Artificial neural network for dipole localization

Pei-Chen Lo*, C. T. Su

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

Abstract

This study aims to investigate the problem of focal source localization based on artificial neural network (ANN) techniques. Our major goal is to estimate the source location and strength in a single-source model. Estimating the focal source from the multi-channel EEG (electroencephalograph) is a highly nonlinear approximation process. We applied a 3-layer, homogeneous spherical current dipole model to simulate the brain potential distribution on the scalp. The dipole source and its corresponding brain potentials were used as the training patterns for the neural networks. Then the neural network was trained to learn the correlation between the dipole and brain potential distribution, that is, the neural network model was used to substitute for the mathematical model. This paper first presents the capability of three neural networks, including the multi- layer perceptrons (MLPs), radial basis function (RBF), and the hybrid (cascade of MLPs and RBF) networks, in localizing the dipole sources in five different regions. In comparison with the simple MLPs and RBF networks, the hybrid network requires less training time and has better localization accuracy. Next, a composite network model is proposed to localize the dipole without any a priori knowledge of its depths and orientation. Our experiment shows that the neural network approach, compared with the conventional numerical approach, is much more efficient in computation. And the localization accuracy is comparable to the numerical approach.

Original languageEnglish
Pages (from-to)105-112
Number of pages8
JournalChinese Journal of Biomedical Engineering
Volume18
Issue number3
StatePublished - 29 Oct 1999

Keywords

  • A single current dipole in a three-shell
  • Artificial neural network
  • Homogeneous sphere
  • Localization of focal source
  • Multi-channel EEG (electroencephalograph)

Fingerprint Dive into the research topics of 'Artificial neural network for dipole localization'. Together they form a unique fingerprint.

Cite this