for non-commercial use. The Python front-end is a cross-
platform, object oriented, interpreted language widely used
in scientific computing, and enables hardware accelerated
3-D interaction using OpenGL.
3. Results
The within- and between-observer reliability and accu-
racy of electrode localizations has been validated through
two years of use, including repeated analysis by three
different observers, as well as through comparison with
intraoperative photographs taken both before and after the
CT scans. To quantify the accuracy of the technique, an
analysis of the variability and localization error of 410VTK, the Visualization Toolkit for 3-D visualization and
radiographic image manipulation, which was developed by
Kitware (Kitware, Inc., Clifton Park, NY), a private
company funded by National Library of Medicine to
develop high quality visualization libraries. The surface
rendering pipeline consisted of a marching cubes algorithm
followed by connectivity filters; the algorithms and filters
utilized were vtkMarchingCubes and vtkPolyDataConnec-
tivityFilter (Schroeder et al., 2002). The marching cubes
J.D. Hunter et al. / Clinical Neu1986subdural electrodes implanted in four different patients wasspacing, and were arranged in 8!8 arrays or 1!8 strips
placed over the parietal, frontal and temporal lobes,
according to the needs of the patients. The findings reveal
that when high-resolution CT scans of 1–1.5 mm slice
thickness are obtained, the average error of the technique is
less than 1 mm. Although the error increases when lower
resolution CT scans (5 mm slices) are utilized, it still
appears to be less than the diameter of an electrode. In either
case, the empirical error is less than the other sources of
error in dipole localization studies (Towle et al., 1997), and
should make this technique useful for testing and improving
the utility and validity of multichannel intracerebral
recordings.
4. Discussion
We have demonstrated that subdural electrodes can be
accurately located with an 80–90% reduction in time
compared to manual identification of the electrodes on CT
slices. Knowledge of the location of recording electrodes is
increasingly important for investigations in which electro-
physiologic sources are modeled. Several different strat-
egies have been employed to locate implanted electrodes,
including intraoperative photographs, frameless stereotacticperformed (Table 1). The patients gave their informed
written consent. Each electrode was identified by three
independent observers on two separate days. In this
situation, by accuracy we mean the degree to which
electrodes located by the rendering technique approximate
the location of electrodes determined manually by careful
visual inspection of the electrode as seen on individual
slices. We calculated the Cartesian distance between the
electrodes as determined by both of these techniques. A
large Euclidian discrepancy between the two techniques
Table 1
Variability and accuracy of 410 electrode localizations from four patients,
based on observations made by three independent observers
Source Mean (SD) (mm) Maximum (mm)
Within observer variability 0.25 (G0.14) 1.2
Between observer variability 0.31 (G0.13) 0.9
Overall Accuracy 0.91 (G 0.41) 3.3
siology 116 (2005) 1984–1987is not easily viewed. We have found that this problem is

immediately solved using the non-orthogonal slice feature
and the rendered image. The only ambiguous situation we
encountered was when two electrodes were located directly
on top of each other. Viewing the intraoperative photo-
graphs usually solved this problem. In our experience, the
possible misplacement of an electrode by one-half of its
diameter would not likely influence the interpretation of
EEG data. We are not sure that the discrepancy between the
two techniques is not due to deviations from reality by our
‘gold standard’ as evidenced by Winkler’s findings, above,
and that the rendering technique is more accurate than
conventional analysis. One limitation of our technique is
Research Foundation. We are indebted to Jake Reimer and
Anna Hill for aiding in the development and validation of
l’epilepsie. Informations neurophysiopathologiques apporte´es par
l’investigation fonctionelle ste´re´otaxique. Paris: Masson; 1965.
Barnett GH, Kormos DW, Steiner CP, Morris H. Registration of EEG
determination using three-dimensional reconstruction of MR data of
the brain as a component of the presurgical diagnosis of epilepsy.
Radiologe 1993;33:185–8.
Burchiel KJ, Clarke H, Ojemann GA, Dacey RG, Winn HR. Use of
stimulation mapping and corticography in the excision of arteriovenous
malformations in sensorimotor and language-related neocortex. Neu-
rosurgery 1989;24:322–7.
Cuffin BN, Cohen D, Yunokuchi K, Maniewski R, Purcell C, Cosgrove GR,
et al. Tests of EEG localization accuracy using implanted sources in the
human brain. Ann Neurol 1991;29:132–8.
Goldring S, Gregorie EM. Surgical management of epilepsy using epidural
recordings to localize the seizure focus-Review of 100 cases.
J Neurosurg 1984;60:457–66.
Grzeszczuk R, Tan KK, Levin DN, Pelizzari CA, Hu XP, Chen GTY, et al.
J.D. Hunter et al. / Clinical Neurophysiology 116 (2005) 1984–1987 1987electrodes with three-dimensional neuroimaging using a frameless,
armless stereotactic wand. Stereotact Funct Neurosurg 1993;61:32–8.
Berger MS, Kincaid J, Ojemann GA, Lettich E. Brain mapping techniques
to maximize resection, safety, and seizure control in children with brain
tumors. Neurosurgery 1989;25:786–92.
Bootsveld K, Traber F, Kaiser WA, Layer G, Elger CE, Hufnagel A,
Gieseke J, Reiser M. Intracranial ECoG electrodes. Locationthis research project. Portions of these data were presented
at the annual meeting of the American Epilepsy Society,
December, 2004, New Orleans.
References
Bancaud J, Talairach J, Bonis A. La ste´re´o-E´lectroence´phalographie dansthat it does not currently allow visualization of the
electrodes relative to cortical gyri. Even so, because the
electrode localizations can be obtained so quickly and
easily, we have found it useful to visualize the location of
chronic subdural electrodes as an aid to surgical decision
making.
Acknowledgements
Supported in part by NIH NS40514 and The BrainRetrospective fusion of radiographic and MR data for localization of
subdural electrodes. J Comput Assist Tomogr 1992;16:764–73.
Homan RW, Herman J, Purdy P. Cerebral location of international 10–20
system electrode placement. Electroencephalogr Clin Neurophysiol
1987;66:376–82.
Merlet I, Gotman J. Reliability of dipole models of epileptic spikes. Clin
Neurophysiol 1999;110:1013–28.
Myslobodsky MS, Bar-Ziv J. Locations of occipital EEG electrodes
verified by computed-tomography. Electroencephalogr Clin Neurophy-
siol 1989;72:362–6.
Schroeder W, Martin K, Lorenson B. The visualization toolkit: an object
oriented approach to 3D graphics. 3rd ed.: Pearson Education, Inc.;
2002.
Sutherling WW, Akhtari M, Mamelak AN, Mosher J, Arthur D, Sands S,
Weiss P, Lopez N, DiMauro M, Flynn E, Leahy R. Dipole localization
of human induced focal afterdischarge seizure in simultaneous
magnetoencephalography and electrocorticography. Brain Topogr
2001;14:101–16.
Towle VL, Khorasani L, Berkson EM, Syed I, Pelizzari CH, Jewett D,
Spire J-P. Human somatosensory evoked potential dipoles: how
accurate are the localizations? Proc IEEE-EMBS 1997;19:2616–21.
Towle VL, Carder RK, Khorasanil L, Lindberg D. Electrocorticographic
coherence patterns. J Clin Neurophysiol 1999;16:528–47.
Towle VL, Khorasani L, Uftring S, Pelizzari C, Erickson RK, Spire J-P,
Hoffmann K, Chu D, Scherg M. Noninvasive identification of the
human central sulcus: a comparison of gyral morphology, functional
MRI dipole localization and direct cortical mapping. NeuroImage 2003;
19:684–97.
Winkler PA, Vollmar C, Krishnan KG, Pfluger T, Bru¨ckmann H,
Noachtar S. Usefulness of 3-D reconstructed images of the human
cerebral cortex for localization of subdural electrodes in epilepsy
surgery. Epilepsy Res 2000;41:169–78.
Wyler AR, Ojemann GA, Lettich E, Ward AA. Subdural strip electrodes for
localizing epileptogenic foci. J Neurosurg 1984;60:1195–200.
Wyllie E, Lu¨ders H, Morris III HH, Lesser RP, Dinner DS, Rothner AD,
et al. Subdural electrodes in the evaluation for epilepsy surgery in
children and adults. Neuropediatrics 1988;19:80–6.