Combining EEG Oscillation Analysis and Explainable Artificial Intelligence for Characterizing Visuospatial Attention

Covert visuospatial attention is the ability of human brain to voluntarily direct the attentional focus in the visual space without moving eyes. EEG studies support the involvement of alpha rhythm, mainly acting on posterior regions, in mediating visuospatial attention mechanisms. However, most of the studies considered a small set of regions, focusing on only one or just a few posterior regions, selected a priori. Thus, it remains unclear how other parietal-occipital regions contribute to visuospatial attention. Here, we collected EEG signals during a covert attention-orienting task, requiring to orient attention either to left or right hemifield. We examined EEG-source level signals, performing a whole cortex analysis. Two different approaches were employed to evidence the parietal occipital regions that have a key role in attention-related alpha mechanisms. The first is based on more traditional analyses, evaluating modulation of both alpha power in cortical regions and of alpha-band interregional cortical connectivity, depending on the attended hemifield. The second is based on explainable artificial intelligence (XAI), combining a convolutional neural network and an explanation technique, to reveal the regions more relevant for discriminating left vs. right attention orienting. Both approaches point to superior parietal cortex as a key node in both hemispheres for visual spatial attention, concurrently with other occipital regions (e.g., lateral occipital cortex). This study, although preliminary, shows the potentiality of combining a XAI approach with more traditional methods for increasing the robustness of EEG-source analysis. The same combined approach can be easily transposed to investigate other cognitive tasks.