A Kullback Leibler-nearest neighbor rule classification of EEGs: The EEG population screening problem, an anesthesia level EEG classification application
✍ Scribed by Will Gersch; F. Martinelli; J. Yonemoto; M.D. Low; J.A. McEwen
- Publisher
- Elsevier Science
- Year
- 1980
- Tongue
- English
- Weight
- 940 KB
- Volume
- 13
- Category
- Article
- ISSN
- 0010-4809
No coin nor oath required. For personal study only.
✦ Synopsis
A new methodology, a Kullback Leibler-nearest neighbor (KL-NIV) rule method is introduced for the EEG population screening problem. It is applied to the classification of anesthesia levels of humans in surgery by the analysis of EEGs alone. Stationary epoch multichannel EEGs are considered. In the EEG population screening problem, the category or state of an individual is classified by comparison of his/her EEG with labeled sample EEGs taken from other individuals. In the KL-NN method, a measure of dissimilarity is computed between a new to-be-classified EEG and each of the labeled sample EEGs. The measure of dissimilarity is a Kullback Leibler measure of the dissimilarity between the sample covariance functions of the EEG time series, computed as if the EEGs were Gaussian distributed. The new EEG is classified to have the same categorical label of its nearest neighbor EEG, (or of a majority of its nearest neighbor EEGs) in the minimum KL number sense. The KL-NN rule has the important statistical property that, with only a relatively small number of categorically labeled sample EEGs, the KL-NN rules yield a statistically reliable estimate of the best achievable discriminability between categorical EEG populations. In an application of the KL-NN rule method, the level of anesthesia insufficient for deep surgery was distinguished from the anesthesia level sufficient for deep surgery on humans in surgery under halothane-nitrous oxide anesthesia. Analysis was carried out on 2@ set EEG epochs from 18 different individuals. The results were that the probabilities of correct classification of anesthesia levels, based on the analysis of two and four EEG data channels, were 85 and 89% respectively. Ninty-seven percent of 62 independent twochannel EEGs, from the same 18-individual population, were classified correctly against the labeled sample EEG data.