Enhancements of EEG Artifact Automatic Correction Algorithm and Processing of Hemiplegic Patients' EEG

Patients suffering from severe brain injuries may experience the unfortunate condition of being conscious, but unable to interact or communicate with the outside world. In these cases, the lack of communication does not allow doctors to have a complete understanding of patients' state of consciousness leading to misdiagnosis compromising the choice of the best treatment. The use of Brain-Computer Interface (BCI) is a promising strategy to analyse brain signals and evaluate the presence of consciousness. The proposed project aims at realizing a low-cost non-invasive BCI able to detect and distinguish consciousness states through the use of electroencephalographic (EEG) and electromyographic (EMG) signals. We initially recorded from healthy subjects to train our algorithm. The method involves a pre-processing part, entirely realized on MATLAB, to clean the raw EEG signals and obtain Readiness Potentials (RP), a special type of Event Related Potentials (ERP), giving a measure of the intentionality behind motion. Machine learning algorithms are successively devoted to the classification of the movement between voluntary, semi-voluntary and involuntary with the extraction of peculiar RP features. After an introduction on consciousness and its states, the brain anatomy and the recording methods, this thesis will focus on enhancing the pre-processing algorithm devoted to the automatic correction of EEG artifacts. In particular, it will be discussed (1) the different artifacts affecting the EEG signal; (2) the use of Independent Component Analysis (ICA) to detect and correct them in frequency and time domain; (3) the analysis of different detrending method to remove slow drifts; (3) the improvements obtained on the RPs. We recorded EEG and EMG also from hemiplegic patients asking them to perform or imagine to perform a motor task. The application of the pre-precessing chain to these signals will be discussed to prove the presence of intention and motor programming exploiting Lateralized Readiness Potentials (LRP).