Name: Richard Bayford

Email: r.bayford@ucl.ac.uk

CompanyName: Middlesex University

Country: UK

Abstract: Development of algorithms to image impedance changes
inside the human head

R. H Bayford*+, AP Bagshaw*, AD Liston+, A Tizzard+ and D S Holder*

+Middlesex University, Archway Campus, Furnival Building, Highgate Hill, London N19 5LW, UK
*Department of Clincal Neurophysiology, Middlesex Hospital, Cleveland Street, London W1T 3AA, UK

Abstract.

It is assumed that inclusion of head geometry in the forward model will improve
the localisation accuracy of image reconstruction algorithms for imaging
impedance changes inside the human head. Initially we developed an algorithm,
which assumed that the human head could be modelled as a simple homogenous
sphere or multiple shelled spheres. This is an assumption often used in
localising dipoles in the human head from measurement of surface potentials
(Electroencephalogram EEG). This did not always yield a solution with correct
localisation.
To test the idea that including geometric information will improve localisation
we have developed a new difference reconstruction algorithm for Electrical
Impedance Tomography (EIT) that incorporates a forward model derived from a
Finite Element model of the human head.
The new algorithm uses a sensitivity approach. The linear sensitivity matrix is
created from a geometrically accurate model of the human head obtain from
segmented MRI data. Generating meshes of the human head is difficult due to the
complex geometry and multiple shells, i.e. scalp, skull, CSF and brain. This
can produce distorted elements, which impact of the quality on the reconstructed
image. Also the number of elements in scalp and CSF are limited reducing the
accuracy of the final solution. We examine the effect of the mesh quality on
the reconstructed image.
The inverse solution was obtained by normalising the sensitivity matrix then
inverting it using truncated Singular Value Decomposition (SVD).
We present the results of this new image reconstruction algorithm and compare
its image quality and location accuracy against an algorithm which models the
head as four concentric spheres.

Close Window