Projection-based preprocessing for electrical impedance tomography to reduce the effect of electrode contacts
A. Jääskeläinen, J. Toivanen, A. Hänninen, V. Kolehmainen, N. Hyvönen · 2024
Read Projection-based preprocessing for electrical impedance tomography to reduce the effect of electrode contacts by A. Jääskeläinen, J. Toivanen, A. Hänninen, V. Kolehmainen, N. Hyvönen (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-15009.txt.
Introduction
Altti Jääskeläinen and co-authors present a data cleaning step that removes electrode contact errors from electrical body imaging, in a paper from 2024. The imaging method infers inner conductivity from skin electrodes, but loose contacts distort the picture. Their step avoids estimating the contacts directly.
The idea is to project the data and the model away from the directions that contact changes can produce. Reconstructions stayed good in tests with simulated and real tank data, even when the projection was built at values far from the truth. The same trick can tell whether a signal change came from the body or from a contact. The authors note uses such as bedside stroke monitoring.
Catalog record
| CREATOR | A. Jääskeläinen, J. Toivanen, A. Hänninen, V. Kolehmainen, N. Hyvönen |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 65 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-15009
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Projection-based preprocessing for electrical impedance tomography to reduce the effect of electrode contacts A. Jääskeläinen, J. Toivanen, A. Hänninen, V. Kolehmainen, N. Hyvönen Abstract This work introduces a method for preprocessing measurements of electrical impedance tomography to considerably reduce the effect uncertainties in the electrode contacts have on the reconstruction quality, without a need to explicitly estimate the contacts. The idea is to compute the Jacobian matrix of the forward map with respect to the contact strengths and project the electrode measurements and the forward map onto the orthogonal complement of the range of this Jacobian. Using the smoothened complete electrode model as the forward model, it is demonstrated that inverting the resulting projected equation with respect to only the internal conductivity of the examined body results in good quality reconstructions both when resorting to a single step linearization with a smoothness prior and when combining lagged diffusivity iteration with total variation regularization. The quality of the reconstructions is further improved if the range of the employed projection is also orthogonal to that of the Jacobian with respect to the electrode positions. These results hold even if the projections are formed at internal and contact conductivities that significantly differ from the true ones; it is numerically demonstrated that the orthogonal complement of the range of the contact Jacobian is almost independent of the conductivity parameters at which it is evaluated. In particular, our observations introduce a numerical technique for inferring whether a change in the electrode measurements is caused by a change in the internal conductivity or alterations in the electrode contacts, which has potential applications, e.g., in bedside monitoring of stroke patients. The ideas are tested both on simulated data and on real-world water tank measurements with adjustable contact resistances. keywords Electrical impedance tomography, projection, contact conductivity, total variation, lagged diffusivity iteration, preprocessing of data AMS 65N21, 35R30, 35J25 1 Introduction Electrical impedance tomography (EIT) is a method for sensing the internal conductivity of an object by driving electrical currents through it using electrodes placed on the object’s surface and measuring resulting voltages on the electrodes. This measurement process defines a nonlinear inverse elliptic boundary value problem for determining the internal conductivity distribution of the object given the measured voltages and the applied currents [ 6 , 11 , 29 ] . Forming an accurate conductivity reconstruction with EIT can often be difficult due to model uncertainties that may clearly exceed the effect that conductivity changes in the object’s interior have on the measurement data [ 4 , 7 , 23 ] . One of the significant and common sources of error arise from inaccurately known contacts between the electrodes and the surface of the object. The strengths of the contacts can be estimated as auxiliary unknowns (in addition to the conductivity) in the nonlinear reconstruction problem with the cost of a more complicated inverse problem and slightly increased computational cost [ 17 , 32 ] . Many applications of EIT, such as monitoring of stroke [ 27 , 28 ] or lung function [ 1 ] , are based on dynamic imaging, aiming at the reconstruction of temporal changes in the conductivity using measurements at different time instants. In these applications, estimating the contacts may be nontrivial, especially if employing the widely popular linearized difference imaging algorithms; see, e.g., [ 1 , 3 ] . The linearized approach is often used due to its simplicity, but the reconstructions may suffer from significant artifacts if the contacts change between the measurements, which is a potential hindrance in monitoring of stroke and constitutes the main motivation for this work. We introduce a method for dealing with the uncertainty in the contacts by projecting the measurement data and the forward map onto the orthogonal complement of the range of the Jacobian matrix of the electrode measurements with respect to the contact strengths. The performance of the approach is further improved if the image space of the projection is also constructed to be orthogonal to the range of the Jacobian with respect to the electrode positions. We show experimentally that while the contact Jacobian depends on the initial estimates for the contacts and the internal conductivity of the imaged object, the subspace onto which one projects is almost independent of them, implying that the approach has high tolerance for uncertainty in the initial estimates of the contact parameters. Although not tested in this work, this observation also motivates forming the projection matrix only once even if resorting to some nonlinear reconstruction method. The EIT forward problem of solving the electrode potentials given the internal conductivity and applied currents is modelled in this work by the smoothened complete electrode model (CEM) [ 20 ] . The CEM models the surface electrodes as subsets of the examined object’s boundary, and it has been found to predict real-world experiments with high accuracy [ 12 ] . In the smoothened CEM, the contacts are characterized by the contact conductivities over the electrodes, with the option to use smooth contact profiles to achieve better regularity properties for the forward solution. We use the same predefined contact conductivity profile on each electrode and parametrize the strengths of the contacts via their peak values at the centers of the electrodes, which means that the aforementioned contact Jacobian is formed by computing Fréchet derivatives with respect to these peak values. We demonstrate the effects of the projections by computing three-dimensional reconstructions from experimental data measured on a water tank with significant errors in the initial estimates for the contact strengths. Two test cases are considered: the electrode contacts are altered either by partially covering some electrodes by duct tape or by using electrode leads that are equipped with tunable resistors for changing their resistances. Two Bayesian reconstruction algorithms are applied to the forward model linearized with respect to the internal conductivity only, with an accurate initial guess for the conductivity of the salt water filling the tank but highly inaccurate estimates for the contacts on some electrodes. The total variation (TV) prior (cf. [ 25 ] ) combined with the lagged diffusivity iteration [ 33 ] is used to compute reconstructions with clear edges between regions of near-constant conductivity, while the other algorithm resorts to a simple smoothness prior. Although in difference imaging applications the reference data, with respect to which the linearization is performed, are typically measured at a different time instant (to cancel out modeling errors), we simulate the reference measurements to demonstrate that our approach also has potential for absolute imaging. For both test cases and algorithms, employing the contact conductivity projection significantly reduces the reconstruction artifacts caused by mismodeling of the contacts, and additionally incorporating the projection with respect to the electrode positions in the algorithms entirely removes the artifacts. In addition to testing the projection approach in the linearized reconstruction problem of EIT, we also numerically study how the considered projections affect changes in the electrode measurements caused by alterations in the internal conductivity or the contacts. These studies are based on water tank measurements and the three-layer head model introduced in [ 8 , 9 ] , and they demonstrate that the signal in the measurements due to altered contacts is significantly reduced by