SSVEP recognition by modeling brain activity using system identification based on Box-Jenkins model

• Published in: Computers in biology and medicine Vol. 101; pp. 82 - 89
• Main Authors: Safi, Seyed Mohammad Mehdi, Pooyan, Mohammad, Motie Nasrabadi, Ali
• Format: Journal Article
• Language: English
• Published: United States Elsevier Ltd 01.10.2018; Elsevier Limited
• Subjects: Accuracy; Autoregressive moving average; Autoregressive processes; Box-jenkins model (BJM); Brain; Brain-computer interface (BCI); Canonical correlation analysis (CCA); Computer applications; Correlation analysis; EEG; Electroencephalography; Fourier transforms; Human-computer interface; Implants; Least absolute shrinkage and selection operator (LASSO); Methods; Modelling; Multivariate linear regression (MLR); Noise; Pattern recognition; Recognition; Regression analysis; Signal processing; Statistical analysis; Steady-state visual evoked potential (SSVEP); Stimulation; System identification; Visual evoked potentials; Visual signals; Visual stimuli; Wavelet transforms; Brain-computer interface (BCI); Steady-state visual evoked potential (SSVEP); Multivariate linear regression (MLR); Least absolute shrinkage and selection operator (LASSO); Box-jenkins model (BJM); Canonical correlation analysis (CCA)
• ISSN: 0010-4825; 1879-0534; 1879-0534
• DOI: 10.1016/j.compbiomed.2018.08.011

The steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI) has received increasing attention in recent years. The present study proposes a new method for recognition based on system identification. The method relies on modeling the electroencephalogram (EEG) signals using the Box-Jenkins model. In this approach, the recorded EEG signal is considered as a combination of an SSVEP signal evoked by periodic visual stimulation and a background EEG signal whose components are modeled by a moving average (MA) process and an auto-regressive moving average (ARMA) process, respectively. Then, the target frequency is determined by comparing the modeled SSVEP signals for all stimulation frequencies. The experimental results of the proposed method for recorded EEG signals from five subjects (each subject with four stimulation frequencies) demonstrated a significant improvement in the accuracy of the SSVEP recognition in contrast to canonical correlation analysis, least absolute shrinkage and selection operator, and multivariate linear regression methods. The proposed method exhibits enhanced accuracy especially for short data length and a small number of channels. This superiority suggests that the proposed method is an appropriate choice for the implementation of real-time SSVEP based BCI systems.