An Ensemble Learning Based Classification of Individual Finger Movement from EEG

Abstract

Brain computer interface based assistive technology are currently promoted for motor rehabilitation of the neuromuscular ailed individuals. Recent studies indicate high potential of utilising electroencephalography (EEG) to extract motor related intentions. Limbic movement intentions are already exhaustively studied by the researchers with high accuracy rate. But, capturing movement of fingers from EEG is still in nascent stage. In this study, we have proposed an ensemble learning based approach for EEG in distinguishing between movements of different fingers, namely, thumb, index, and middle. Six healthy subjects participated in this study. Common spatial patterns (CSP) were extracted as features to classify with the extra tree or extremely randomized tree binary classifier. The average classification accuracy of decoding a finger from rest condition was found to be 74%, wheres in discriminating of movement of pair of fingers average accuracy was 60%. Furthermore, error correcting output coding (ECOC) was added to the binary classifier to use it in multiclass classification. The proposed algorithm achieved a maximum kappa value of 0.36 among the subjects.

Individual finger movement decoding using a novel ultra-high-density electroencephalography-based brain-computer interface system

“…Peak accuracy ranged from 85.4(S1), 79.2(S2), 69.4(S3), 71.5(S4) to 66.2(S5). Mean accuracies of 70.7(S1), 65.3(S2), 63.7(S3), 63.7(S4), and 60.6(S5) are in the range of other motor movement studies without feedback (Quandt et al, 2012;Bera et al, 2019). The difference is that now we are classifying single fingers, not left/right motor imagery.…”

“…The difference is that now we are classifying single fingers, not left/right motor imagery. As stated in section “1,” the possibility of distinguishing between finger movements was sufficiently shown by ECoG studies ( Kubánek et al, 2009 ; Hotson et al, 2016 ; Gruenwald et al, 2019 ) but few for EEG with limited performance ( Liao et al, 2014 ; Stankevich et al, 2016 ; Bera et al, 2019 ). Essential to push the classification accuracy much further is to include feedback which will be done in the subsequent study.…”

“…The obtained classification accuracies [64.8 (6.3)% on average] can only be indirectly compared to Bera et al (2019) and Liao et al (2014) . While as they also performed pairwise classification of finger movements, they used different features and classification models.…”

“…While as they also performed pairwise classification of finger movements, they used different features and classification models. Specifically, Bera et al (2019) conducted finger extension tasks including thumb, index, and middle fingers but used CSPs, LDA and Extra Trees Classifiers for classification, achieving an overall average accuracy of 60.3% with two preceding training days. Liao et al (2014) conducted two repetitive finger flexions and extensions tasks during 2 s. They used a SVM with a radial basis kernel function and achieved an average classification accuracy of 58.6 and 57.7% using mu and beta band power features, respectively.…”

“…Decoding more precise movements, such as individual finger movements, is highly challenging for conventional EEG systems. Quandt et al (2012) , Stankevich et al (2016) , and Bera et al (2019) decoded individual finger movements using frequency bands under 40 Hz. They resulted in an average classification accuracy of 60, 50, and 43%, where the chance level was 33, 25, and 25%, respectively.…”