Decoding covert visual attention to motion direction using graph theory features of EEG signals and quadratic discriminant analysis

Visual attention is a type of selective attention that plays an important role in prioritizing and processing the information received from the visual scenes around us. The brain is an intricate network composed of numerous regions, each with distinct functions. As different tasks are performed, brain regions regularly synchronize and correlate with each other through intricate networks of connections. The aim of this study is to decode two states of attention by examining the interactions and connections between different brain regions using graph theory. This is demonstrated by EEG recordings from 15 participants who performed visual attention task. Pearson's correlation coefficient and coherence have been used to measure the functional connections between brain regions. In fact, each of these two criteria is regarded as an individual feature, and we perform decoding using each criterion separately. With an optimal selection of 40 connectivity features, the QDA classifier attained accuracies of 79.83% and 83.28% using correlation and coherence features, respectively. The results of attention decoding using the coherence criterion are more promising, indicating the superior effectiveness of coherence-based methods. Therefore, this study employed graph theory to analyse a neural network derived from coherence measurements. The study focused on three graph-theoretical metrics: degree centrality, efficiency, and betweenness centrality. The QDA classifier, using an optimal set of 40 features that includes degree, betweenness, and channel efficiency, achieved an accuracy of 86.46%. In comparison, the QDA classifier with 40 features based solely on degree centrality reached an accuracy of 89.96%. Finally, the results of this research indicate that analysing brain connections and brain network graphs can effectively decode different covert visual attention states.