Abstract
When sustained for long periods of time, goal-directed attention eventually succumbs to distraction. We hypothesized that such lapses partly result from an inability to effectively monitor one’s own internal state to detect impending failures — and that, if such states could be made more accessible, observers may learn to better predict and prevent lapses before they occur. Here we test this idea by externalizing attentional states with real-time fMRI during a sustained attention task. Observers viewed blocks of face/scene composite images. Before each block, they were cued to selectively attend to one category (e.g., scenes). For each trial, observers performed a go/no-go task in which they responded if the image was from one subcategory (e.g., outdoor scenes, 90% probability) but not another (e.g., indoor scenes, 10% probability). We measured behavioral performance in this task before and after an fMRI session to assess training effects. The fMRI session employed the same task, but observers received neurofeedback. This feedback derived from an online multivariate pattern classifier trained to detect whether attention was allocated to the correct category during each block. Feedback was delivered by altering the proportion of each category in the composite images. Namely, we externalized observers’ attentional state by fading the to-be-attended category when the neural measure of attention declined and strengthened this category when it recovered. There was a general improvement in behavior from before to after training. Across participants, the amount of behavioral improvement was correlated with the amount of useful feedback that each participant received (where "useful feedback" was operationalized using a measure of oscillations in the image proportions). These results were not found in a yoked control group that received neurofeedback from the brains of different observers. Thus, real-time fMRI may enable more powerful and customized cognitive training, here enhancing attentional abilities after only one feedback session.
Meeting abstract presented at VSS 2013