Análisis de Estimulacion Electrica Funcional en Señales de EEG


Javier Mínguez Zafra


Electroencephalographic (EEG) recordings usually contain a lot of artifacts besides the signal
representing the neural activity of the brain. Especially when using the EEG signal as an input for brain-computer interfaces (BCI) the removal of those artifacts is crucial.
This project had a focus on nding methods in order to remove artifacts generated by
functional electrical stimulation (FES) from EEG recordings. Information gathered during this
project will serve for a research project with the objective of controlling a robotic prothesis by
a BCI using FES at the University of Zaragoza.
This project basically consisted of four parts. The rst step was a literature research regarding
the following topics: EEG ([1]) and artifacts ([2]), BCI basics and Blind Source Separation (BSS)
such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA).In
the second part frequency analysis of several EEG recordings containing FES artifacts were
performed. Several BSS methods were applied in order to extract and remove the FES artifacts from the EEG signals. Neither PCA nor ICA achieved satisfying results, so new methods had to be searched. In the third part of the project new ICA algorithms as well as their combination with wavelet transformation ([3]) were investigated. Every method was applied for three di erent EEG datasets without obtaining any satisfying results. Due to the fact that no satisfying algorithm was found to remove the FES artifact from the EEG recording the practical part of the project consisted of doing experiments with brain-computer interfaces at Bit&Brain Technologies, a spin-o company of the University of Zaragoza. In order to do BCI experiments the electrodes had to be placed on the scalp of the subject as well as the preparation of the EEG acquisition equipment had to be done for monitoring the EEG signal in real time. Further I took part as one of the subjects in an experiment concerning emotion recognition. New ICA algorithms and wavelet transforms, as well as ICA algorithms enhanced with wavelet transforms ([3]) were applied at this part of the project in order to achieve the main