EEG Data Mining Using PCA

EEG Data Mining Using PCA

Lenka Lhotská (Czech Technical University in Prague, Czech Republic), Vladimír Krajca (Faculty Hospital Na Bulovce, Czech Republic), Jitka Mohylová (Technical University Ostrava, Czech Republic), Svojmil Petránek (Faculty Hospital Na Bulovce, Czech Republic) and Václav Gerla (Czech Technical University in Prague, Czech Republic)
DOI: 10.4018/978-1-60566-218-3.ch008
OnDemand PDF Download:


This chapter deals with the application of principal components analysis (PCA) to the field of data mining in electroencephalogram (EEG) processing. The principal components are estimated from the signal by eigen decomposition of the covariance estimate of the input. Alternatively, they can be estimated by a neural network (NN) configured for extracting the first principal components. Instead of performing computationally complex operations for eigenvector estimation, the neural network can be trained to produce ordered first principal components. Possible applications include separation of different signal components for feature extraction in the field of EEG signal processing, adaptive segmentation, epileptic spike detection, and long-term EEG monitoring evaluation of patients in a coma.
Chapter Preview


An electroencephalogram (EEG) is a recording of spontaneous brain electrical activity by means of electrodes located on the scalp. The placing of the electrodes is constrained by natural physical limits, namely by the size of the electrodes, which limits the maximum number of electrodes that can be used. Another limitation is the mutual influence of electrodes located close to each other. Standardized placement of the basic number of electrodes is done in accordance with the scheme designed by Dr. Jasper (Jasper, 1958). This is nowadays known as the International 10-20 system.

In the frequency domain we can distinguish four basic frequency bands on an EEG signal, namely delta, theta, alpha, and beta activities.

The delta band corresponds to the slowest waves in the range of 0-4 Hz. Its appearance is always pathological in an adult in the waking state. The pathological significance increases with increasing amplitude and localization. The existence of a delta wave is normal for children up to three years of age, in deep sleep and hypnosis. During sleep the waves can be higher than 100 µV in amplitude.

The theta band corresponds to waves in the range of 4-8 Hz. Their existence is considered as pathological if their amplitude is at least twice as high as the alpha activity or higher than 30 µV if alpha activity is absent. The presence of a theta wave is normal if its amplitude is up to 15 µV and if the waves appear symmetrically. In healthy persons they appear in central, temporal and parietal parts. This activity is characteristic for certain periods of sleep.

The alpha band corresponds to waves in the range of 8-13 Hz. In the waking state in mental and physical rest the maximum appears in the occipital part of the brain. Its presence is highly influenced by open or closed eyes. The amplitude is in the range of 20-100 µV, most frequently around 50 µV.

The beta band corresponds to the fastest waves in the range of 13-20 Hz. The maximum of the activity is mostly localized in the frontal part, and it decreases in the backward direction. The rhythm is mostly symmetrical or nearly symmetrical in the central part. The amplitude is up to 30 µV. The activity is characteristic for concentration, logical reasoning and feelings of anger and anxiety.

An EEG contains a great deal of information about the state of a patient’s health. It has the advantage of being non-invasive and applicable over a comparatively long time span (up to 24 hours, if necessary). This is an important feature in cases where we want to follow disorders that are not permanently present but appear incidentally (e.g., epileptic seizures) or under certain conditions (various sleep disorders) (Daube, 2002).

Complete Chapter List

Search this Book: