[IEEE 2012 ICME International Conference on Complex Medical Engineering (CME) - Kobe, Japan...
Transcript of [IEEE 2012 ICME International Conference on Complex Medical Engineering (CME) - Kobe, Japan...
Proceedings of 20121CME International Conference on Complex Medical Engineering
July I - 4, Kobe, Japan
Uncertainty Assessment of Target Localization
for rTMS Treatment Y oshihiro Yasumuroa,b Koichi Hosomi b, Y oichi Saitoh b Taiga Matsuzaki b,c
a Faculty of Environmental and Urban
Engineering, Kansai University
3-3-35 Yamate, Suita, Osaka, Japan
b Department ofNeuromodulation and
Neurosurgery, Office for Univers ity
Industry Collaboration, Osaka University
2-1, Yamada-oka, Suita, Osaka, Japan
C Home Healthcare Research and Develop
ment Department,
Teijin Pharma Limited
3-2-1, Kasumigaseki, Chiyoda, Tokyo, Japan
[email protected], [email protected]
Abstract - Repetitive transcranial magnetic stimulation
(rTMS) is a non-invasive method for treating various neurological
and psychiatric disorders. This paper focuses on a technical as
pect of this treatment method for neuropathic pain that can be
caused by a lesion or disease of the central or peripheral nervous
system, including stroke, trauma or surgical operation. rTMS
uses electromagnetic induction to induce weak electric currents
by rapidly changing the magnetic field. Targeting a specific part
of the brain to locate the magnetic field works as a treatment for
pain relief. This is the reason why the current style of rTMS
treatment is still developing and is so technically specialized that
only a limited number of hospitals and only a handful of special
ists can provide this therapy. The existing systems of rTMS are
based on an optical maker-based 3-dimensional (3D) sensing
technique for positioning the stimulation coil to target the small
spot in the region of interest in the brain, and for referring pre
scanned MRI data to check the target position. Furthermore, the
existing systems require the patient to be fixed on a bed in which
optical markers for 3D sensing are placed during the treatment to
maintain positioning precision. With the growing demands of
neuropathic pain patients and their increasing numbers, new
approaches and systems to achieve more supportive and easy-to
handle navigation have been proposed lately for rTMS treat
ment. This paper proposes a quantitative index for localization
precision, considering an uncertainty measure. Uncertainty is
technically defined as a parameter, associated with the result of a
measurement that characterizes the dispersion of the values that
could reasonably be attributed to the measurement. This paper
shows a detailed example of the uncertainty derivation for rTMS
treatment, based on trial tasks for searching the spots on the
brain that cause muscle twitch.
Index Terms - Repetitive Transcranial Magnetic Simulation,
Stimulation Spot, 3-dimensional Localization, Uncertainty Analy
sis
I. INTRODUCTION
A. rTMS Treatment
Repetitive transcranial magnetic stimulation (rTMS) has been gathering attention as a non-invasive method for treating various neurological and psychiatric disorders including strokes, Parkinson's disease, and depression. This paper focuses on a new treatment to alleviate neuropathic pain that is caused by a lesion or disease of the central or peripheral nervous system, including stroke, trauma, or a surgical operation.
rTMS uses electromagnetic induction to induce weak electric
978-1-4673-1618-7112/$31.00 ©2012 IEEE 784
currents by rapid change of a pulsed magnetic field; thus, rTMS treatment is capable of stimulating specific parts of the brain, such as the primary motor area for neuropathic pain, with minimal discomfort in a non-invasive manner [1].
B. Existing and Developing System for rTMS
Since the effects of rTMS last only several hours, rTMS treatments need to be available whenever the patient needs these treatments. In fact, rTMS treatments are only available at
specialized clinics such as university hospitals, because only experienced physicians in a limited number of hospitals can use the expensive and complicated rTMS system in its present circumstances [2,3]. The existing treatment method of rTMS is achieved by an optical 3-dimensional (3D) sensing technique for positioning the stimulation coil to target the spot in the brain, especially in the primary motor cortex (Ml) re
gion[4]. Before treatment, the doctors are supposed to measure
the 3D position of the feature points on the patient's head.
These positions allow the system to register the MRI data of the preoperative examination in the sensor coordinate by using the corresponding feature points in the MRI data. Then the positioning sensor tracks the coil, using a set of optical markers installed on the sensor. Showing the relative position of the coil and the registered MRI data helps the doctor to target the spot in the brain. To ensure positioning accuracy during treatment, the patient must be immobilized on the bed throughout the treatment, because the registered patient's feature points are based on the optical markers fixed on the bed as shown in Fig. 1.
Fig. 1 A typical style of rTMS treatment (Osaka University Hospital)
Fig. 2 Brain surface model for rTMS navigation. The primary motor cortex (MI ) region is highlighted in the left picture and recorded stimulation positions for the right hand and right foot regions are indicated by groups of bars in the right picture.
To enhance the localization precision, Richter et al. proposed a robot-guided TMS system that utilizes a direct object that tracks 3D shapes of the markers and the coil without additional
markers [6]. This approach is a straightforward way to track the object by matching the online captured shape of the object with pre-registered shape data. The pre-registered shape for the patient's head is prepared from MRI and the coil shape is
from CAD data. A laser scanner captures the current shapes and an ICP (iterative closest points) algorithm performs the matching. Fukushima et al. proposed a magnetic navigation system designed for home use of rTMS[5] using inexpensive and small-sized magnetic sensors. By collecting the spatial data samples of the magnetic field to record the proper position and orientation of the stimulation coil at the initial rTMS treatment, the system can help the user navigate and reproduce
the coil position to target the proper spot for subsequent treatments at home. This system theoretically allows any users, even those without medical knowledge and techniques, to easily locate the coil for subsequent treatment. The magnetic sen
sors are fixed on a glasses-style mount device that the user can easily wear when applying the treatment with no need to be constrained on the bed in a treatment room. Since the positioning accuracy depends on the reproducibility of wearing the
glasses-style mount device, that device must be elaborately designed for adjustment to the individual patient in practical steps. Yasumuro et al. proposed an rTMS system of free-ofconstraints style for the patient by a marker-less localization, using shape-based tracking techniques incorporated with im
age processing to recognize the patient's face and face parts[7]. A voiding the computing cost for ICP process, the patient's face and its parts; e.g. the eye-edges, the nose top and the lips are tracked on the video frames with the stereo camera. Only selected parts are triangulated to compute the 3D positions, which are used for estimating head movement of 6DOFs.
As mentioned above, a number of rTMS navigation methodologies and supportive systems have been proposed and developed with a variety of sensing techniques and design approaches. Their common direction is to achieve precise 3D localization of the stimulation coil relative to a specific portion of the brain. The required specification for localization of the effective target size basically relies on the observations of muscle twitches; Hirayama et al. mentioned that the effective
785
spot size is 1 cm region, for example [1]. However, no detailed quantitative analysis for describing the tolerant localization error has ever been conducted. This paper introduces a suitable metric of measurement uncertainty for effective rTMS lo
calization based on a quantitative data collection incorporated with twitch observation and data classifications. This approach also takes into account the further advent of newly developed systems in order to support or replace the specialized handling procedures for coil positioning required of the doctors. Our
metric can be used for designing such new systems as well as for analysis of performance and for checking the qualities of existing equipment.
II. DATA COLLECTION
We employ 4 medical doctors in a department of neurosurgery to perform a task to search for the "sweet spot", where muscle twitch by the rTMS is most clearly observed. After repeating the task 10 times, the doctors acquire a set of 10 coil positions to target the same sweet spot. The data collection task was conducted at Osaka University Hospital where they use a navigation system, Brainsight (Rouge Research, Inc.) for the usual rTMS treatment. Brainsight consists of a stimulation unit, MagPro XI00 with a coil MC-B70 (MagVenture, Inc.), a 3D positioning sensor unit, Polaris (ND!, inc.) [3] to perform 6
DOFs tracking with optical markers, and a visual software with a graphical user interface (QUI). The QUI is capable of
showing a brain surface model constructed from the patient's MR!, on which the doctors can localize the coil by themselves.
This software also has a registration function that can align the patient's head posture with the pre-scanned MRI data as well as track the live coil position and orientation in an identical 3D space. The relative geometry between the coil coordinate and
the convolutional pattern on the brain can be monitored. We utilize Brainsight to record 6 DOFs of the coil coordinate when the doctor reaches the sweet spot. Table I shows the 9 combinations of the 4 medical doctors, A, B, C, and D who played the role of task operators, and then the patient in turn.
For each combination, 10 trials to localize the sweet spot have been done, and a total of 90 sweet spots data were acquired for both hand and foot twitches.
TABLE I TRIAL COMBINATIONS OF OPERATORS AND SUBJECTS
Operator \ Subject A B C D
A ---------- 0 0 -
B 0 =:::::::::::: 0 -
C 0 0 ---- -
D 0 0 0 ----------
III. UNCERTAINTY IN LOCALIZATION FOR RTMS
A. Data Classification by PCA
Fig. 3 and Fig. 4 show the standard deviation of the coil position and targeted positions along with the depth from the
coils. The coil directions are independently measured, but the set of coil directions has a low dispersion with almost linear
..............
20 30 Depth from the Coil Position (mm)
Fig. 3 Standard Deviation of the coil positions and targeted positions relative to the depth in the hand region of MI
1800 ,---,-----,----,-----.-----,-----,-----,-----, 16 .00 1--+--+--+--+---+---+---+---1
E S
14.00 " 12.00 1--+--+--+--+---+---+---+---1 .� 10.00 1--+--+---±--+----F---+---+---I .:; 8.00 +---+--+--+--+---+---+--+--1 8 6.00 1--+-:-: .. -:-: .. -:-: . . "" . .... ...... � . ......... �+-� ........ � . . '-'. .. '-'.. 'f-'-" '-'-' '=-'-" '-"'1'--1
] 4 .00 "'" 8 2.00 Ci5 0.00
10 20 30 Depth from the Coil Position (mm)
Fig. 4 Standard Deviation of the coil positions and targeted positions relative to the depth in the foot region of Ml
spread. The dispersion ratio is 3.1 nun relative to the 30 nun
depth, which is equivalent to 5.8 deg spread in the hand re
gion. Similarly for the foot region, the dispersion ratio is
equivalent to angle 2.0 deg spread. Fig. 5 shows an example of the schematic relation between
the head skin surface, the brain surface, the coil positions, and directions. The green-colored coil positions could trigger a right-hand twitch, and the red positions a right-foot twitch. The length of each bar indicates the depth of 30 nun from the coil position. The concentric circles show the 1 nun interval for a scale reference. This figure also shows that the coil directions are stable toward the inside of the skull without large disper
sions. The coil tracking is performed in 6 DOFs, but recorded positions and orientations in the 3D coordinate have anisotropic distributions. Since the coil is placed on the head surface, displacement of the coil positions along the depth direction is physically constrained. The distribution of the coil positions is along the skin surface, but not anisotropic. Fig. 6 is an example of variation that shows dispersions in targeting the same sweet spot for the same subject; e.g., forms of linear or curved path. These dispersion tendencies may be caused by multiple factors such as bias by the doctor's knowledge, the doctor's own procedure to handle the coil, or the effects by the brain convolution pattern in the navigation image display.
Based on this observation, we apply principal component analysis to each data set of 10 sweet spots. The 3rd principal axis shows the lowest variance direction, which is along the depth direction. The 1 st principal axis finds the highest variance direction, which approximates the direction along the
786
lined distribution of the sweet spots. The second axis shows the width direction of the lined distribution. Since the width
size of the lined distribution should not be missed, we thus select the 2nd axis distribution to measure the uncertainty. We project the 10 sweet spot coordinate onto the 2nd principal axis to acquire a 1 DOF distribution. Figs. 7, 8, and 9 show the distribution project onto the 1st, 2ndm and 3rd principal axes, respectively.
B. Uncertainty Metric
As described in the Guide to the Expression of Uncertainty in Measurement (GUM) [8,9], the above mentioned disper
sions of the coil position (J"c can be categorized as A-type un-
certainty. This includes the uncertainty of human factors such as coil manipulation by the doctors as well as the psychological and biological response of the patient's nerves. As mentioned in the previous subsection A, we use the 2nd principal axis distribution from the PCA result and sum up the 90 sweet spot distributions from 9 data sets (see Fig.8). Assuming that
the same system can be used for rTMS treatment for the hand and the foot, a more severe localization should be the standard specification. Therefore, we adopt the distribution data of the foot twitch, whose standard deviation is 2.8 nun . (Fig. 8 right)
As for the registration of the MRI, 4 feature points are used; the nasal point, nose top, and anterior auricular points, for instance, are specified on both the MRI image and the actual subject's head. A decisive scale depends on the MRI im
age on which the doctor manually specifies the feature position
Fig. 5 Schematic relation between the head skin surface, the brain surface, the coil positions, and the directions. The green-colored coil positions found the right-hand twitch and red coils found the right -foot twitch. The bars indicate the depth of 30 mm from the coil positions and the concentric circles show the I mm interval for a scale reference.
Fig. 6 Variation examples for the same subject with different operators.
with a mouse cursor on a computer screen. The milllmum error for pixel selection can be approximated by the MRI image pixel resolution, which is (JR =1.1 mmJpixel in this case.
The fluctuation by this registration can be also categorized as A-type uncertainty. On the other hand, accuracy (Jp of the
3D positioning sensor is 0.25 mm in RMS according to the specification, and is categorized as B-type uncertainty. Eventually, the combined standard uncertainty u can be calculated
in the form of RMS as follows:
25% 20% 15% 10%
:: � � � � � 0 5 W � m
Range from the Average Position (mm)
25% 20% 15% 10% 5% 0%
� � � � 0 5 W � m Range from the Average Position (mm)
Fig. 7 Frequency distribution of the range of the sweet spots projected on the 1st principal axis
(1) 25% 25% 20% 20%
To give an approximate coverage probability, the combined 15% 15% standard uncertainty u is multiplied by the standard GUM k- 10% 10% value. k = 2 could be regarded as an approximate coverage
probability of 95%, that is if k*2 or if k =2 gives another
level of confidence other than 95%. The resulting uncertainty
range U is derived as follows:
u = k u = 6.0 mm . (2)
This uncertainty range can be used as a minimum requirement for the total localization performance of a rTMS system, including all the errors and ambiguities that arise from the device components and human activities throughout? the treatment.
IV. CONCLUDING REMARKS
This paper proposed a logical scenario to introduce a systematic index for an accuracy assessment of localization for rTMS treatment, which may include a number of ambigu
ous factors not only from the physical aspect of devices or computation errors, but also from human-related causalities. This scenario is applicable for different methodologies and protocols by enumerating the error components and combining them as a standard uncertainty. Another contribution of this paper is that we clarified the anisotropic nature of the existing localization method by applying the PCA approach. In other word, it is possible to examine the redundancy in the performance of the existing tracking systems for rTMS treatment. This kind of analysis leads to efficient simplification and improvement of the mechanism and the process for further design
of rTMS systems. Our next step is to increase the number of data to acquire
a more generic uncertainty specification for the rTMS system. To support both of the necessary and the sufficient conditions
in our logic, clinical validation based on the derived uncertainly value is also within our focus.
ACKNOWLEDGMENTS
This work was supported in part by a grant from Teijin
Pharma Limited.
5% 5% 0% 0%
� � � � 0 5 W � m � � � � 0 5 W � m Range from the Average Position (mm) Range from the Average Position (mm)
Fig. 8 Frequency distribution of range of the sweet spots projected on the 2nd principal axis
50% 50% 40% 40% 30% 30% 20% 20% 10% 10%
O% -_�.I 0% ·20 ·15 ·W ·5 10 15 20 ·20 · 15 · 10 ·5 5 10 15 20
Range from the Average Position (mm) Range from the Average Position (mm)
787
Fig. 9 Frequency distribution of the range of the sweet spots projected on the 3rd principal axis
REFERENCES
[11 A. Hirayama et aI., "Reduction of intractable deafferentation pain by navigation-guided repetitive transcranial magnetic stimulation of the primary motor cortex, " Pain, vol. 122, no. 1-2, pp. 22-27, 2006.
[21 Rouge Research Inc., http://www.rogue-research.com/ [3] Advanced Neuro Technology: http://www.ant-neuro.com/products/visor/ [4] NDI, Inc. POLARIS.http://www.ndigital.com/medicaIlpolarisfamily.php/. [5] L. Richter, R. Bruder, A. Schlaefer, and A. Schweikard, 'Towards direct
head navigation for robot-guided transcranial magnetic stimulation using 3D laserscans: idea, setup and feasibility, " In Proceedings of the 2010 Annual international Cotiference of the iEEE Engineering in Medicine and Biology SOCiety, pp. 2283-2286, 2010.
[6] T. Fukushima et aI. , A. Nishikawa, F. Miyazaki, K. Uchida, M. Sekino, Y. Saitoh, "Magnetic guided transcranial magnetic stimulation: newly convenient navigation system, " International Journal of Computer Assisted Radiology and Surgery, vol. 5, no. suppl.l , pp.36-37, 2010.
[7] Yoshihiro Yasumuro et al." 3-Dimensional Visual Navigation for Repetitive Transcranial Magnetic Stimulation Treatment, The 21st International Conference on Artificial Reality and Teleexistince (ICAT2011), pp. 78-82, 20 II.
[8] Guide to the Expression of Uncertainty in Measurement, first edition, 1993, corrected and reprinted 1995, International. 14/30. 98-3:2008. Organization for Standardization (Geneva, Switzerland)
[9] D. Deavor, 1998 Guardbanding and the world of ISO Guide 25: is there only one way? NCSL Workshop & Symposia; Nicholas M 2004, Guardbanding using automated calibration software, NCSL International Symposium & Workshop.