The Combination of Principal Component Analysis with Neural Network as a Highly Efficient 3D EIT Solution

Magdalena Stasiak1, Krzysztof Siwek2, Ryszard Sikora3, Stefan F Filipowicz2 and Jan Sikora2

1 The Institute of Electrical Apparatus, Technical University of Lodz, Stefanowskiego 18/22, 90-924 Lodz, Poland (e-mail: stasiak@ck-sg.p.lodz.pl)

2 The Institute of Theory of Electrical Engineering and Electrical Measurements, Warsaw University of Technology, Koszykowa 75, 00-661 Warsaw, Poland (e-mail: 2xf@iem.pw.edu.pl)

3 The Chair of Theoretical Electrotechnics and Computer Science, Technical University of Szczecin, Wl.Sikorskiego 37, 70-313 Szczecin, Poland

ABSTRACT

The recent industrial and medical interest is focused on real-time reconstruction. That is not possible in Electrical Impedance Tomography (EIT) using the classical approach, especially in 3D space. Therefore, in this paper 3D Boundary Element Method (BEM) for thin layers in EIT and the solution to the inverse problem using Neural Network is presented.

Thin layers like skull in EIT cause many problems to geometrical representation. Finite Element Method, which is time-consuming in 3D space, is usually used to the solution of the forward problem. The BEM, which represents only a discretization of the surface, reduces the number of necessary elements as a consequence the computation time.

In order to solve the inverse problem, the Neural Network method was applied. For selection of neural network size, which is the one of more important and complex problems, the Principal Component Analysis (PCA) is used. PCA decomposes high-dimensional data into a low-dimensional subspace component. In this way, the size of input vector to train the neural networks is limited.

Keywords Impedance Tomography, Boundary Element Method, Principal Component Analysis, Neural Network