Electrical impedance tomography with neural networks and fuzzy sets
Various imaging procedures, providing different types of information and based on different physical properties of the tissues, are characterized by specific complexities, degrees of hazard, resolutions and price. Electrical impedance tomography (EIT) has the advantage of a low price and simplicity, as well as of not having any known hazards, such as ionising radiation, which makes it particularly attractive. However, it has the disadvantage of a low resolution. The information provided by EIT could be used directly for medical diagnosis or in combination with other imaging systems. In EIT the electric current is applied to the periphery of the body and the corresponding voltage is measured in order to find the internal distribution of conductivity and permittivity. In this thesis, the EIT performance has been improved in three different aspects: (1) improving the network approximation method for solving the electrode voltages when the distribution of conductivity and permittivity within the object as well as injecting current pattern are known, namely, the forward problem; (2) utilizing neural networks in an iterative procedure for solving the conductivity distribution when the electrode voltages as well as injecting current pattern are known, namely, the inverse problem; and (3) improvising a new method for image fusion based on fuzzy set theory to be used in a multifrequency scheme to increase certainty and accuracy.