DECODING VISUAL COVERT SELECTIVE SPATIAL ATTENTION BASED ON MAGNETOENCEPHALOGRAPHY SIGNALS

Abstract

This paper proposes a hybrid approach for inferring the target of visual covert selective spatial attention (VCSSA) from magnetoencephalography (MEG) signals. The MEG signal offers a higher spatial resolution and a lower distortion as compared with their competing brain signaling techniques, such as the electroencephalography signal. The proposed approach consists of removing global redundant patterns of MEG channels by surface Laplacian, feature extraction by Hurst exponent (H), 6th order Morlet coefficients (MCs), and Petrosian fractal dimension (PFD), standardization, feature ranking by statistical analysis, and classification by support vector machines (SVM). The results indicate that the combined use of the above elements can effectively decipher the cognitive process of VCSSA. In particular, using four-fold cross-validation, the proposed approach robustly predicts the location of the attended stimulus with an accuracy of up to 92.41% for distinguishing left from right. The results show that the fusion among wavelet coefficients and non-linear features is more robust than in other previous studies. The results also indicate that the VCSSA involves widespread functional brain activities, affecting more regions than temporal and parietal circuits. Finally, the comparison of the results with six other competing strategies indicates that a slightly higher average accuracy is obtained by the proposed approach on the same data.