Small-Dimension Feature Matrix Construction Method for Decoding Repetitive Finger Movements From Electroencephalogram Signals

• Published in: IEEE access Vol. 8; pp. 56060 - 56071
• Main Authors: Jia, Tianyu, Liu, Ke, Lu, Yijia, Liu, Yali, Li, Chong, Ji, Linhong, Qian, Chao
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Brain computer interfaces; Classification; Classification algorithms; Clustering; Decoding; Electrodes; Electroencephalography; event-related desynchronization topography; Feature extraction; finger movement; Frequencies; Graph theory; Hemispheres; Human-computer interface; Indexes; Support vector machines
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2020.2982210

Brain computer interface (BCI) has been widely studied to allow people to control external devices as an extension of capabilities or a replacement of lost functions. The decoding algorithm of brain signals is a crucial part in BCI, since its performance determines the efficiency of the interface. Decoding performance can be improved by generating optimal feature matrix. The objective of this paper is to propose and implement a decoding algorithm with optimized small dimension feature matrix on identifying motor intention of finger movement using electroencephalogram (EEG) signals. An experiment was designed and conducted, in which EEG was acquired from 10 healthy subjects during the left or the right index finger movement. Event-related desynchronization (ERD) topography was analyzed during motor tasks. A degree feature extraction algorithm was proposed based on the graph theory together with Support Vector Machine (SVM) to classify two kinds of index finger movement, which takes three factors into consideration: frequency bands, amplitude and range of ERD. The results showed that the algorithm can classify the finger movement effectively for 7 subjects based on a three-dimension optimized feature matrix, consisting of the maximum degree, average degree and clustering range. The proposed algorithm is not limited by the size of samples and can indicate the source area of the neural activities. Results also demonstrate that the proposed degree feature extraction algorithm can smooth signal noise and enlarge the feature differences between the contralateral and the ipsilateral hemispheres.