Weighted inverse model for source localization

Melissa Elizabeth Acosta Muñoz and Hugo Alberto Paredes Argoty, Universidad de Nariño, San Juan de Pasto - Colombia, 2015

Epilepsy is a neurological disorder mainly consisting of imbalances of neural activity in certain brain regions. After cerebrovascular accidents (also known as strokes) and dementia, epilepsy has become the most common neurological disorder – it is believed that it affects 0.5-1.5% of the world population. Mostly,
this disorder affects children under age 10 and adults over 65, being more frequent in developing countries and less favored classes. The localization of epileptic sources can be done in a non-invasive way through the analysis of the measurements of electric potential on the scalp- electroencephalographic signals (EEG). Specifically, using mathematical models, EEG signals can be mapped onto geometrical coordinates revealing the epileptic sources localization. In this sense, one of the most widely used and recommended model by the scientific community is the so-called inverse problem model. Some alternatives and variants have been proposed, being weighted approaches a recent and suitable alternative.

In this work, we perform an exploratory study of weighted inverse models aimed at identifying the benefit of incorporating weighting factors effect into the solution of the inverse model problem. To this end, we first implement a versatile formulation of the inverse problem model followed from a weighted approach. Particularly, we use the LORETA method is. On this model, we evaluate different weighting approaches for scaling each signal EEG coming from every channel.  Such scaling is done by the estimated relevance of the EEG channels, which is based on variance, and energy criteria. Experiments are carried out over a standard simulated human EEG, which is based on a head model comprises 4001 dipoles placed only on the tessellated cortex surface to generate the lead field matrix. The performance of considered weighting approaches is assessed in terms of Earth Mover’s Distance. As a result, a proper methodology to compare epileptic source localization is presented as well as key aspects to select a proper method are widely discussed.

Figure 1. Source localization results for the three considered approaches. All the methods are tested regarding the same simulated source (blue point).

Figure 2. Boxplots of studied weighting approaches regarding the EMD values at 20 iterations.



        See full thesis

  • On the effect of inverse problem weighted solutions for epileptic sources localization.

          The Twentieth Symposium on Signal Processing, Images and Computer Vision, STSIVA-2015 
          H. A. Paredes-Argoty, M. E. Acosta-Muñoz, E. J. Revelo-Fuelagán, D. H. Peluffo-Ordóñez,

          See abstract.        See full paper.

  • Exploratory study on epileptic sources localization via weighting inverse problema approaches.

          See abstract.        See full paper.


Figure 3. Source Localization

Source Localization