Automatic detection of alertness level from electroencephalogram signals and cortical auditory evoked potential responses

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description Size
Thumbnail01front.pdf144.07 kB
Adobe PDF
Thumbnail02whole.pdf5.58 MB
Adobe PDF
This research aims to identify the degree of alertness of subjects that undergo the Cortical Auditory Evoked Potential (CAEP) based hearing test. One of the important factors that influence this is the alertness state of subjects. Research has shown that for this test to be useful, subjects need to stay at a constant state of engagement. Accordingly, this thesis focuses on developing a system that will be able to classify each portion of the recorded signal into one of four states; engaged, calm, drowsy and asleep. In order to achieve this, we studied the relationship between CAEP responses and the alertness states, and we validated the existence of this relationship. We have also developed a method to search for the best channel/rhythm combination for each alertness state. In the first study, two sets of features were considered to represent the recorded data. The first set was based on the wavelet transform of the background EEG, while the second set was obtained from the peaks of the CAEP responses. Obtained results suggest that the CAEP-based features were very comparable, in terms of classification accuracy, to the well-established wavelet-based features of EEG signals (79% compared to 80%). In the second study, the EEG rhythms of subjects were analysed. Investigation of the importance of the different EEG rhythms in terms of their capabilities in differentiating between the different alertness states was conducted. This is followed by considering subsets that contain 2, 3, 4 as well as all 5 EEG rhythms. Finally, a feature subset selection method based on differential evolution 9DE) that has been proposed particularly to deal with multi-channel signals is used to search for the best subset of EEG rhythms for the various channels. It was shown that higher frequency EEG rhythms (𝛾, 𝛽) are better classifiers for the subject’s alertness state than 𝛼, 𝜃, and 𝛿 (lower frequency EEG rhythms). Optimal combinations of different EEG rhythms have been described. The proposed differential evolution feature selection algorithm is shown to produce better results than the ranking and sequential forward selection approaches. Obtained results suggest that the best subsets are formed using combinations of channels and features that are influenced by high frequency rhythms.
Please use this identifier to cite or link to this item: