Neural networks in transmission and emission tomography

• Grant number: 05/03414-4
• Support Opportunities: Scholarships in Brazil - Master
• Start date: May 01, 2006
• End date: March 27, 2007
• Field of knowledge: Physical Sciences and Mathematics - Mathematics - Applied Mathematics
• Principal Investigator: Alvaro Rodolfo de Pierro
• Grantee: Juliana Verga
• Host Institution: Instituto de Matemática, Estatística e Computação Científica (IMECC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

Emission Computed Tomography (ECT) aims at determining the emission densities of a metabolic compound tagged with a radioactive isotope from data obtained by detectors outside the body. Transmission tomography (CT) is the determination of the tissue atenuation inside the body. In the absence of errors both can be modelled by the Radon transform and the mathematical problem consists of computing its inverse. In practice many corrections are necessary to obtain reasonable images (high levels of attenuation in ECT, lack of data in CT, etc). These corrections avoid the necessity of dealing with a more realistic but nonlinear model. One alternative is to deal with the nonlinearities using artificial neural networks (ANN). We will study the development of appropriate ANNs for the case of estimating the attenuation from activity data only in ECT and for eliminating artifacts in the case of incomplete data in CT. (AU)