Premium
Eigen analysis on ultrafast FMRI data shows altered dynamic functional connectivity in MCI individuals
Author(s) -
Park Sung Min,
Rane Swati
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.040052
Subject(s) - resting state fmri , dynamic functional connectivity , leverage (statistics) , functional connectivity , human connectome project , artificial intelligence , computer science , pattern recognition (psychology) , neuroscience , psychology
Background Resting state functional connectivity (rsFC) is a powerful tool to detect brain dysfunction. Recent studies show that resting state networks are not static, but dynamic and alter connectivity in seconds. The use of simultaneous multislice imaging technology allows sub‐second fMRI acquisitions in ADNI3. We leverage eigenanalysis to compare brain connectivity states and their temporal dynamicity between cognitive normal older adults (NC) and MCI participants. Method We analyzed 105 ADNI subjects (50 MCI [22M], age = 74 ± 8 and 55 NC [35M], age = 75 ± 8) with T1 and BOLD fMRI acquisition (TR = 600 ms). Processing included motion correction, detrending, despiking, physiological noise correction and registration to standard space. We performed harmonization using ComBat to minimize site‐related variability. We employed a sliding window approach followed by eigenanalysis 1 to extract spatially independent connectivity states between regions derived from 4 networks (see Table 1). For each connectivity state, we identified whether the timecourse was different between NC and MCI subjects and created a final difference network comparing the brain state between the two groups. Result Figure 1 shows the 10 brain connectivity states, spanning 32 regions and 4 networks, identified using eigennetwork analysis. Figure 2 shows the timecourse for each of the connectivity states. Finally, Figure 3 shows the difference network for the MCI group with and without harmonization. Harmonization appears to have a considerable effect on the connectivity state pattern. We then compared the harmonized difference network to within‐site difference networks. RMSE was smaller for the harmonized network than that of the unharmonized network. Conclusion We demonstrate the use of eigennetwork analysis to determine a dynamic brain connectivity state in MCI that is different than NC. The advantages of this approach are two‐fold: One, all brain regions and resting state networks are considered as a whole instead of studying each network separately. Two, it allows us to study the temporal properties of the networks in addition to their spatial characteristics. This analysis enables us to capture a more accurate depiction of the connectivity of the brain and its dynamic nature.