A Matrix Determinant Feature Extraction Approach for Decoding Motor and Mental Imagery EEG in Subject-Specific Tasks

• Published in: IEEE transactions on cognitive and developmental systems Vol. 14; no. 2; pp. 375 - 387
• Main Authors: Sadiq, Muhammad Tariq, Yu, Xiaojun, Yuan, Zhaohui, Aziz, Muhammad Zulkifal, Siuly, Siuly, Ding, Weiping
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.06.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Biomarkers; Biomedical imaging; Brain–computer interface (BCI); Classification; Datasets; Electroencephalography; electroencephalography (EEG); Feature extraction; Graphical user interface; Graphical user interfaces; Machine learning; matrix determinant; mental imagery; motor imagery (MI); multiscale principal component analysis (MSPCA); Neural networks; Physicians; Principal component analysis; Principal components analysis; Real-time systems; Task analysis
• ISSN: 2379-8920; 2379-8939
• DOI: 10.1109/TCDS.2020.3040438

This study introduces a novel matrix determinant feature extraction approach for efficient classification of motor and mental imagery activities from electroencephalography (EEG) signals. First, the multiscale principal component analysis was utilized to obtain clean EEG signals. Second, denoised data were sequentially arranged to form a square matrix of different orders (e.g.,10, 13, 16, and 20) and determinant was computed for each order matrix. Finally, the extracted matrix determinant features were provided to several machine learning and neural network classification models for classification. All experiments were carried out using a 10-fold cross-validation approach on three publicly accessible data sets: 1) data set IV-a; 2) data set IV-b; and 3) data set V of BCI competition III. Also, this study designs a computerized automatic detection of motor and mental imagery graphical user interface that can assist physicians/experts to efficiently analyses motor and mental imagery data. The experimental results reveal that the highest average classification accuracy of 99.55% (for data set IV-a), 99.52% (for data set IV-b), and 91.80% (for data set V) was obtained for motor and mental imagery, respectively, with 20-order matrix determinant using a feedforward neural network classifier. The experimental results suggest that the proposed framework provides a robust biomarker with the least computational complexity for the development of automated brain-computer interfaces.