Study on the computational cost of EEG dynamic modeling methods

Abstract

The recording of brain activity at the scalp level, also known as electroencephalography (EEG), is a brain imaging technique commonly used in the clinical environment. Adequate modeling of the recorded signals could help to improve the diagnosis of several illnesses such as sleep disorders and epilepsy. This paper presents a computational cost analysis for dynamic modeling methods and considers their suitability to real-time biomedical applications. The analyzed state-of-the-art methods are Dynamic Bayesian Networks (DBN) and Sequential Independent Component Analysis Mixture Modeling (SICAMM). The results show that the ICA-based methods have a lower computational cost than the BN-based methods. The applicability of these methods to patient monitoring using EEG signals is discussed, considering the improvement of the time response by means of parallelization techniques.