![Page 1: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/1.jpg)
Electromagnetic inverse problems in Electromagnetic inverse problems in biomedical engineering biomedical engineering
Jens HaueisenJens Haueisen
Institute of Biomedical Engineering and Informatics, Technical University of Ilmenau, Germany
Biomagnetic Center, Department of Neurology, University Jena, Germany
Presentation at the Politecnico di Milano on March 14, 2008
![Page 2: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/2.jpg)
Overview
1. Introduction2. Localization of magnetic markers in the alimentary
tract3. The influence of forward model conductivities on
EEG/MEG source reconstruction4. Optimization of magnetic sensor arrays for
magnetocardiography5. Validation of source reconstruction procedures
![Page 3: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/3.jpg)
Magnetocardiography (MCG)• Measurement of magnetic
field produced by the heart• Reconstruction of electric
sources causing the field
![Page 4: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/4.jpg)
MCG
Magnetocardiography (MCG) provides non-invasively information about the electrical activity of the heart. 0
8.9
0
22.0µAmm
lateral
inferobasal
sept
al
QRS
apical
Zoomleft ventricle
Rightventricle
Rightlung
Leftlung
Leftventricle
(a) (b)
(c)
![Page 5: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/5.jpg)
Introduction
• New room temperature optical magnetometers allow customized and flexible sensor arrangements
• Arising question: how do we arrange the sensors optimally?
• Goal function: condition number (CN) of the lead field (LF) matrix
![Page 6: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/6.jpg)
BEM model
Torso LungsVentricular blood masses
![Page 7: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/7.jpg)
Source space
13 current dipoles, distributed around the left ventricle of the heart
![Page 8: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/8.jpg)
The objective function
• LF matrix contains information on geometry of the source space, the boundary element model and the sensor array
• A minimal CN implies an optimal sensor arrangement for a given setup
![Page 9: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/9.jpg)
Discretization of the search space
• Optimization: iterative search for a sensor setup with minimal CN
• But LF computation is slow, therefore pre-computation for a fixed grid of positions & orientations is needed
![Page 10: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/10.jpg)
Constraint Framework for Continuous Optimizers
• Discrete search volume → snap into grid before each CN evaluation
• Minimum distance (MD) of sensors, here 2 cm→ while mean(MD violation) > tolerance
1. pick a sensor with max #clashes2. move all clashing sensors away radially3. snap into grid
• Pro: one representative sensor out of the clashing sensors is kept
![Page 11: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/11.jpg)
Restoring the minimum distance2 cm
UntouchedRepresentative
Physical search volume
![Page 12: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/12.jpg)
Particle Swarm Optimization (PSO)
• A set of candidate solutions (= particles) is randomly initialized
• Each particle has a position and velocity in high-dim. search space
• Each particle has informant particles, whose state it can access
• Iteration = move particles + update velocities + fix constraint
• After constraint fix, the velocities are corrected
![Page 13: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/13.jpg)
PSO algorithm
![Page 14: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/14.jpg)
PSO velocity correction
High-dim. search space
Particle = current solution
Velocity
Shift due to constraint
NewVelocity
½ the shift length
![Page 15: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/15.jpg)
Tabu Search (TS)
• Discrete search: combinatorial selection of s out of r sensors with minimal CN
• The minimum distance constraint is satisfied for all sensor selections
• In each iteration step: find a better selection of s sensors (with lower CN) in the neighborhood of the current solution by exchanging n sensors (during the search n was decreased from s/2 to1)
![Page 16: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/16.jpg)
PSO vs. TS
• TS prevents reevaluations of sensor configurations by memorizing them
• TS is robust against local minima • But: no use of spatial closeness or
gradient, limited to combinations of predefined sensor positions/orientations
• Dense grids (i.e. a higher number of sensors on the same area) may be more difficult to optimize than sparse ones because of the combinatorial complexity
![Page 17: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/17.jpg)
Numerical Results• PSO and TS are implemented in C++ in
SimBio: TS (green) and PSO (blue) optimized setups are very similar
![Page 18: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/18.jpg)
Reduction of CN• Both optimizations significantly reduce CN
![Page 19: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/19.jpg)
Conclusion• Comparable results indicate that optimization of vectorial
sensor setups may be significantly improved• Reconstruction robustness may be improved and the
number of sensors may be reduced while retaining information in terms of CN
• The new quasi-continuous PSO optimization incorporates the gradient and spatial closeness information while being robust against local minima in the goal function
• A fine 3D search volume, projection method based and lower error bound based sensor setup optimizations are planed
Lau, Eichardt, Di Rienzo, Haueisen: Tabu Search Optimization of Magnetic Sensor Systems for Magnetocardiography. IEEE Transactions on Magnetics, to appear 2008
![Page 20: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/20.jpg)
Overview
1. Introduction2. Localization of magnetic markers in the alimentary tract3. The influence of forward model conductivities on EEG/MEG
source reconstruction4. Optimization of magnetic sensor arrays for
magnetocardiography
5. Validation of source reconstruction procedures1. Simulations2. Phantom measurements 3. Animal measurements
![Page 21: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/21.jpg)
Simulations
Wolters et al. SIAM Journal on Scientific Computing, in press, 2007
134 electrodes4-layer sphere model: Radii: 92, 86, 80, 78mm; 0.33, 0.0042:0.042, 1.79, 0.33 S/m
Nodes: 161,086
![Page 22: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/22.jpg)
Simulations
Wolters et al. SIAM Journal on Scientific Computing, in press, 2007
Forward: J.C. de Munck Inverse: FEM
Dipole localization error
![Page 23: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/23.jpg)
Simulations
Wolters et al. SIAM Journal on Scientific Computing, in press, 2007
Forward: J.C. de Munck Inverse: FEM
Dipole orientation error
![Page 24: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/24.jpg)
Simulations
Wolters et al. SIAM Journal on Scientific Computing, in press, 2007
Forward: J.C. de Munck Inverse: FEM
Dipole magnitude error
![Page 25: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/25.jpg)
Phantom measurements
0.5 mm0.5 mm 1 mm
electrodes
![Page 26: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/26.jpg)
Phantom measurements
Liehr, Haueisen et al. Annals of Biomedical Engineering, 2005
![Page 27: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/27.jpg)
Phantom measurements
![Page 28: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/28.jpg)
![Page 29: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/29.jpg)
Wetterling, Haueisen et al. IEEE TBME, in revision, 2008
Phantom measurements
![Page 30: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/30.jpg)
Animal measurements
• Combined ECoG and MEG measurements in rabbits
• median nerve / tibial nerve– current 0.2 - 0.5 mA– Interstimulus interval 503 ms– 2048 averages– latency
• 15 - 20 ms (median nerve)• 20 - 24 ms (tibial nerve)
![Page 31: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/31.jpg)
Animal measurements
Combined electric measurements (ECoG) with Compumedics Neuroscan Synamps
![Page 32: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/32.jpg)
Median nerve Tibial nerveTibial nerve
1.25 mm1.25 mm
Animal measurements
Electric measurements
![Page 33: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/33.jpg)
Median nerve Tibial nerveTibial nerve
Magnetic measurements
Animal measurements
![Page 34: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/34.jpg)
Median nerve Tibial nerveTibial nerve
Magnetic measurements
8.4 mm 8.4 mm
Time point: 17 ms (P1)Increment: 5 fT
Time point: 21 ms (P1)Increment: 5 µV
Animal measurements
![Page 35: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/35.jpg)
Source localization setupSource localization setup
•• 16 MEG pick up coils 16 MEG pick up coils •• 16 electrodes16 electrodes•• One compartment model One compartment model
Animal measurements
![Page 36: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/36.jpg)
Comparison median and tibial nerve
dip 1 - median nerve: 44.8/46.6/50.5 mm; dip 2 - tibial nerve: 46.2/48.2/50.3); calculated dipole distance 2.1 mm
Animal measurements
![Page 37: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/37.jpg)
Influence of anisotropy
median and tibialmedian and tibialnervenerve
isotropicisotropic isotropicisotropic
with anisotropywith anisotropy
![Page 38: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/38.jpg)
Validation results
Validation in a spherical model successfulValidation in a spherical model successful
Validation with two stimulus modalities successfulValidation with two stimulus modalities successful
Validation BEM and FEM successfulValidation BEM and FEM successful
Influence of anisotropy within the procedural limitsInfluence of anisotropy within the procedural limitsfor median and tibial nerve stimulationfor median and tibial nerve stimulation
![Page 39: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/39.jpg)
Thanks to: Thanks to:
Financial support: EU, DFG, BMBF, AIF, TMWTA
Cesare Mario ArturiJohn W. BelliveauLuca Di RienzoJohn S. George Thomas KnöscheYoshio OkadaCeon RamonPaul H. SchimpfDavid S. TuchVan J. Wedeen
The SimBio Team: Carsten Wolters Alfred Anwander Thomas KnöscheMatthias Dümpelmann…
Roland EichardtLars FlemmingFrank GießlerUwe GraichenMaciej Gratkowski Daniel GüllmarBernd HilgenfeldRalph Huonker Mario LiehrUwe Schulze
Hartmut BrauerMichael EiseltHannes NowakJürgen R. ReichenbachHerbert Witte Otto W. Witte
![Page 40: Electromagnetic inverse problems in biomedical engineeringing.univaq.it/emc-chap-it/...Part2_14_mar_2008.pdf · Optimization of magnetic sensor arrays for magnetocardiography 5. Validation](https://reader031.vdocument.in/reader031/viewer/2022011922/6044f04b8aadd822b13b5d19/html5/thumbnails/40.jpg)