WIT Press


Evolving Neural Networks To Flow Cytometric Data Interpretation

Price

Free (open access)

Paper DOI

10.2495/DATA050421

Volume

35

Pages

10

Published

2005

Size

362 kb

Author(s)

G. C. Pereira, A. Bonomo & N. F. F. Ebecken

Abstract

Flow cytometry is a method largely employed in research and medical practice as a diagnostic exam on grave diseases, such as infections by HIV viruses and mainly leukemia. This paper aims to introduce a new methodological approach. Artificial neural networks based on supervised training are applied to pattern recognition of their optical blood cell population coming from patients. Nonlinear system identification via artificial neural networks consists of adjusting the network in such a way that it approximately describes the inputoutput mapping of the system. In its complete form, network induction involves both parametric and structural learning. The two tasks can be seen as optimization problems and, as such, they may present multiple local minima, non-differentiability and discontinuity. The idea of applying Genetic Algorithms to the problem of complete network acquisition comes naturally. The preliminary results demonstrate that this approach is promising and can soon contribute to protocols related to this disease. Keywords: flow cytometry; neural network; diagnostic. 1 Introduction Flow cytometers are equipments which use light dispersion and quantify different cellular populations (Van Dam and Tjalma [28]). Depending of the arrangement it can obtain two or thirteen simultaneous parameters for each cell or organism. Many diseases depend on the right \“classification” from flow cytometers to give an accurate diagnosis and subsequent treatment. Patients infected, for instance by HIV viruses need their CD4+ cells quantified to suitable

Keywords

flow cytometry; neural network; diagnostic.