Hybrid dendritic computing with kernel-LICA applied to Alzheimer's disease detection in MRI

  • Authors:
  • Darya Chyzhyk;Manuel Graña;Alexandre Savio;Josu Maiora

  • Affiliations:
  • Facultad de Informatica, Paseo Mauel Lardizabal 1, 20018 San Sebastian, Spain;Facultad de Informatica, Paseo Mauel Lardizabal 1, 20018 San Sebastian, Spain;Facultad de Informatica, Paseo Mauel Lardizabal 1, 20018 San Sebastian, Spain;Facultad de Informatica, Paseo Mauel Lardizabal 1, 20018 San Sebastian, Spain

  • Venue:
  • Neurocomputing
  • Year:
  • 2012

Quantified Score

Hi-index 0.01

Visualization

Abstract

Dendritic computing has been proved to produce perfect approximation of any data distribution. This result guarantees perfect accuracy training. However, we have found great performance degradation when tested on conventional k-fold cross-validation schemes. In this paper we propose to use Lattice Independent Component Analysis (LICA) and the Kernel transformation of the data as an appropriate feature extraction that improves the generalization of dendritic computing classifiers. We obtain a big increase in classification performance applying with this schema over a database of features extracted from Magnetic Resonance Imaging (MRI) including Alzheimer's disease (AD) patients and control subjects.