Diverse Feature Blend Based on Filter-Bank Common Spatial Pattern and Brain Functional Connectivity for Multiple Motor Imagery Detection

Abstract

Motor imagery (MI) based brain-computer interface (BCI) is a research hotspot and has attracted lots of attention. Within this research topic, multiple MI classification is a challenge due to the difficulties caused by time-varying spatial features across different individuals. To deal with this challenge, we tried to fuse brain functional connectivity (BFC) and one-versus-the-rest filter-bank common spatial pattern (OVR-FBCSP) to improve the robustness of classification. The BFC features were extracted by phase locking value (PLV), representing the brain interregional interactions relevant to the MI, whilst the OVR-FBCSP is used to extract the spatial-frequency features related to the MI. These diverse features were then fed into a multi-kernel relevance vector machine (MK-RVM). The dataset with three motor imagery tasks (left hand MI, right hand MI, and feet MI) was used to assess the proposed method. Experimental results not only showed that the cascade structure of diverse feature fusion and MK-RVM achieved satisfactory classification performance (average accuracy: 83.81%, average kappa: 0.76), but also demonstrated that BFC plays a supplementary role in the MI classification. Moreover, the proposed method has a potential to be integrated into multiple MI online detection owing to the advantage of strong timeefficiency of RVM. INDEX TERMS Multiple motor imagery, filter-bank common spatial pattern (FBCSP), phase locking value (PLV), brain functional connectivity (BFC), multi-kernel relevance vector machine (MK-RVM).

Trial Regeneration With Subband Signals for Motor Imagery Classification in BCI Paradigm

“…The non-stationary property is considered to calculate CSP (NS-CSP) in [18] and bi-spectrum based channel selection is implemented in [25]. Park et al [11] has proposed MI classification algorithms using various decomposition methods such as Butterworth filter (BF), continuous wavelet transformation (CWT), synchrosquizing transformation (SST), EMD, ensemble EMD (EEMD), multivariate EMD (MEMD), and noise assisted MEMD (NA-MEMD).…”

“…[7]. In terms of neurophysiology, motor imagery accompanies attenuation or enhancement of rhythmical synchrony over the sensorimotor cortex with the frequency bands of alpha (8)(9)(10)(11)(12)(13) and beta (14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) [8] - [11]. This paper focuses on EEG based classification of two motor imagery tasks.…”

“…The multi-kernel relevance vector machine (RVM) is utilized for classification. In [18], a selection scheme of CSP features is implemented considering the non-stationarity nature of EEG. The outlier suppression method is also applied to increase the interclass distances.…”

Single-Trial Kernel-Based Functional Connectivity for Enhanced Feature Extraction in Motor-Related Tasks

“…Meanwhile, using features that were extracted from the EEG dynamic brain network analysis is feasible for increasing the classification accuracy [25][26][27]. Nonetheless, dealing with EEG variability because of evoked nonstationary responses remains challenging, particularly in MI-BCI inefficient subjects [28,29]. Some single-trial functional connectivity measures have been introduced from EEG signals, i.e., Cross-Correlation Coefficient (CCF) and Phase Lag Value (PLV) [30], to support motorrelated classification tasks.…”

Discriminating three motor imagery states of the same joint for brain-computer interface

“…In the past few decades, researchers have proposed various feature extraction methods and classification algorithms to classify MI tasks efficiently. The most classical feature extraction methods include wavelet transform (WT) ( You, Chen & Zhang, 2020 ), empirical mode decomposition (EMD) ( Taran et al, 2018 ), common spatial pattern (CSP) ( Yang et al, 2016 ; Selim et al, 2018 ), and filter-bank CSP (FBCSP) ( Ang et al, 2008 ; Wang et al, 2020 ). The widely used classification algorithms include linear discriminant analysis (LDA) ( Aljalal, Djemal & Ibrahim, 2019 ), extreme learning machine (ELM) ( Rodriguez-Bermudez, Bueno-Crespo & Martinez-Albaladejo, 2017 ), k-nearest neighbors (KNN) ( Bashar, Hassan & Bhuiyan, 2015 ), support vector machine (SVM) ( Selim et al, 2018 ) and least squares support vector machine (LS-SVM) ( Taran et al, 2018 ; Taran & Bajaj, 2019 ).…”