A fully unsupervised clustering on adaptively segmented long-term EEG data

Abstract

Several new procedures for efficient processing of long-term EEG recordings has been validated on long-term comatose EEG data. The proposed solution is based on the use of cluster analysis over a set of various features derived from adaptively segmented EEG data. Special attention was given to utilization of clinically relevant information from multichannel EEG data. Methods for validation of the cluster analysis results were implemented and tested. Suggested algorithms speeds up a subsequent evaluation of the data and simplify a tedious and time-consuming work of neurologists or sleep technicians, making the evaluation more objective, and represent results in an understandable form.