Identifying and Manipulating
Functional Networks of the Human Brain
Evidence from modern neuroscience studies shows us that cognition involves not just a single brain structure but multiple brain structures that are coupled, forming a network that represents coordinated activity. Networks are transient and dynamically formed to support different cognitive processes. At the neuronal level, activity of neuronal populations are coordinated by rhythmic brain activity, measured as local field potentials (LFPs), across a relatively large spatial extent. This coordination is observed when neurons synchronize their activity according to particular frequencies and phases of the LFPs.
To identify functional networks from the EEG (i.e., far-field measurements of the LFP), we have developed tools to compute association matrices (e.g., phase-amplitude coupling) of the EEG source estimates. These association matrices are then decomposed using non-linear methods, such as artificial neural networks (ANN), with connectivity constraints, derived from structural (tracts) data, to identify networks associated with particular cognitive processes. These networks may include subcortical structures whose activities are not seen in the EEG but are revealed as “latent” structures of a given network.
As we identify the networks, we can bring tools to manipulating them, such as recently demonstrated by the work of Reinhart (2017, 2019). We propose a feasibility project wherein we take the Individual Neuromorphic Model (IN Model) and understand the network of slow-wave (SW) sleep. The IN Model will be tuned to mirror SW sleep features of the individual. In the next stage, we predict, based on the IN Model, the effects of noninvasive transcranial electrical stimulation (TES) on the network dynamics of SW for the individual. The results will then be compared against the actual responses of the network to TES during SW sleep. In multiple iterations, data from the individual will be used to refine the IN model, with the goal of minimizing the prediction error. The end result, in concept, will be a tuned IN Model of the individual’s SW sleep, representing a first step towards Individual Neuromorphic Emulation.
The Role of Sleep in Brain-Machine Communication
As we understand sleep, this knowledge may provide insights into how we can understand and replicate various states of consciousness. Perhaps there is a continuum of consciousness that resembles the development of the brain, from early inutero rhythmic neurophysiological signatures, to their recapitulation in sleep-wake states, and to those seen during coma and/or anesthesia. Just as the scientific analysis of the mind requires that we understand the nightly consolidation of memory in sleep, training machine intelligence to characterize an individual’s sleep process may be a basic method for improving brain-machine communication during the person’s waking cognition.