Tracking Frequency Specific Instantaneous Brain Connectivity Dynamics in Resting State fMRI
Poster No:
1762
Submission Type:
Abstract Submission
Authors:
Muchen Li1, Zirui Song1, Yuyang Xu1, Yang Zhou1, Yanxi Wang1, Zhao Qing1
Institutions:
1Southeast University, Nanjing, JiangSu
First Author:
Co-Author(s):
Introduction:
Typical resting-state functional magnetic resonance imaging (fMRI) studies measured brain functional connectivity (FC) and brain networks based on calculating the correlations between fMRI time series within a typical frequency band, typically 0.01-0.1Hz, which was supposed to represent neural activities. However, these process implicated assumptions of the temporal stationarity throughout the measurement period, as well as no frequency specificity within this 0.01-0.1Hz band.
There were amounts of studies reported fMRI signal have frequency-specific features even within 0.01-0.1Hz. Moreover, lots of studies have already investigated the "dynamic FC" and showed the importance temporal variation of FC. However, the parameters used to divide fMRI signal is often chosen arbitrarily, and limited the generalizability and comparability across these studies.
The Hilbert Huang Transform (HHT) method is an adaptive iterative algorithm widely used in engineering. HHT decomposes signals into intrinsic model functions (IMFs) and define instantaneous phase, amplitude, and frequency for each IMF. Therefore, we apply HHT on fMRI to achieve adaptive frequency decomposition and high temporal resolution synchronization measurement based on instantaneous phase difference.
There were amounts of studies reported fMRI signal have frequency-specific features even within 0.01-0.1Hz. Moreover, lots of studies have already investigated the "dynamic FC" and showed the importance temporal variation of FC. However, the parameters used to divide fMRI signal is often chosen arbitrarily, and limited the generalizability and comparability across these studies.
The Hilbert Huang Transform (HHT) method is an adaptive iterative algorithm widely used in engineering. HHT decomposes signals into intrinsic model functions (IMFs) and define instantaneous phase, amplitude, and frequency for each IMF. Therefore, we apply HHT on fMRI to achieve adaptive frequency decomposition and high temporal resolution synchronization measurement based on instantaneous phase difference.
Methods:
We downloaded one subset of the public database "Consortium for Reliability and Reproducibility, CORR," which is named"BNU1". 50 healthy young subjects (male/female: 30/20; age: 19–30 years) were included in the current study. The fMRI data here have 200 time points with TR=2s, voxel size=3.125×3.125×4.2 mm, matrix = 64×64, 33 slices. The preprocessing was performed by in DPABI, including realignment, slice-timing, covariates regression and spatial normalization.
After preprocessing the time series were extracted for each Brainnetome atlas region. Then HHT was applied on each fMRI time series and instantaneous FC between each two regions for different frequency bands were decided by the following steps: 1). Using empirical model decomposition subdivided signals from two regions into different IMFs. 2). Instantaneous phase, amplitude, and frequency for each IMF were calculated. 3). Hilbert weighted frequency (HWF) was calculated from each IMF. 4). IMFs with HWF out of 0.01-0.1Hz were excluded, and the remaining IMFs from two brain regions were coupled based on the similarity of their HWF. 5). The Cosine values of the difference of the coupled IMFs' instantaneous phases were defined as instantaneous FC value at the frequency corresponded to the IMFs used. Similarly, we extend such steps and generated FC matrices among all 246 Brainnetome regions
Then, similar to previous studies measured brain states using dynamic FC by K-means and further compared the results across different frequency bands.
After preprocessing the time series were extracted for each Brainnetome atlas region. Then HHT was applied on each fMRI time series and instantaneous FC between each two regions for different frequency bands were decided by the following steps: 1). Using empirical model decomposition subdivided signals from two regions into different IMFs. 2). Instantaneous phase, amplitude, and frequency for each IMF were calculated. 3). Hilbert weighted frequency (HWF) was calculated from each IMF. 4). IMFs with HWF out of 0.01-0.1Hz were excluded, and the remaining IMFs from two brain regions were coupled based on the similarity of their HWF. 5). The Cosine values of the difference of the coupled IMFs' instantaneous phases were defined as instantaneous FC value at the frequency corresponded to the IMFs used. Similarly, we extend such steps and generated FC matrices among all 246 Brainnetome regions
Then, similar to previous studies measured brain states using dynamic FC by K-means and further compared the results across different frequency bands.
Results:
As shown in Figure.1, the IMFs from all brain regions and subjects have similar frequency distribution. There were 3 typical IMFs have HWF within 0.01-0.1Hz for all subjects and regions, which mean HWF were 0.01Hz, 0.02Hz and 0.05Hz, while the IMFs have higher frequency have larger HWF variation among regions and subjects.
We further investigated brain states based on these 3 frequency bands. As shown in Figure. 2, in each band, there were 5 brain states. No matter to which frequency, there was one state which have longest duration, and showing that brain was organized into clusters: default mode, attention related and primary regions (visual, auditory and sensorimotor). A Chi-square test showed this main brain states across 3 bands significantly (P<0.001) to occur at the same time points. Besides, the Chi-square test also showed some of the minor states in different bands were temporal coupled across frequencies.
We further investigated brain states based on these 3 frequency bands. As shown in Figure. 2, in each band, there were 5 brain states. No matter to which frequency, there was one state which have longest duration, and showing that brain was organized into clusters: default mode, attention related and primary regions (visual, auditory and sensorimotor). A Chi-square test showed this main brain states across 3 bands significantly (P<0.001) to occur at the same time points. Besides, the Chi-square test also showed some of the minor states in different bands were temporal coupled across frequencies.
Conclusions:
The current study applied HHT on fMRI and reveal brain network dynamics at instantaneous level and also among different frequency. Moreover, brain dynamics have 3 characteristic frequency bands while the brain network dynamics showed coupling across bands.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Keywords:
Other - dynamic functional connectivity; Hilbert transform; brain states
1|2Indicates the priority used for review
Provide references using author date format
Kringelbach, M. L. (2020). "Brain States and Transitions: Insights from Computational Neuroscience." Cell Rep 32(10): 108128.
Schumacher, J. (2019). "Dynamic functional connectivity changes in dementia with Lewy bodies and Alzheimer's disease." Neuroimage Clin 22: 101812.
Yang, H.(2022). "Frequency-specific coactivation patterns in resting-state and their alterations in schizophrenia: An fMRI study." Hum Brain Mapp 43(12): 3792-3808.