Date/heure
Date(s) - 13 décembre 2012

Catégories Pas de Catégories

$Fully unsupervised detection and clustering of EEG epileptic spikes By Johann Marchi, INS. Nowadays, large amounts of electroencephalogram (EEG) data remain unexploited because of a lack of unsupervised data mining programs. Indeed, the visual analysis of EEG signals is a time-consuming task for the physician, and some variability may appear across expert advice, hence the need of automated and objective algorithms. We will present a project under progress. This work aims at providing a method to detect and classify abrupt changes (epileptic spikes) in interictal EEG signals recorded on epileptic patients. The method used for the detection of spikes is inspired from the local false discovery rate (Efron 2001) which is based on very few modeling assumptions. Various methods are studied to perform efficient clustering with an adaptive number of classes. To the best of our knowledge, this is the first fully automated method for the detection and classification of epileptic EEG spikes, as opposed to most of the already available algorithms requiring either the intervention of an expert to form learning sets or a tedious adjustment of complex parameters.$

Date/heure
Date(s) - 13 décembre 2012

Catégories Pas de Catégories

$Fully unsupervised detection and clustering of EEG epileptic spikes\nBy Johann Marchi, INS.\n\nNowadays, large amounts of electroencephalogram (EEG) data remain unexploited because of a lack of unsupervised data mining programs. Indeed, the visual analysis of EEG signals is a time-consuming task for the physician, and some variability may appear across expert advice, hence the need of automated and objective algorithms.\n\nWe will present a project under progress. This work aims at providing a method to detect and classify abrupt changes (epileptic spikes) in interictal EEG signals recorded on epileptic patients. The method used for the detection of spikes is inspired from the local false discovery rate (Efron 2001) which is based on very few modeling assumptions. Various methods are studied to perform efficient clustering with an adaptive number of classes.\n\nTo the best of our knowledge, this is the first fully automated method for the detection and classification of epileptic EEG spikes, as opposed to most of the already available algorithms requiring either the intervention of an expert to form learning sets or a tedious adjustment of complex parameters.[$