Automatic Detection of Motor Imagery EEG Signals Using Swarm Decomposition for Robust BCI Systems

Abstract

The development of brain-computer interface (BCI) systems faces a major challenge in achieving a reliable classification of motor imagery-based electroencephalogram signals (MI-EEG). The MI-EEG signals consist of nonstationary multicomponent modulation and complex artifact interference, so accurate decoding and establishing the correlation between MI tasks is always considered the bottleneck in the BCI application. Therefore, the signal decomposition methods can be used to extract MI-EEG-specific rhythms to get the most significant discriminative features for the accurate classification of BCI tasks. However, several weaknesses, including mode mixing, predefined mode number selection, and poor noise suppression, severely limit the use of decomposition for a wide BCI application range. To overcome these issues, a novel swarm decomposition (SWD)-based classification framework has been proposed for improving the classification accuracy of the MI-EEG signals. SWD adopts the swarm iterative filtering to decompose oscillatory components related to the MI-EEG signal accurately. With the BCI Competition IV benchmark dataset, extensive experiments have been conducted on the new hybrid features (HFs) and bidirectional long short-term memory (BiLSTM) classifiers in the proposed framework. In the subject-independent cross-validation scheme, the proposed SWD-HF-BiLSTM framework outperformed all other state-of-the-art approaches utilized for MI-EEG classification with an accuracy of 78.62%. © 2024 selection and editorial matter, Ravichander Janapati, Usha Desai, Shrirang A. Kulkarni, Shubham Tayal.