Towards EEG-based biomarkers of large scale brain networks

Abstract

Several major functional networks in the brain have been identified, based on sub-regions in the brain that display functionally correlated, synchronous activity and perform common cognitive functions. Three such brain networks (default mode network - DMN, central executive network - CEN, and salience network - SN) form a tri-network model of higher cognitive functioning and are found to be dysregulated in a number of psychopathologies, such as PTSD, autism, schizophrenia, anxiety, depression, bipolar disorder and fronto-temporal dementia (FTD). Current therapies that improve the patient’s cognitive and behavioural states are also found to re-normalize these dysregulated networks, suggesting a correlation between network dysfunction and behavioural dysregulation. Hence, assessing tri-network activity and its dynamics can be a powerful tool to objectively assess treatment response in such psychopathologies. Doing so would most likely rely on functional magnetic resonance imaging (fMRI), as one of the most commonly used modalities for studying such brain networks. While fMRI allows for superior spatial resolution, it poses serious challenges to widespread clinical adoption due to MRI's high operational costs and poor temporal resolution of the acquired signal. One potential strategy to overcome this shortcoming is by identifying the activity of these networks using their EEG-based temporal signatures, greatly reducing the cost and increasing accessibility of using such measures. This thesis takes a step towards improving the clinical accessibility of such brain network-based biomarkers.

Doing so first required the exploration of a popular EEG-based method currently being used to study brain networks in mental health disorders - Microstates. This work uncovered flaws in the core assumptions made in assessing Microstates, necessitating the development of an alternate method to detect such network activity using EEG. To accomplish this, it was important to understand the healthy dynamics between the three brain networks constituting the tri-network model and test one of the core predictions of this model, i.e. the SN gates the DMN and CEN activation based on interoceptive and exteroceptive task demands. Probing this question next uncovered mechanistic details of this process, discovering that the SN co-activates with the task-relevant network. Using this information, a novel machine learning pipeline was developed that used simultaneous EEG-fMRI data to identify EEG-based signatures of the three networks within the tri-network model, and could use these signatures to predict network activation. Finally, the novel machine learning pipeline was trialed in a study investigating the effects of lifestyle interventions on the network dynamics, showing that CEN-SN synchrony can predict response to intervention, while DMN-SN synchrony can develop in those that fail to respond. The understanding of healthy network dynamics gathered from the earlier study helps interpret these results, suggesting that the non-responders persistently activated DMN as a maladaptive strategy.

In conclusion, the studies discussed in this thesis have improved our understanding of healthy network dynamics, uncovered critical flaws in currently popular methods of EEG-based network analysis, provided an alternative methodology to assess network dynamics using EEG, and also validated its use in tracking changes in network synchrony. The identified EEG signatures of widely used functional networks, will greatly increase the clinical accessibility of such brain network measures as biomarkers for neuropathologies. Monitoring the level of network activity in affected subjects may also lead to the development of novel individualized treatments such as brain network-based neurofeedback interventions.