Assessment of EEG Brain Dynamics in Time and Frequency Domains through Information-Theoretic Measures

Understanding the functions of the human brain is crucial for preventing and monitoring neurological diseases and developing new neurorehabilitation strategies. In the field of information theory, several measures have been developed for the data-driven analysis of human brain dynamics. For instance, the entropy rate quantifies the information content of a system by taking into account its temporal structure and, under the assumption of stationarity, it can be linked to the well-known concept of conditional entropy which, in turn, quantifies the complexity of a dynamical system in the time domain. Entropy rate can be easily derived directly from the computation of the power spectral density of the analyzed system. However, this method is not widely used in computational neuroscience, although its spectral implementation can disclose the complexity of specific brain rhythms. In this work, the behaviour of entropy rate in the time and frequency domains is illustrated in a theoretical example of six linearly interacting Gaussian processes, demonstrating that the information content of the system is predominantly localized in specific bands of the spectrum. ER measures are then applied to EEG signals recorded from ten healthy subjects performing a motor task, demonstrating how the use of features extracted from the information domain and relevant to the hemisphere contralateral to the movement performed can be used to discriminate between two experimental conditions.