What happens in the brain during rest? And is this resting state different in people sleeping or suffering from depression? This question is difficult to answer because rest is a complex, multifaceted process involving cognition, (lack of) attention, and sporadic activation of the senses. The answer, however, may provide new insight into brain function in both health and disease, as well as explain individuality, paving the way to precision medicine.
The brain has traditionally been divided into regions based on specific functional responsibilities. However, it has become widely accepted that cognition and background brain activity involves not only single regions but rather networks of anatomically distinct brain areas. Such networks become apparent when studying average region-to-region signal similarity using brain scanning techniques such as functional magnetic resonance imaging (fMRI) or electroencephalography (EEG).
However, studying connectivity networks through time averages assumes that brain activity is stationary. This complicates the assessment of changes to the resting state, such as falling asleep or the gradual strengthening of the effect of a psychedelic drug. Even the resting state itself is a dynamic corpus of thoughts and emotions. Dynamic functional connectivity is the study of “states” of brain connectivity. The brain switches dynamically between these states, and each state represents a connectivity strength between brain regions, as seen in the figure. We compute these states using unsupervised machine learning.
This project aims to model dynamic brain connectivity by clustering instantaneous interregional coherence estimates. Coherence is a measure of oscillatory similarity (how much signals swing together) and is a circular measure bounded by
and
. We will cluster the coherence estimates using distributions from directional statistics that acknowledge the inherent geometry of the data. Specifically, we will 1) develop the theoretical and computational aspects of the matrix-variate directional distributions and 2) describe commonly occurring brain states as an atlas of the healthy brain to compare, e.g., disease or perturbations to the resting state. Also, we will 3) develop models for multimodal studies combining the virtues of several brain scanning techniques.
With dynamic functional connectivity, we work towards describing interregional information processing in human brains in detail. The ultimate goal is precision medicine. In the distant future, we hope to offer patients suffering from complex brain-related disorders, such as depression, a set of brain scans that, when analyzed, can predict the optimal treatment for each individual.