Dynamical Feature Extraction from Brain Activity Time Series

Abstract

Neurologists typically study the brain activity through acquired biomarker signals such as Electroencephalograms (EEGs) which have been widely used to capture the interactions between neurons or groups of neurons. Detecting and identifying the abnormal patterns through visual inspection of EEG signals are extremely challenging and require constant attention for well trained and experienced specialists. To resolve this problem, data mining techniques have been successfully applied to the analysis for EEG recordings and other biomarker data sets. For example, Chaovalitwongse et al., (2006; 2007), Prokopyev et al., (2007) and Pardalos et al., (2007) reported the EEG patterns can be classified through dynamical features extracted from the underlying EEG dynamical characteristics. Moreover, in the area of cancer research, Busygin et al., (2006) showed promising results via Bi-clustering data classification technique using selected features from DNA microarrays. Ceci et al., (2007); Krishnamoorthy et al., (2007) also reported that data mining techniques enable protein structure characterization and protein structure prediction. From data mining aspects, feature extraction and selection for time series data sets not only play an important role in data preprocessing but also provide opportunities to uncover the underlying mechanisms of data under study. It also keeps the essential data structure and makes a better representation of acquired data sets that need to be classified. In this work, the properties and descriptions of the most common neurological biomarker namely EEG data sets will be given as well as the motivations and challenges for abnormal EEG classification. The dynamical features for EEG classification will be reviewed and described in the second part of this work. The described dynamical features can also be applied to other types of classification applications for discovering the useful knowledge from obtained data sets. Finally, the potential applications for EEG classification will be discussed and comments for further research directions will be given in the future trends and conclusion sections.