Publication Details
Overview
 
 
 

Proceedings 15th Annual International Conference IEEE Engineering in Medicine & Biology Society

Contribution To Book Anthology

Abstract 

Building on the progress that has been made in the use of recurrent neural networks to solve optimisation problems, a new self-adaptive neural network is proposed that promises to improve upon existing image reconstruction techniques in Electrical Impedance Tomography. Interpreting image reconstruction in terms of interacting relaxations processes, a neural network formulation emerges that naturally leads to a parallel, cooperative algorithm that can easily be mapped to a massively parallel processor. The dynamics of the network are surprisingly simple and retain the possibility of analog implementation. Compared to algorithms currently used in EIT, the proposed method differs in its emphasis on relaxation processing and rigorous mathematical support.

Reference 
 
 
DOI ieeexplore