Networks in nonlinear fMRI connectivity are present during infancy and exhibit associations with age
Poster No:
1770
Submission Type:
Abstract Submission
Authors:
Spencer Kinsey1, Masoud Seraji1, Sarah Shultz2, Vince Calhoun1, Armin Iraji1
Institutions:
1Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Atlanta, GA, 2Emory University School of Medicine, Atlanta, GA
First Author:
Spencer Kinsey
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Co-Author(s):
Masoud Seraji
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Vince Calhoun
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Armin Iraji
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)
Atlanta, GA
Introduction:
Independent component analysis (ICA) is often used to estimate brain intrinsic connectivity networks (ICNs) from functional magnetic resonance imaging (fMRI) time series. Although such analyses are typically designed to extract ICNs that reflect linear (LIN) functional connectivity (FC), we previously showed that ICNs can be extracted from voxel-wise distance correlation patterns that move beyond signed Pearson correlation patterns (i.e., from explicitly nonlinear (ENL) whole-brain FC (ENL-wFC)) (Iraji et al., 2023; Kinsey et al., 2023). Here, we extend our connectivity domain (Iraji et al., 2016) ICA framework by estimating ICNs from distance correlation patterns that move beyond Pearson correlation magnitudes. We find that ICNs are reliably extracted from infant resting-state fMRI (rsfMRI) data using our approach. Moreover, we show that the LIN and ENL estimates of many canonical ICNs exhibit spatial variation during infancy, and that some ENL estimates show associations with age that are missed by LIN counterparts.
Methods:
We analyzed 442 rsfMRI scans collected from infants during a study conducted by Marcus Autism Center, Atlanta, GA, i/o CHOA. Subjects were classified as low-likelihood (LL) or elevated likelihood (EL) for autism spectrum disorder (ASD). The subject pool included 94 LL infants and 59 EL infants (n = 153). Data were collected with Siemens TIM Trio (TR = 720ms; TE = 33ms) and Siemens MAGNETOM Prisma (TR = 800ms; TE = 37ms) scanners.
Preprocessing involved 1) discarding the first ten scans, 2) head motion, distortion, and slice timing correction, 3) normalization to MNI 152 space, 4) spatial smoothing with a 6mm FWHM Gaussian kernel, 7) motion regression, detrending, and despiking, 8) temporal resampling to TR = 720ms, and 9) time series Z-scoring.
We constructed LIN whole-brain FC (LIN-wFC) and ENL-wFC matrices for each scan (Fig. 1A). We calculated the voxel-wise Pearson correlation coefficient (PCC), which is a conventional linear FC estimator, to construct LIN-wFC. Then, we calculated the voxel-wise distance correlation (Székely et al., 2007) to construct nonlinear whole-brain FC (NL-wFC). To extract ENL-wFC, we removed the effect of the absolute value of LIN-wFC on NL-FC by estimating a value of ∝ which minimized the sum of squared errors between NL-wFC and |LIN-wFC| for each scan.
The top 30 components from principal component analysis were used as input to group-level spatial independent component analysis via Group ICA of fMRI Toolbox (GIFT v4.0). A component was identified as an ICN if and only if 1) it exhibited peak weight in and high overlap with gray matter, 2) it passed visual inspection, and 3) it exhibited ICASSO stability > .80. ICNs were matched based on spatial similarity.
ICNs reconstructed from LL infant scans via group information-guided ICA (Du & Fan, 2013) were statistically analyzed. Voxel-wise GLMs were constructed to assess spatial variation between LIN and ENL ICNs. Voxel-wise GLMs were also constructed to determine LIN and ENL ICN associations with age.
Preprocessing involved 1) discarding the first ten scans, 2) head motion, distortion, and slice timing correction, 3) normalization to MNI 152 space, 4) spatial smoothing with a 6mm FWHM Gaussian kernel, 7) motion regression, detrending, and despiking, 8) temporal resampling to TR = 720ms, and 9) time series Z-scoring.
We constructed LIN whole-brain FC (LIN-wFC) and ENL-wFC matrices for each scan (Fig. 1A). We calculated the voxel-wise Pearson correlation coefficient (PCC), which is a conventional linear FC estimator, to construct LIN-wFC. Then, we calculated the voxel-wise distance correlation (Székely et al., 2007) to construct nonlinear whole-brain FC (NL-wFC). To extract ENL-wFC, we removed the effect of the absolute value of LIN-wFC on NL-FC by estimating a value of ∝ which minimized the sum of squared errors between NL-wFC and |LIN-wFC| for each scan.
The top 30 components from principal component analysis were used as input to group-level spatial independent component analysis via Group ICA of fMRI Toolbox (GIFT v4.0). A component was identified as an ICN if and only if 1) it exhibited peak weight in and high overlap with gray matter, 2) it passed visual inspection, and 3) it exhibited ICASSO stability > .80. ICNs were matched based on spatial similarity.
ICNs reconstructed from LL infant scans via group information-guided ICA (Du & Fan, 2013) were statistically analyzed. Voxel-wise GLMs were constructed to assess spatial variation between LIN and ENL ICNs. Voxel-wise GLMs were also constructed to determine LIN and ENL ICN associations with age.
Results:
Our analysis uncovered 11 common ICNs (similarity > .80) and 1 ICN unique to each dataset (similarity < .40). GLM statistics revealed that matched LIN and ENL counterparts exhibit unique spatial variation during infancy (Fig. 1B-E). For instance, the posterior default mode (pDM) ICN exhibits a visible ENL-LIN gradient between the precuneus and bilateral angular gyri. GLM testing also revealed distinct age association patterns, with many ENL voxels showing associations with age that are missed by LIN (Fig.2).
Conclusions:
Here, we leveraged connectivity domain ICA to estimate ICNs from ENL-wFC patterns within infant rsfMRI data. Our results show that the ENL counterparts of many large-scale canonical ICNs are present during early postnatal life, are differentiated via their spatial distributions, and show associations with age that may provide valuable insight into typical and atypical early life brain development.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism)
Lifespan Development:
Early life, Adolescence, Aging 2
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling 1
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Autism
Development
FUNCTIONAL MRI
Statistical Methods
Other - NONLINEAR; FUNCTIONAL CONNECTIVITY; DISTANCE CORRELATION; INDEPENDENT COMPONENT ANALYSIS (ICA); INTRINSIC CONNECTIVITY NETWORK (ICN)
1|2Indicates the priority used for review
Provide references using author date format
Iraji, A., et al. (2016), “The connectivity domain: Analyzing resting state fMRI data using feature-based data-driven and model-based methods,” NeuroImage, vol. 134, pp. 494-507. https://doi.org/10.1016/j.neuroimage.2016.04.006
Iraji, A., et al. (2023), 'The nonlinear brain: towards uncovering hidden brain networks using explicitly nonlinear functional interaction,' IEEE International Symposium on Biomedical Imaging.
Kinsey, S., et al. (2023), “Networks extracted from nonlinear fMRI connectivity exhibit unique spatial variation and enhanced sensitivity to differences between individuals with schizophrenia and controls,” bioRxiv. https://doi.org/10.1101/2023.11.16.566292
Székely, G. J., et al. (2007), “Measuring and Testing Dependence by Correlation of Distances,” The Annals of Statistics, vol. 35, no. 6, pp. 2769-2794. 35(6), 2769–2794. http://www.jstor.org/stable/25464608