Publisher’s version / Version de l'éditeur:
IEEE Transactions on Biomedical Engineering, 59, 12, pp. 3491-3497, 2012-12
READ THESE TERMS AND CONDITIONS CAREFULLY BEFORE USING THIS WEBSITE. https://nrc-publications.canada.ca/eng/copyright
Vous avez des questions? Nous pouvons vous aider. Pour communiquer directement avec un auteur, consultez la première page de la revue dans laquelle son article a été publié afin de trouver ses coordonnées. Si vous n’arrivez pas à les repérer, communiquez avec nous à PublicationsArchive-ArchivesPublications@nrc-cnrc.gc.ca.
Questions? Contact the NRC Publications Archive team at
PublicationsArchive-ArchivesPublications@nrc-cnrc.gc.ca. If you wish to email the authors directly, please see the first page of the publication for their contact information.
This publication could be one of several versions: author’s original, accepted manuscript or the publisher’s version. / La version de cette publication peut être l’une des suivantes : la version prépublication de l’auteur, la version acceptée du manuscrit ou la version de l’éditeur.
For the publisher’s version, please access the DOI link below./ Pour consulter la version de l’éditeur, utilisez le lien DOI ci-dessous.
https://doi.org/10.1109/TBME.2012.2220356
Access and use of this website and the material on it are subject to the Terms and Conditions set forth at
Improved localization accuracy in magnetic source imaging using a 3-D
laser scanner
Bardouille, Timothy; Krishnamurthy, Santosh V.; Ghosh Hajra, Sujoy; D'Arcy,
Ryan C. N.
https://publications-cnrc.canada.ca/fra/droits
L’accès à ce site Web et l’utilisation de son contenu sont assujettis aux conditions présentées dans le site LISEZ CES CONDITIONS ATTENTIVEMENT AVANT D’UTILISER CE SITE WEB.
NRC Publications Record / Notice d'Archives des publications de CNRC:
https://nrc-publications.canada.ca/eng/view/object/?id=a3fe32d3-6ef8-40b3-940e-5e79a2e7be6d https://publications-cnrc.canada.ca/fra/voir/objet/?id=a3fe32d3-6ef8-40b3-940e-5e79a2e7be6d
Abstract — Brain source localization accuracy in
magnetoencephalography (MEG) requires accuracy in both digitizing anatomical landmarks and co-registering to anatomical magnetic resonance images (MRI). We compared the source localization accuracy and MEG-MRI co-registration accuracy of two head digitization systems - a laser scanner and the current standard electromagnetic digitization system (Polhemus) - using a calibrated phantom and human data. When compared using the calibrated phantom, surface and source localization accuracy for data acquired with the laser scanner improved over the Polhemus by 141% and 132%, respectively. Laser scan digitization reduced MEG source localization error by 1.38mm on average. In human participants, a laser scan of the face generated a 1000-fold more points per unit time than the Polhemus head digitization. An automated surface matching algorithm improved the accuracy of MEG-MRI co-registration over the equivalent manual procedure. Simulations showed that the laser scan coverage could be reduced to an area around the eyes only while maintaining co-registration accuracy, suggesting that acquisition time can be substantially reduced. Our results show that the laser scanner can both reduce setup time and improve localization accuracy, in comparison to the Polhemus digitization system.
Index Terms— Magnetoencephalography (MEG), magnetic
resonance imaging (MRI), co-registration, laser scanner, localization accuracy
I. INTRODUCTION
AGNETOENCEPHALOGRAPHY (MEG) is a non-invasive technique for mapping neuronal activity by measuring magnetic fields generated from intracellular electrical currents [1-3]. MEG has high spatiotemporal resolution, which makes it a valuable tool for both basic and clinical brain research and is clinically-indicated for localization in epilepsy and pre-surgical mapping [4-6]. Neuromagnetic sources are localized with respect to the spatial configuration of the helmet-shaped
Manuscript received October 31, 2011. (Write the date on which you submitted your paper for review.) This work was supported in part by the National Research Council of Canada, the Atlantic Innovation Fund, and the T. Bardouille is with the National Research Council, Halifax, NS Canada (Corresponding author: Phone: 902-473-1865; Fax: 902-473-1851; E-mail: tim.bardouille@nrc-cnrc.gc.ca).
S. V. Krishnamurthy is with the IWK Health Centre, Halifax, NS Canada (E-mail: santosh.vema@iwk.nshealth.ca).
S. G. Hajra is with the National Research Council, Halifax, NS Canada (E-mail: sujoy.ghoshhajra@nrc-cnrc.gc.ca).
R. C. N. D’Arcy is with the National Research Council Institute for Biodiagnostics (Atlantic), Halifax, NS Canada (E-mail: Ryan.D'Arcy@nrc-cnrc.gc.ca) .
MEG sensor array through inverse modeling [2]. These locations are translated to sources within the brain by combining MEG data with the structural image of the brain generated using magnetic resonance imaging (MRI). By this approach, MEG can identify the specific brain areas that are involved in function or pathology. The precise co-registration of the MEG and MRI coordinate systems is essential for achieving accurate localization. Errors in coregistration lead to errors in source localization.
MEG/MRI co-registration is achieved by defining transformations between three coordinate frames: the MEG scanner, the participant’s head in the scanner (“MEG head”), and the anatomical MRI of the participant’s head (“MRI head”) [7]. The MEG scanner coordinate frame is based on the manufacture of the sensor array and remains constant. The transform between MEG scanner and MEG head coordinate frames is based on head position indicator (HPI) coils. HPI coils are 3 or 4 small coils that emit known magnetic fields that can be easily localized within the MEG scanner coordinate frame during the scan. Prior to scanning (i.e, outside of the magnetically shielded room), the HPI coils are placed on the subject’s head and the 3-D positions of the coils on the head are digitized. If the HPI coils are not placed over the so-called fiducial landmarks (nasion and peri- or pre-auricular points), then the positions of the landmarks are also digitized. Also, the positions of additional points on the face and scalp are often digitized. The transform between MEG head and MRI head coordinate frames is based on this 3-D digitization and the manual identification of the same fiducial landmarks on the MRI using image-viewing software. In general, increasing the number of digitization points helps constrain this transform.
It is not surprising that the precision of the MEG/MRI co-registration is strongly dependent upon the accuracy of 3-D digitization and MRI fiducial landmark identification. The digitization procedure represents a significant factor affecting the localization accuracy for neuromagnetic source imaging. Localization accuracy in MEG is generally thought to be on the order of millimeters. However, there is evidence that MEG source localization accuracy can be improved. For example, Suk et al. (1991) estimated the limit of localization accuracy by imaging cortical responses to consecutive activation of tactile receptors on three adjacent fingers [8]. They identified different activation locations for each finger over approximately 5 mm of cortex, which estimates the localization accuracy as 1.7 mm. This example, which nicely
Improved Localization Accuracy in Magnetic
Source Imaging using a 3D Laser Scanner
Timothy Bardouille, Santosh V. Krishnamurthy, Sujoy Ghosh Hajra, and Ryan C.N. D’Arcy
controlled for localization error by restricting the comparison to within subjects and within scans, highlights the need to focus on further error reduction strategies. MEG-MRI co-registration represents an important factor in reducing localization error, and therefore requires further attention.
The most commonly used device for 3-D digitization in MEG is the Polhemus Fastrak 3-D digitization unit (Polhemus Incorporated, VT, USA). The Polhemus uses alternating current electromagnetic transmitters and sensors to measure positions in space. As a result, the surrounding electromagnetic environment, and even the Polhemus itself, can adversely affect the digitization result, making it error prone [9]. Furthermore, the low sampling rate of the device makes digitization of the whole head time consuming.
To overcome these challenges, we investigated 3-D laser scanning as an alternative (NextEngine, CA, USA). Laser scanners are insensitive to magnetic objects in the nearby environment and offer improved spatial sampling rate [9, 10]. However, the relative accuracy of laser scanning for surface and, consequently, source localization remains to be demonstrated. Accordingly, the aim of the current study was to compare the performance of laser scanning against the standard Polhemus digitization method for both a calibrated phantom and human data.
We completed localization of known points on the surface of a calibrated phantom. We hypothesized that the 3-D laser scanner would offer superior surface localization accuracy in comparison to the Polhemus digitizer. We also activated a current source of known properties inside of a calibrated phantom. We hypothesized that source localization accuracy would be improved using coordinate frame co-registration based on the 3-D laser scanner in comparison to the Polhemus digitizer. Finally, we performed digitization of the human head using both the Polhemus and the laser scanner and co-registered results to anatomical MRI to compare data quality. Manual and automated co-registration was performed. Using these data combined with the automated co-registration algorithm, we simulated laser scans with reduced spatial sampling rate and coverage to determine optimized scan parameters that provide the necessary precision in MEG-MRI co-registration in a timely manner.
A. Phantom Study – Surface Localization Accuracy
The calibrated phantom supplied by the manufacturer (Elekta Neuromag, Finland) contains 32 current sources and four HPI coils at known locations inside a head-sized hemisphere (see Figure 1). The HPI coils are molded into the calibrated phantom at equidistant locations around the circumference, at a distance of 79.5 mm from the center. We compared the accuracy of localization for the HPI coils on the phantom using the Polhemus and laser digitization methods.
Phantom 3-D digitization was performed by mounting the calibrated phantom on a non-magnetic stand within 50 cm of the field generator. The Polhemus measured the positions of two magnetic field sensors with respect to a generator of a known magnetic field. One sensor was a “reference” sensor, secured to the head to correct for movement during the digitization process. The second sensor was a stylus, which was held on the points to be digitized. The tip of the stylus contains the magnetic field sensor, which is activated by a button press. The reference sensor was secured on the phantom. The stylus sensor digitized the position of the centre of each HPI coil.
The phantom was then installed in a mounting system for laser digitization. The laser scanner used harmless invisible laser light (FDA approved) to triangulate the distance to points on the surface of an object. A line of human-safe laser light was emitted to the object surface and the reflection was measured using a video camera. The camera was spatially offset from the laser source. The information on the position of the object was obtained based on the physical relationship between the emitted and reflected laser light. Color was also extracted from the digital image. The vendor-supplied laser scanner software then generated a list of locations and colors for all points on the surface.
To identify landmarks in the laser scanner image, a yellow sticker was placed on the location of each HPI coil. As the phantom is axially symmetric, a distinct mark was made on each quadrant of the hemisphere to distinguish laser scans. Four side-on laser scan images were taken of the calibrated phantom at equal angles around its circumference. The individual images were concatenated into a 3-D reconstruction using the laser scanner software. The locations of the centers of the yellow stickers were extracted from the reconstruction.
For each method, the measurement was repeated ten times. The difference between the known and measured distances between HPI coils defined the per-point localization accuracy of each device. A repeated measures analysis of variance (ANOVA) [11] revealed any significant differences in the per-point localization accuracy between devices (p < 0.01).
B. Phantom Study – Current Source Localization Accuracy
Once HPI localization was completed ten times with each method, we estimated the variability in localization of the phantom sources for both methods. To do this, the phantom was installed in the MEG. A brief HPI localization scan was performed, followed by activation of a current source with a Figure 1 - The calibrated phantom is shown inside the MEG helmet.
known location inside of the phantom for 2 cycles of a 20 Hz sinusoid at a magnitude of 25 nAm. During this time, MEG data were collected at 1000 Hz with an inline low-pass filter at 330 Hz for 100 repetitions of the source activation.
MEG data for each period of activation were averaged and environmental noise was subtracted using signal-space separation [12]. Data were then low-pass filtered at 40 Hz offline to further improve the signal-to-noise ratio. The magnetic field strength at the peak latency of activation was then visualized topographically.
A recursive least squares fitting algorithm estimated the location of the current dipole source within the phantom for which the projected magnetic field strength matched the measured topography [2]. This source localization procedure was repeated 20 times; using co-registration based on the HPI positions acquired during each of the ten Polhemus and laser scans. For each digitization device, the mean and standard error of the difference between the known and estimated source location was determined. An unpaired t-test revealed any significant difference in source localization accuracy based on the scanner used for co-registration (p < 0.01). Systematic error was evaluated by plotting the estimated locations versus know in all three cardinal planes.
C. Human Study – Data Acquisition
Polhemus and laser digitization scans, and anatomical MRI scans were acquired on eight healthy adult participants enrolled in a MEG neuroimaging study. All participants underwent both Polhemus and laser digitization scans to establish the position of HPI coils and anatomical landmarks and to derive a head shape. Each participant also had a high-resolution (1mm3 voxel size) T1-weighted anatomical MRI scan completed as part of the study. Four subjects were excluded from further analysis, as the MRI images did not include coverage of the face. The study had full research ethics approval and all participants provided informed consent before involvement in the study.
Before the MEG scan, two HPI coils were placed on the participant’s forehead and one in front of each ear (four in
total). The positions of the HPI coils, three fiducial landmarks (nasion and left/right pre-auricular points), and a 100-150 point head shape were digitized using the Polhemus digitizer. For head digitization, the participant was seated such that the head was positioned within 50 cm of the magnetic field generator (attached to the back of a chair). Digitization data were acquired through positioning the stylus at fixed points, with the 3-D location data transferred to the MEG acquisition software in real time. The Polhemus digitization process was completed in about 3 minutes.
After the MEG scan, participants underwent three 60s laser scans: one of the front and one of each side of the head (i.e., total scan time of 3 minutes). For ease of identification, colored stickers were placed on the fiducial landmarks and HPI coils prior to digitization. Also, participants wore a swimming cap to image the head shape above the hairline. The locations of the colored stickers were determined automatically to identify the fiducial landmarks and HPI coils in the 3-D image.
D. Human Study –MEG-MRI Co-Registration
For each subject, a preliminary manual MEG-MRI co-registration was achieved by marking the fiducial landmarks on the MRI in the available viewing software (MRI Lab, Elekta Neuromag). In turn, surface data from each scanner (Polhemus and laser) were then superimposed on the anatomical MRI image. The quality of the co-registration was verified based on three constraints; (1) the triangle defined by the three fiducial landmarks should match for the MRI and the digitization scan, (2) the fiducial landmarks should overlay the correct anatomy, and (3) visual surface matching between the scanner data and the MRI. If necessary, the marked locations of the fiducial landmarks were then manually updated in the software to ensure that all three constraints were satisfied.
For comparison to the manual approach, automated MRI co-registration of the Polhemus and laser scanner data was also performed. The scalp surface was obtained using surface segmentation applied to the anatomical MRI [13] as implemented in the SPM2 analysis package, with the scalp Figure 2 - Head surface digitization data acquired by the (a) Polhemus and (b) laser scanner, and (c) a 3-D reconstruction image from the laser scanner. Laser scanner digitization acquires a thousand-fold more points per unit time and compiles scans of the front and both sides of the head.
intensity threshold specified as 10% of the maximum voxel intensity. For the laser scanner, the extracted scalp surface and laser scanner data were then manually restricted to include the face above the upper lip only, as these are easily identifiable in both MRI and laser images. For the automated co-registration with Polhemus data, we used all surface data available. A coordinate frame transformation was then applied to the scanner data to place both datasets in a similar orientation. Automated co-registration of the MRI and digitization scanner datasets was then achieved using the Iterative Closest Point (ICP) algorithm [14] using a K-dimensional tree based alignment [15] and point-to-plane error minimization [16]. A maximum of 50 iterations of the ICP were allowed for the alignment to converge, to provide the necessary accuracy in a reasonable amount of processing time.
The co-registration was manually verified in the available MRI viewing software, as above. As well, the distance to the nearest co-registered MRI point was calculated for each digitization scanner point to quantify the co-registration accuracy. For each subject, the mean and standard error of co-registration accuracy was calculated between the digitization scanner data and the MRI surface.
E. Human Study – Simulations
We completed three simulations studies to establish an optimal sampling rate and range of coverage that could decrease the laser scan acquisition time without negatively impacting on co-registration accuracy. The first simulation investigated the effect of reducing the spatial sampling rate of the laser scan. Automated co-registration was performed on downsampled subsets of the laser scan data - including as little as one tenth of the entire set. The second and third simulations investigated the effect of reducing the coverage of the laser scan. These two simulations did not require detailed coverage of the fiducial landmarks or HPI coils, as we assume that MEG-MRI co-registration can be achieved without the need for HPI coils in the MEG scan (as addressed in the
spatial sampling, but varied the range of coverage of the laser scanner in the inferior-superior (defined as “z”) direction by defining possible regions of coverage based on centre, zcentre,
and range, zrange. In the third simulation, we used the entire
spatial sampling and optimized coverage in the z-direction defined by the second simulation, but restricted the range of coverage to exclude points below a minimum anterior distance, defined as yminimum.
For each simulation, the defined subsets of laser scanner data were automatically co-registered with the MRI data as described above. The resultant co-registration transformation matrix was then applied to the entire laser dataset, and the distance to the nearest neighbor on the MRI surface was calculated for each point. The mean of nearest-neighbor distances defined the mean co-registration accuracy of the ICP algorithm for each iteration of the simulation. For each subject, we determined the parameters for spatial sampling and coverage that minimized the MEG-MRI co-registration accuracy.
III. RESULTS
A. Phantom Study
We compared the accuracy of position estimation for the calibrated phantom HPI coils between the Polhemus and laser digitizers. The mean difference between measured and known inter-coil separation on the phantom defined the surface localization accuracy of each device. Surface localization accuracy for estimating the spatial configuration of the HPI coils is 0.8 ± 0.2 mm (mean ± standard error) for the laser scanner. This is an improvement of 141% over the Polhemus (1.9 ± 0.3 mm; p < 0.01).
We also compared the accuracy for localization of a current source inside of the calibrated phantom based on HPI coil digitization using the Polhemus and laser scanner The mean difference across repeated measures between measured and known location of the current source defined the source localization accuracy of each device. Source localization accuracy for estimating the position of the known source is 1.04 ± 0.03 mm for the laser scanner. This is an improvement of 132% over the Polhemus (2.42 ± 0.08 mm; p < 0.0001) for estimating the location of the current source. Notably, the laser scanner improved source localization accuracy by 1.38 mm on average. Examination of these results in the cardinal planes revealed that localization error for Polhemus tended to dominate in the z-direction.
B. Human Study
Manual MEG/MRI co-registration was completed in all four subjects based on Polhemus and laser digitization data. All three co-registration constraints were satisfied for all subjects and both digitization scanner types. Automatic co-registration was also achieved for the laser scanner data in all four subjects. Although automatic co-registration of Polhemus data converged in all four subjects, the resultant surfaces did not always match well with the underlying anatomical MRIs. Figure 3 - MEG-MRI co-registration of a single subject based on (a)
Scatter plots of the surface digitization generated by both scanners for a representative subject are shown in Figure 2, along with the 3-D reconstructed image of the face. Figure 3 shows cross-sections of the anatomical MRI for the same subject in three cardinal planes with digitization data highlighted for the manual Polhemus and laser scanner co-registrations. The automated laser scanner co-registration is not shown, as it is qualitatively indistinguishable from the manual laser scanner data. The laser scanner acquires approximately 150,000 points in three 60-second scans. In contrast, the Polhemus generates a sparse map of the head surface, collecting 150 points in approximately the same amount of time. Structures on the face (i.e. nose, forehead, ears, eyes, and cheeks) that offer ideal landmarks for co-registration can be easily identified in the high-resolution laser scan. These landmarks are either not apparent or sparsely sampled in the Polhemus scan.
C. Simulations
Reducing the spatial sampling before applying the ICP
algorithm reduced the co-registration accuracy for the entire dataset, as shown in Figure 4a for the representative subject. Thus, the full sampling was used in the second and third simulations. Figure 4b shows a surface plot of co-registration accuracy for various combinations of zcentre and zrange for the
same representative subject used in Figures 2 and 3. Optimal co-registration accuracy was achieved using a 10 cm inferior-superior range centered at the nasion. Using this optimized coverage in the z-direction, the third simulation showed little change in co-registration accuracy when excluding laser scanner data with y-value less than 8 cm, as shown in Figure 4c. Co-registration accuracy worsened rapidly when the laser scanner coverage was reduced above 8 cm in the y-direction. The results of the three simulations were consistent across the subject group, suggesting that optimal co-registration accuracy is achieved with high spatial sampling, and that coverage can be dramatically reduced without a major loss of accuracy.
Figure 4d indicates the optimized region of coverage as well as the spatial characteristics of co-registration accuracy. Co-registration accuracy is poor only around the tip of the nose, Error [mean ± standard error; mm]
Subject Manual Polhemus Manual Laser ICP (Polhemus) ICP (Full Laser) ICP (Reduced Laser) 1 12 ± 2 4.70 ± 0.02 14 ± 1 2.3 ± 0.02 2.41 ± 0.02 2 4.8 ± 0.2 5.50 ± 0.03 3.8 ± 0.2 2.90 ± 0.03 2.99 ± 0.03 3 3.0 ± 0.2 3.17 ± 0.01 2.2 ± 0.2 1.76 ± 0.01 1.77 ± 0.01 4 7.9 ± 0.9 5.46 ± 0.03 7.6 ± 0.9 1.87 ± 0.01 1.89 ± 0.02
MEAN 7 ± 2 4.7 ± 0.5 7 ± 3 2.2 ± 0.3 2.3 ± 0.3
Figure 4 - Automated MEG-MRI co-registration with laser scanner data. A) The effect on mean co-registration error of spatial sampling. B) The effect of adjusting inferior-superior laser scanner coverage in two dimensions (Z-range and Z-centre). C) The effect of reducing anterior-posterior laser scanner coverage. D) The co-registration error for each point in the optimized region, superimposed on the MRI data (gray). Same subject as in Figures 2 and 3.
Table 1 summarizes the mean co-registration accuracy for all subjects. Accuracy is calculated using the entire digitization scanner dataset, although reduced ICP co-registration is defined based on the optimized coverage area (as shown in Figure 4d). MEG-MRI co-registration based on laser scanner data is more accurate using the ICP algorithm as compared to the manual approach. Both manual and automated co-registration is more accurate using the laser scanner than the Polhemus scanner. There is very little change in co-registration accuracy when the laser scanner coverage is reduced to an area around the eyes only.
IV. DISCUSSION
Digitization is a critical step in MEG-MRI co-registration, which ultimately affects the accuracy of source localization. In the current study, we hypothesized that laser scanning technology would out-perform the existing Polhemus method, which uses an electromagnetic position digitization system.
With respect to accuracy, we used a calibrated phantom to evaluate performance of Polhemus versus laser digitization methods. Laser digitization improved surface and source localization accuracy by 141% and 132%, respectively. Importantly, this result reflected a 1.38 mm average improvement in source localization accuracy.
With respect to application, the laser scanner generated a thousand-fold higher resolution image of the head, in the same amount of time as the Polhemus device. It provided high-resolution imaging of structures on the face that was easily co-registered with the anatomical MRI, both manually and using the ICP algorithm. Co-registration based on the laser scanner data required that high-resolution MR images include coverage of structures on the face.
The phantom results emphasized the importance of co-registration for improving source localization accuracy in MEG. The phantom results and the first simulation demonstrate that higher resolution digitization leads to more effective MEG-MRI co-registration. The improvement likely relates to both the increased spatial sampling and reduced localization errors that occurred with the laser scanner method. While Polhemus outliers can be compensated via their removal, this involves subtraction from the already low spatial sampling rate. In contrast, the laser scanner provided sufficient spatial sampling coverage to permit accurate and reliable identification of anatomical landmarks, such as the nose and eyebrows.
Another advantage of the laser scanner is the ability to properly evaluate the MEG-MRI co-registration (see Figure 3). Given that MEG-MRI co-registration represents a point where error can readily be introduced into source localization, the ability to evaluate the fit between critical coordinate frames represents an important quality control step. Having a 3-D reconstruction of the head, with automatically identified positions of the fiducial landmarks, HPI coils and additional anatomical structures, greatly enables this evaluation.
Co-registration accuracy between the MRI and laser surfaces improved with the use of an automated algorithm, as
has been reported elsewhere [17], was maintained even with a marked reduction in the coverage of the laser scan. Thus, laser scan acquisition time can be substantially reduced leading to a reduction in MEG setup time. This is an important issue with difficult patient populations.
Importantly, the reduced laser scan coverage could be optimized and shown to still produce robust co-registration results. This reduction is noteworthy because it allows for a shorter acquisition time (1 minute instead of 3), which can be critical when scanning patient populations. However, the reduced laser scan coverage does not include the fiducial landmarks or the HPI coils that monitor the head position within the MEG helmet. As stated earlier, these coils are essential to defining the transformation from the MEG scanner to MEG head coordinate frames. This problem may be addressed if the HPI coils can be replaced by a MEG-compatible system for real-time 3-D position tracking of the face (possibly based on laser scanning technology). In this case, the MEG scanner to MRI coordinate frame transformation would be defined in a single step without the necessity for HPI coils, which are prone to shifting during scanning. Our results to date suggest that such a system may provide improved co-registration and localization accuracy over the current combination of HPI coils and Polhemus 3-D digitization.
Further validation of the laser scanner in terms of gains in localization accuracy associated with human MEG scans is required. This is difficult to achieve because a gold standard, in terms of correct location of activation in the brain, is elusive. However, there may be both technical (e.g., magnetic stimulation) and experimental (e.g., within subject localization error) approaches to demonstrate in vivo MEG localization improvements using laser digitization. Further, repeated measures within a human subject, similar to the phantom study described here, may provide some useful confirmation.
V. CONCLUSION
Three-dimensional imaging of the head shape with a laser scanner offers improved spatial resolution per unit time and localization accuracy in MEG. Simulation studies show that laser scanner acquisition time can be significantly reduced without any cost to MEG-MRI co-registration accuracy by scanning the area around the eyes only. In the case of calibrated phantom measurements, the laser scanner provides an improvement in source localization accuracy of 1.38 mm over the Polhemus.
ACKNOWLEDGMENT
This work was supported by the National Research Council of Canada, the Atlantic Innovation Fund, and the IWK Health Centre. We thank Elekta Atlantic and Elekta Neuromag for their helpful contributions to this work.
REFERENCES
[1] S. Baillet, et al., "Electromagnetic brain mapping," 2001. [2] M. Hamalainen, et al., "Magnetoencephalography - theory,
instrumentation, and applications to noninvasive studies of the working human brain," Reviews of Modern Physics, vol. 65, pp. 413-496, 1993.
[3] R. Hari, et al., "The brain in time: insights from neuromagnetic recordings," Annals of the New York Academy of Sciences, vol. 1191, pp. 89-109, 2010.
[4] J. S. Ebersole and S. M. Ebersole, "Combining MEG and EEG source modeling in epilepsy evaluations," J Clin Neurophysiol, vol. 27, pp. 360-71, Dec 2010.
[5] S. M. Stufflebeam, "Clinical magnetoencephalography for neurosurgery," Neurosurg Clin N Am, vol. 22, pp. 153-67, vii-viii, Apr 2011.
[6] J. P. Makela, et al., "Magnetoencephalography in neurosurgery,"
Neurosurgery, vol. 59, pp. 493-510; discussion 510-1, Sep 2006.
[7] P. Adjamian, et al., "Co-registration of magnetoencephalography with magnetic resonance imaging using bite-bar-based fiducials and surface-matching," Clin Neurophysiol, vol. 115, pp. 691-8, Mar 2004.
[8] J. Suk, et al., "Anatomical localization revealed by MEG recordings of the human somatosensory system,"
Electroencephalogr Clin Neurophysiol, vol. 78, pp. 185-96, Mar
1991.
[9] L. Engels, et al., "Factors influencing the spatial precision of electromagnetic tracking systems used for MEG/EEG source imaging," Neurophysiologie Clinique/Clinical Neurophysiology, vol. 40, pp. 19-25, 2010.
[10] L. Koessler, et al., "Spatial localization of EEG electrodes,"
Neurophysiol Clin, vol. 37, pp. 97-102, Apr-May 2007.
[11] S. W. Greenhouse and S. Geisser, "On methods in the analysis of profile data," Psychometrika, vol. 24, pp. 95-111, 1959.
[12] S. Taulu, et al., "Suppression of interference and artifacts by the Signal Space Separation Method," Brain Topogr, vol. 16, pp. 269-75, Summer 2004.
[13] J. Ashburner and K. Friston, "Multimodal image coregistration and partitioning--a unified framework," NeuroImage, vol. 6, pp. 209-17, Oct 1997.
[14] P. Besl and N. McKay, "A method for registration of 3-D shapes," in IEEE Trans. Pattern Analysis and Machine Intelligence, 1992, pp. 239-256.
[15] G. Bulan and C. Ozturk, "Comparison of two distance based alignment method in medical imaging," in Proceedings of the 23rd
Annual EMBS International Conference, Istanbul, Turkey, 2001,
pp. 2606-2608.
[16] C. Yang and G. Mediono, "Object modelling by registration of multiple range images," Image and Vision Computing, vol. 10, pp. 145-155, 1992.
[17] C. Whalen, et al., "Validation of a method for coregistering scalp recording locations with 3D structural MR images," Hum Brain
Mapp, vol. 29, pp. 1288-301, Nov 2008.
Timothy Bardouille received his M.Sc in physics from Dalhousie University
in Halifax, Canada, in 1999, and his Ph.D in medical sciences from the University of Toronto in Toronto, Canada in 2010.
Following his Master’s degree, he worked as a Physicist for a MEG system manufacturer (VSM Medtech Inc.) and a neuroscience research facility (Rotman Research Institute). He is now a Research Officer with the National Research Council of Canada (NRC) in Halifax, Nova Scotia, and has a Scientific Staff appointment at the IWK Health Centre. He has published a number of articles in the field of MEG neuroimaging. His current research focuses on enabling imaging of neuronal networks using MEG for the purposes of clinical diagnosis and treatment monitoring.
Santosh V. Krishnamurthy received his B.Eng from Visvesvaraya
Technological University in Belgaum, India and his M.A.Sc in Electrical and Computer Engineering from Dalhousie University in Halifax, Canada, in 2010.
Following his Master’s degree, he works as a Medical Devices Engineer at the IWK Health Centre. He is also a guest worker with the NRC-IBD Atlantic division in Halifax, Nova Scotia.
Sujoy Ghosh Hajra (M’10) received his B.A.Sc. in electrical and computer
engineering from the University of Toronto in Toronto, Canada, in 2008. He is currently pursuing advanced course work within the Faculty of Graduate Studies at Dalhousie University.
He is a Research Engineer with NRC in Halifax, Nova Scotia where he works closely with clinicians and neuroscientists to develop innovative brain imaging and diagnostic technologies. He has played a key role in the development and clinical translation of several medical technologies recently developed by the institute.
Mr. Ghosh Hajra is also a member of Professional Engineers Ontario (PEO) as an EIT.
Ryan C.N. D’Arcy received his M.Sc. and Ph.D. in neuroscience from
Dalhousie University in Halifax, Canada, in 2002.
Following his training as a Post-Doctoral Research Associate with the NRC’s Institute for Biodiagnostics (IBD) in Winnipeg, Manitoba, he established and continues to lead the Atlantic Institute in Halifax, Nova Scotia. IBD (Atlantic) specializes in developing new biodiagnostic methods and technology in neuroscience. He is an Associated Professor of Diagnostic Radiology, Psychology and Neuroscience, and Anatomy/Neurobiology at Dalhousie University, and has additional appointments with the Queen Elizabeth II Health Care Center and the IWK Health Centre. He has over 15 years of experience in developing neurophysiologically-based diagnostic methods to replace traditionally behavioral tests.
Dr. D’Arcy is a member of a number of committees for neuroscience and biomedical imaging and has received several awards of distinction.
