Features Domains and Classification Algorithms in Motor Imagery Brain Computer Interface

Abstract

Motor imagery brain-computer interface (MI-BCI) is well approved to help people with movement impairments due to neural disorders. The procedure for MI-BCI to translate MI brain signals to understandable instructions for external devices involves extracting features from recorded signals and predicting desired movements. This paper summarizes and compares feature extraction methods and classification algorithms and their modifications that are commonly used for MI EEG signals. Feature extraction techniques are discussed based on their feature domains: time, frequency, and spatial. Classification algorithms are divided into classical machine learning and deep learning. This paper aims to provide a straightforward view of common ways to extract features and classifying movements in MI-BCI and show difficulties in MI-BCI signal processing. The nature of MI EEG signals and MI-BCI applications points to several promising field such as transfer learning and deep learning neural networks.