Implicit decoded neurofeedback of spatial and feature-based visual selective attention

Full description

In this project we sought to assess the cognitive specificity of an implicit decoded neurofeedback protocol for training spatial and feature-based visual selective attention. If participants can learn to regulate only the cognitive function most strongly associated with decoder-based feedback, we expect that feedback designed to enhance one mode of attention (e.g., spatial) should not influence the other (feature). By contrast, if feedback is also integrated more broadly toward other cognitive functions, then neurofeedback aiming to enhance one mode of attention should also improve the other. To assess these possibilities, we developed a four-day EEG-based training protocol to enhance the patterns of neural activity associated with visual selective attention in three separate groups of participants (spatial neurofeedback group, feature-based neurofeedback group, sham control group, Total N = 108). Neurofeedback was delivered through dynamic changes in the difficulty of an attentional cueing task. Participants were tasked with identifying bursts of coherent motion in a frequency-tagged display containing fields of black and white moving dots while we recorded their brain activity using EEG. Attentional cues on each trial indicated that participants should monitor a subset of the dots, defined by the intersection of a cued colour and spatial location, for brief bursts of coherent motion. To generate neurofeedback, frequency-tagged EEG data were passed in real-time to machine learning classifiers trained to classify either the attended colour (feature neurofeedback group), or spatial configuration (space neurofeedback group, Figure 1b). Neurofeedback representing the output of the classifier was presented by modulating the contrast of phase-scrambled noise masks presented behind the dot fields, where increased contrast made target bursts of motion direction more difficult to discriminate . Critically the contrast of the noise masks varied over time but was always matched across locations. Thus, neurofeedback training reinforced endogenous neural states associated with strong attentional selectivity, rather than exogenously eliciting those states through changes in salience across targets (i.e. top-down rather than bottom-up attention). To assess the efficacy and specificity of this neurofeedback training, participants performed an attention test before and after neurofeedback training. This task was similar in form to the training task but did not contain any phase-scrambled noise masks. Further, while we used combined spatial/feature-based cues during neurofeedback, participants were cued to independently engage either spatial or feature-based attention during the attention test. A third “sham neurofeedback” group of participants each completed the same protocol but were shown the feedback which had previously been generated for a different active neurofeedback participant. As such, all three neurofeedback groups viewed identical displays and received identical instructions. Participants were blind to the existence of separate neurofeedback groups and to the intended purpose of neurofeedback training. Thus, any differences between groups could only be attributed to the nature of the link between task-difficulty and attentional state. The dataset contains data from three groups of subjects (Feature-based attnetion training, spatial attention training, sham training). Each group's data are in their own folder ('TrainFeature', 'TrainSpace', 'TrainSham'). Each folder contains BIDS formatted data for each subject in the experiment. These folders contain EEG and behavioural data for each of the four experimental test days.