Automated Ocular Artifacts Identification and Removal from EEG Data Using Hybrid Machine Learning Methods

Abstract

Brain Computer Interface (BCI) is an emerging field of research, which helps neuromuscular disabled people to communicate and control external devices with the brain. Electroencephalogram (EEG) signals have been used by doctors to diagnose brain activities. EEG signals are very sensitive to the physical movement of most of the body parts. Electrooculogram (EOG) are the signals which are directly related to eye movements. In the field of BCI, artifacts refer to those signals which are not required but will be present in data because of the involuntary actions of human muscles. In the proposed research, BCI Competition III Dataset IV A is used, which consists of two classes of signals right hand and right foot, so ocular artifacts were a prime hindrance for this type of data. The research involves the use of the following feature detection methods: ICA, ICA-RLS, ICA-LMS, ICA-DWT, EMD-ICA. The first common step is using ICA to obtain independent components(ICs). Using ICA, ICs that contain ocular artifacts can be identified and then further exposed to another round of ICA or adaptive filtering that uses LMS, RLS. The adaptive filtering technique reduces the need for parallel EOG recordings by removing the ocular ICs using reference signals. Hybrid' methods have gained much advantage over the traditional ICA process, which leads to better detection and removal of ocular artifacts with these techniques. Further, the research proposes a system involving the use of SVM-PSO for the classification of motor imagery signals obtained after the removal of artifacts. ICA-RLS and CSP as pre-processing and feature extraction techniques provided the best classification accuracy of 92.94%.