Improving the forward model for electrical impedance tomography of brain function through rapid generation of subject specific finite element models

Electrical Impedance Tomography (EIT) is a non-invasive imaging method which allows internal electrical impedance of any conductive object to be imaged by means of current injection and surface voltage measurements through an array of externally applied electrodes. The successful generation of the image requires the simulation of the current injection patterns on either an analytical or a numerical model of the domain under examination, known as the forward model, and using the resulting voltage data in the inverse solution from which images of conductivity changes can be constructed. Recent research strongly indicates that geometric and anatomical conformance of the forward model to the subject under investigation significantly affects the quality of the images. This thesis focuses mainly on EIT of brain function and describes a novel approach for the rapid generation of patient or subject specific finite element models for use as the forward model. After introduction of the topic, methods of generating accurate finite element (FE) models using commercially available Computer-Aided Design (CAD) tools are described and show that such methods, though effective and successful, are inappropriate for time critical clinical use. The feasibility of warping or morphing a finite element mesh as a means of reducing the lead time for model generation is then presented and demonstrated. This leads on to the description of methods of acquiring and utilising known system geometry, namely the positions of electrodes and registration landmarks, to construct an accurate surface of the subject, the results of which are successfully validated. The outcome of this procedure is then used to specify boundary conditions to a mesh warping algorithm based on elastic deformation using well-established continuum mechanics procedures. The algorithm is applied to a range of source models to empirically establish optimum values for the parameters defining the problem which can successfully generate meshes of acceptable quality in terms of discretization errors and which more accurately define the geometry of the target subject. Further validation of the algorithm is performed by comparison of boundary voltages and image reconstructions from simulated and laboratory data to demonstrate that benefits in terms of image artefact reduction and localisation of conductivity changes can be gained. The processes described in the thesis are evaluated and discussed and topics of further work and application are described.