Conditional Entropy Estimates for Distress Detection with EEG Signals

摘要

Recently, distress has become a major problem in most advanced societies because of its negative side effects in physical and mental health. In this sense, the assessment of different physiological signals such as electroencephalogram (EEG) provides new insights about the body's reaction against distressful stimuli. Moreover, the non-linear and dynamic behaviour of the brain suggests the application of nonlinear methodologies for EEG analysis. In this work, a symbolic technique called conditional entropy was applied for the assessment of 279 32-EEG channel segments of calm and distress emotional states. Results of all EEG electrodes were combined in a simple decision tree classifier, reporting a discriminatory power above 70%. Furthermore, a decreasing tendency of irregularity when changing from calm to distress conditions was observed for all EEG channels. The simplicity of this classification model allows an easy interpretation of the results, together with a possible implementation of the algorithm in a real-time monitoring system.