Seed-based correlation analysis and instantaneous global correlation analysis for resting state fMRI

Abstract

Brain disorders have an increasingly poignant socioeconomic impact, and persons with mental illness and mental disorders, Alzheimer's disease, dementia, Parkinson's disease, and epilepsy are intensely affected by these disorders. The development of novel tools to analyze and interpret brain behavior is thus essential for investigating the activity of the brain. The purpose of this work is to introduce and characterize two new methods for understanding the activity of the brain.

The first method will introduce a technique to address the limitation encountered by depending on a single seed point for seed-based correlation analysis (SCA), and instead submit a probabilistic formulation of SCA which is robust to variations in the initial seed point. The method will firstly produce the strength of the correlations for voxels strongly correlated to the posterior cingulate cortex. Additionally, this approach provides a probabilistic interpretation of functional connectivity network behavior in the brain, and a maximum a posteriori (MAP) estimation of regions belonging to the default mode network (DMN) and dorsal attention network will be demonstrated as an example application of the procedure. In order to establish the ability of proposed method to provide comparably reliable results to those of SCA, the group maps resulting from the method termed seed cloud SCA (SC-SCA) will be compared against those calculated from performing SCA and region-based SCA. The SC-SCA method will also be compared against its region-based counterpart, region-based SC-SCA in order to determine any extra advantages gained by performing a region-based correlation. The statistical significance of the difference between methods will be evaluated by comparing the difference in the median of the standard deviation of voxels across subjects. The components resulting from the decomposition of the FC network from SCA and SC-SCA via independent components analysis will also be compared. This method represents a fully automated approach with probabilistic interpretations which confers augmented understanding of the DMN and its relation to underlying brain functionality.

The second proposed method is three-fold in studying dynamic functional connectivity (dFC) at the time scale of seconds. First, this approach will use data-driven techniques with minimal extrinsic expert knowledge to distinguish between the functional connectivity of the brain during periods of large co-activations and during intervals when the brain has low co-activations. The method is inspired by the point process technique proposed by Tagliazucchi in 2012, but instead defines a time interval which identifies avalanching periods based on the magnitude of whole-brain correlation. This allows the data to be reduced into two classes - avalanching periods and non-avalanching periods. Secondly, a set of dFC co-activation patterns (CAPs) will then be characterized, and are the networks of connectivity that are formed during avalanching periods and non-avalanching periods. Lastly, the spatial propagation of the avalanche will be demonstrated by using the regions of highest activation to track the epicenters of activity. These new tools may help to explain the behavior of the brain, which can be used to study and understand brain disorders to provide better diagnoses and treatments for those affected.