Advanced Pipelines For Artifact Removal From EEG Data

Feature extraction and working with EEG data has become one the most challenging studies these years. The raw EEG signal has various artifacts and needs to be detected and separated from brain components. This study is part of ERC. For removing artifacts from EEG data , this procedure done by a method known as “semi-automatic ICs selection pipeline”.This method was developed and verified by Cosynclab directed by Prof. Betti in Rome (where I spent my internship). In particular, the thesis work aims to investigate another method for complementing semi-automatic ICs selection pipeline and evaluate results which conveys to increasing the accuracy of semi-automatic ICs selection pipeline.The ICA algorithm derives independent sources from highly correlated EEG signals statistically without concern for the actual location or configuration of the EEG signal source . It is used to locate concurrent signal sources that are either too close together or too broadly scattered to be separated using conventional localization techniques. The primary issue in understanding ICA output is determining the right dimension of the input channels and the physiological and/or psychophysiological relevance of the resulting ICA source channels.With semi-automatic ICs selection pipeline method more than 2600 ICs evaluated and 405 ICs labeled as brains and the rest classified as artifacts. To evaluate these 405 ICs and increase possible accuracy another method was used known as ICLabel. ICLabel projects had been proposed by EEGLAB. This is a method based on Deep Learning and provides classification based on EEG IC classifier1 . After running and comparing the two methods pipeline, then,we designed an application for comparison and visualization output for both methods which name is IC selection.With this application we realize some modification needed for future steps for labeling with semi-automatic ICs selection pipeline method and some artifacts could change from artifacts to brain.