PROJECT SUMMARY/ABSTRACT
Neuropsychiatric conditions are increasingly being understood as disorders of intrinsic, functional interactions
within and between widespread, distributed, brain networks. Given recent advances in functional Magnetic
Resonance Imaging (fMRI) data acquisition and computational analysis, it is now possible to reliably map the
functional neuroanatomy of brain networks within individuals, offering a potential avenue for identifying
personalized neurotherapeutic targets. However, gold standard treatments (e.g. pharmacotherapy) in current
psychiatric practice were not originally designed to target specific brain network interactions and lack protocols
that leverage such individual-level data. Real-time neurofeedback— whereby patients observe and learn to
regulate selected aspects of their own brain activity— is a candidate approach to personally tailor the
normalization of unhealthy communication within and between brain networks. However, to target the major
brain networks that function abnormally in neuropsychiatric conditions, neurofeedback relies on fMRI, which is
an expensive procedure involving a complex setup and patient burden. The goal of this project is to develop an
electroencephalography (EEG) “fingerprint” of fMRI network dynamics so that a neurofeedback system based
on EEG (electrodes placed on the scalp) alone can be used to precisely target interactions within and between
brain networks. Because EEG devices can be portable and offer relatively simple setup in flexible settings, our
work could enable a scalable form of network-based neurofeedback training that patients could regularly
access. Our Aim 1 is to identify an optimal, generalizable model of EEG features that are predictive of fMRI-
based default mode network (DMN) “antagonism” within individuals. We focus on this DMN antagonism
because it is a major feature that is relevant to cognitive dysfunction in psychiatric disease at a transdiagnostic
level. We will collect high-quality, simultaneous EEG-fMRI data in 24 healthy adults (>100 mins of sampling per
participant), including three conditions: (1) resting state, (2) continuous task performance, and (3) continuous
fMRI-based neurofeedback from DMN antagonism states. We will apply machine learning-based methods to
identify an optimal mapping between EEG signal components and fMRI-based DMN antagonism. Further, we
will determine how much individual-level EEG-fMRI sampling is needed to successfully predict DMN
antagonism from EEG. Our Aim 2 is to test whether EEG markers of DMN antagonism are predictive of
cognitive task performance fluctuations within individuals. As such, our findings could offer validation of the
behavioral relevance of an EEG neurofeedback system that would target DMN antagonism. If successful, our
work can lead to development of an accessible, computational psychiatry tool that can be tested in clinical
conditions in which DMN antagonism (and related cognitive function) is affected, including attention-
deficit/hyperactivity disorder, depression and schizophrenia.