Evaluation of Structural-Functional Coupling Mechanism on Human Connectome Project Using HoloBrain

1739 

Abstract Submission 

Huan Liu¹, Tingting Dan¹, defu yang¹, Won Hwa Kim², Minjeong Kim³, Paul Laurienti⁴, Guorong Wu¹

¹University of North Carolina at Chapel Hill, Chapel Hill, NC, ²Pohang University of Science and Technology (POSTECH), Pohang, Korea, Republic of, ³University of North Carolina at Greensboro, Greensboro, NC, ⁴Wake Forest School of Medicine, Winston Salem, NC

Huan Liu  
University of North Carolina at Chapel Hill
Chapel Hill, NC

Tingting Dan  
University of North Carolina at Chapel Hill
Chapel Hill, NC

defu yang  
University of North Carolina at Chapel Hill
Chapel Hill, NC

Won Hwa Kim  
Pohang University of Science and Technology (POSTECH)
Pohang, Korea, Republic of

Minjeong Kim  
University of North Carolina at Greensboro
Greensboro, NC

Paul Laurienti  
Wake Forest School of Medicine
Winston Salem, NC

Guorong Wu  
University of North Carolina at Chapel Hill
Chapel Hill, NC

Human brain is a complex wiring system in which diverse behaviors are supported by ubiquitous functional fluctuations. Although striking efforts have been made to investigate the association between brain structure connectivity (SC) and function connectivity (FC), the SC-FC coupling mechanism is largely elusive. Recently, we have developed a novel analytic approach, called HoloBrain (Liu, Dan et al. 2023), to characterize the interference patterns formed by the BOLD signals modulated by a collection of harmonic wavelets (stemming from wirings of white matter fibers) across a widespread graph spectrum. In this work, we have applied our HoloBrain technique to task-fMRI data from HCP-YA. Compared to conventional network analysis methods, HoloBrain offers a new window to investigate the task-specific footprint of SC-FC coupling through the lens of cross-frequency coupling (CFC), demonstrating great potential in discovering novel neurobiological biomarkers for resting-state fMRI studies.

Method overview of HoloBrain. First, we construct harmonic wavelets from the Laplacian matrix of SC by (1) calculating harmonic waves based on the Laplacian matrix, (2) localizing each harmonic wave to each brain region and form region-specific (indexed by i) and frequency-specific (denoted by s) harmonic wavelet ψ_i^s using graph signal processing techniques (Hammond, Vandergheynst et al. 2011). Second, we modulate the snapshot of BOLD signal f(t) at time t with each harmonic wavelets and generate a time series of harmonic power p_i^s. Third, we construct a collection of local CFC patterns CFC_ij^sr by the inner product of harmonic power between i-th and j-th brain regions and across frequencies s and r. Supposing we break down the whole graph spectrum into four frequency bands (Fig. 1(c)), the output of HoloBrain is a 4×4 CFC matrix for each node/edge.
Neuroscience insight of HoloBrain. Following the concept of wave-to-wave interference (Gabor 1948), we have developed a proof-of-concept approach to computationally "record" the CFC of time-evolving interference patterns that are formed by superimposing the harmonic wavelets on the subject-specific neural activities.
Group comparison using HoloBrain. For each CFC value, we apply a linear regression model to examine whether the underlying CFC manifests significant difference between two cognitive tasks, where age and gender are confounders.

We evaluate the statistical power of HoloBrain on task-based fMRI of HCP-YA dataset, where each subject is partitioned into 360 brain regions (HCP atlas). In the following experiments, we only show the significant motor (1) vs. language (0) result at the significance level of 10^-8, along with the effect size greater than 0.2 and adjusted R²>0.2. In Fig. 2 top, it is evident that the identified brain regions and connections are aligned with current findings in the neuroscience field (Glasser, Coalson et al. 2016), which are located at the somatosensory, motor cortex, and Auditory Association cortex. Compared to the conventional network analysis using FC matrix only (Fig. 2(a)), HoloBrain provides a new dimension of brain connectome in the frequency domain. Furthermore, we apply the same analysis to the re-test scans (shown in Fig. 2 bottom). In terms of replicability, the group comparison results by HoloBrain are more consistent than conventional FC analysis, as indicated by circles.

We evaluate the statistical power and replicability performance of our recently developed HoloBrain technique on task-fMRI data in HCP-YA. Since HoloBrain characterizes the SC-FC coupling mechanism via the interference between SC-modulated BOLD signals, we are able to examine the groupwise SC-FC coupling differences on both brain regions and connections across the graph spectrum.

Classification and Predictive Modeling ²

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling ¹

Computational Neuroscience

Data analysis

FUNCTIONAL MRI

Modeling

Statistical Methods

STRUCTURAL MRI

Other - Cross Frequency Coupling