International Journal of Bioelectromagnetism Vol. 4, No. 2, pp. 17-18, 2002. |
www.ijbem.org |
transmembrane potential reconstruction
|
(1) |
where and are the intracellular and extracellular conductivity tensors; and are the extracellular and transmembrane potentials.
Eq. (1) is solved using the finite element method (FEM). Our finite element model comprised 50000 nodes corresponding to 327000 tetrahedrons. About 17000 nodes were placed regularly in the heart to allow for the simulation of electrocardiograms (ECG) and for the construction of the lead-field matrix. The fiber orientation in the myocardium as well as in the skeletal muscles was also available [5].
The simulation of “realistic” TMP distributions during the complete cardiac cycle was performed with a cellular automaton. It is based on a highly-detailed voxel heart model having the resolution of 1´1´1 mm.
The inverse solution was sought in the form of 3D TMP patches. Every patch corresponds to one node of a simplified coarse finite element heart model (955 nodes). The sequence of patches was generated by setting the TMP at the chosen node of the coarse grid to 1 and the rest to 0 and interpolating the TMP distribution onto the nodes of the fine finite element mesh using first-order interpolation functions for tetrahedrons. The correspondent ECG signal is then copied to the corresponding column of the lead-field matrix.
Zero- and second-order Tikhonov regularization were used to solve the inverse problem [6]. Tikhonov zero-order regularization is based on the singular value decomposition of the lead-field matrix A and the inverse solution is written as
(2) |
where are the singular values, and are left and right singular vectors of A, and is the regularization parameter. The optimal regularization parameter was found by the method of L-curve [6].
For second-order Tikhonov regularization, the iterative “preconditioned” conjugate gradient algorithm from [7] was applied. The L-curve was used to find the optimal number of iterations.
The TMP distributions were modeled with a cellular automaton and forward computations were performed to get the ECG.
|
Fig. 1 Transmembrane potential distribution simulated with a cellular automaton at one instant of time. Four horizontal slices through the heart volume are shown as seen from the top of the torso.
An example of a computed TMP corresponding approximately to the middle of the T-wave is presented in Fig. 1. Four slices through the heart volume are shown as seen from the top of the torso.
Fig. 2 demonstrates the results of Tikhonov zero-order reconstruction of the TMP from the simulated ECG signal with 0.5% Gaussian noise added. The range of transmembrane potentials corresponds approximately to that of Fig. 1 although the values are somewhat shifted upwards. It should be noticed that according to Eq. 1 two TMP distributions which differ only by a constant value are considered to be equivalent.
|
Fig. 2 Transmembrane potentials reconstructed from the simulated measurement data with 0.5% noise using Tikhonov zero-order regularization. The same slices and the time instant as in Fig. 1 are shown.
An example of a TMP distribution reconstructed with Tikhonov second-order regularization is shown in Fig. 3. For both approaches it is seen that only low spatial frequency components of source distributions could be reconstructed.
|
Fig. 3 Transmembrane potentials reconstructed from the simulated measurement data with 0.5% noise by Tikhonov second-order regularization. The same slices and the time instant as in Fig. 1 are shown.
DISCUSSION
The work demonstrates a new volume-oriented approach for inverse TMP computations. The method allows for taking into account the anisotropy of the heart muscle which was not possible in the inverse procedures developed so far. Tikhonov second-order regularization delivered generally smoother and less noisy solutions as the zero-order regularization. The simulated TMP distributions could be reconstructed better during the repolarization phase. This can be explained by the relatively low content of high spatial frequencies in the source distributions during this stage.
[1] G. Huiskamp, F. Greensite “A new method for myocardial activation imaging”, IEEE Trans. Biomed. Eng., vol. 44, no. 6, pp. 433-446, 1997
[2] P. Wach, B. Tilg, G. Lafer, W. Rucker “Magnetic source imaging in the human heart: estimating cardiac electrical sources from simulated and measured magnetocardiogram data”, Med. Biol. Eng. Comput., vol. 35, pp. 157-166, 1997
[3] C. D. Werner, F. B. Sachse, O. Dössel “Electrical excitation propagation in the human heart”, In CardioModel 2000, Computer Models of the Heart: Theory and clinical application, International Journal on Bioelectromagnetism, vol. 2-2
[4] C. S. Henriquez “Simulating the electrical behavior of cardiac tissue using the bidomain model”, Crit. Rev. Biomed. Eng., vol. 21, pp. 1-77, 1993
[5] F. B. Sachse, M. Glas, M. Müller, K. Meyer-Waarden “Segmentation and tissue-classification of the Visible Man dataset using the computertomographic scans and the thin-section photos”, In Proc. First Users Conference of the National Library of Medicine’s Visible Human Project. 1996
[6] P. C. Hansen “Rank-deficient and discrete ill-posed problems”, Philadelphia: SIAM, 1998
[7] P. C. Hansen “Regularization Tools. A Matlab package for analysis and solution of discrete ill-posed problems”, Technical Report, Dept. of Mathematical Modeling, Technical University of Denmark, 1998