Real-time Control of Decoded Brain Activity Using Artificial Neural Networks

点击翻译按钮获取中文摘要

Our psychological well-being is greatly affected by our subjective emotional experiences. Yet we know surprisingly little about the brain mechanisms that generate such subjective states. Scientific progress in the field has been limited in part because we possess few methods capable of controlling fine-grained brain activity at will. Here we propose to address this issue with a new method combining artificial intelligence and brain imaging. More precisely, we propose to leverage modern artificial neural networks to generate "synthetic" visual stimuli capable of triggering specific patterns of brain activity. Previous experiments conducted in macaque monkeys showed that such synthetic stimuli were far more effective than the best natural stimuli at increasing the firing rate of specific neurons in the visual system. Importantly, such synthetic stimuli were also not recognized as instances of any particular object category which means that they could potentially be used without participants' awareness of the purpose or content of the manipulated brain activity. This NSERC Discovery grant will investigate if synthetic visual stimuli can be used to robustly and unconsciously generate subjective affective experiences. More precisely, we will determine if subjective experiences of fear can be generated by unconsciously activating the brain representation of feared animals in the ventral visual system. This proposal has two main objectives: First, we will adapt and validate an approach to generate synthetic images from functional magnetic resonance imaging (fMRI) activity. We will use two existing datasets to generate synthetic images associated with the visual presentation of animals from 30 different categories. Second, we will recruit participants reporting fear of some of the animals in our dataset. In a series of pre-registered double-blind placebo controlled fMRI experiments, we will determine if synthetic images generated from feared animal categories can modulate the targeted brain activity, subjective fear and physiological defensive responses. We hypothesize that synthetic stimuli will increase the activation of the brain representation of the feared animal and generate subjective experiences of fear and physiological defensive responses. This research project represents one of the first attempts to use artificial neural networks to control brain activity in humans. The synthetic stimuli created using this approach could present a unique opportunity to control specific brain representations in isolation, a feat that is usually impossible to achieve using natural stimuli. Such an intervention method could allow us to develop a direct means to interrogate how the brain generates subjective emotional experiences, representing a very valuable asset for the study of affective neurosciences and human consciousness.