Professor Stephen LaConte | Harnessing supervised learning-based real-time fMRI to test the dynamic function of resting state networks

Spontaneous “resting-state” neural activity was initially thought to be noise, but the spatially distributed coherence patterns of these fluctuations produce known, reproducible networks. It is now further recognized that these networks are broadly implicated in healthy cognition and across the entire spectrum of brain health issues. Nevertheless, the functional roles of individual networks as well as their coordinated activity remains elusive. In this talk, I will describe the potential of supervised learning-based real-time functional magnetic resonance (rtfMRI) to move beyond correlational studies try to directly examine the functional roles of these networks. In conventional task-based neuroimaging, the type and timing of stimuli are presented to optimize statistical brain mapping. In closed-loop experiments such as rtfMRI, however, the stimulus “input” does not have to be independent of ongoing brain activity. Instead, activity can act as a control signal for ongoing adaptation of the stimulus. This enables tremendous experimental flexibility and can support the causal study of specific brain activity by enabling brain activity to control factors such as stimulus category, intensity, and timing. Because supervised learning is at the core of our approach, much of our effort focuses on developing models that can predict brain function from fMRI scans. The talk will explore a range of supervised learning strategies, but will mostly focus on our recent efforts to model and track resting-state network dynamics. We will attempt to argue that pattern-based real-time fMRI (rtfMRI) enables powerful and flexible approaches to directly test and potentially consolidate the amassed findings related to resting state networks.