Evaluation of less Common Independent Component Analysis Algorithms for Brain Computer Interface Preprocessing

owadays, one of the most challenging issues in Brain-Computer Interface (BCI) systems is tackling with physiological artifacts like Electrooculography (EOG) and Electrooculography (EMG) as a preprocessing step. It is the first step of each BCI systems that is very substantial because for next steps like feature extraction and classification we need clean signals without undesirable artifacts. Using a linear filter to remove these artifacts is common due to their simplicity and acceptable results in recent BCI preprocessing papers especially among winners in BCI VI competition using the same dataset as this paper (Graz 2A). By means of this, we have decided to compare the performance of band-pass filter with thirty well-known Independent Component Analysis (ICA) algorithms to remove undesirable EOG artifacts from EEG signals. The most common ICA algorithms that have been applied on this dataset are FastICA, SOBI and Infomax, but we tend to try less common algorithms to evaluate the real performance of ICA. In the meantime, we choose three optimized algorithm of ICA (FPICA, SANG and SYM-WHITE) that work better in noise reduction with Graz 2A dataset and compare their results with simple 8 to 40 Hz band-pass filter on 23-24-25 channels for EOG signals. The performance of our three algorithms was measured using the signal to interference (SIR) index. Results indicate that among all tested algorithms, Symmetric Pre-Whitening (SYM_WHITE) algorithm has considerable efficiency for EOG reduction and remarkable high speed in runtime. It is worth noting that it is for the first time that this algorithm is applied on Graz 2A dataset and approximately eliminates all EOG artifacts without destroying the original EEG signals.

کلیدواژه ها:

Brain computer interface ، Independent Component Analysis ، ICA ، EOG artifact ، linear filter ، SYM-WHITE