Characterising subpopulation variability in the spatial organisation of functional brain networks
Poster No:
1421
Submission Type:
Abstract Submission
Authors:
Rezvan Farahibozorg1, Mark Woolrich1, Stephen Smith1
Institutions:
1University of Oxford, Oxford, UK
First Author:
Co-Author(s):
Introduction:
Big fMRI data, e.g., UK Biobank[1], provide resources to examine the brain function, not only in population and individuals, but also in various subject subgroups nested within the big data. Spatial topographies of the brain's functional networks vary across individuals, and characteristics shared within each subgroup can carry important information about traits and disease (e.g., dementia subgroups)[2]. Therefore, incorporating subgroup characteristics within the modelling of brain networks has the potential to enhance our ability to extract strong biomarkers of disease using fMRI. Crucially, however, for datasets in the scale of UKB, many subgroups are unknown, and unsupervised techniques that can scale to these datasets and extract reliable and phenotypically relevant subgroups are still lacking. We propose a new approach to characterise subgroups based on cross-individual variability in the spatial topographies of brain networks, and demonstrate novel patterns of subgroup variability in the brain, across a range of cognitive, health and lifestyle phenotypes.
Methods:
fMRI data and network modelling: resting fMRI (rfMRI) from 20,000 subjects in UKB were used. Individual-specific RSNs were modelled using Probabilistic Functional Modes (PFMs)[3]. Subject fMRI data (D) into a set of functional modes, represented by spatial maps (P), time courses (A), amplitudes (H) and noise (E): D=PHA+E. Spatial maps of RSNs were used as bases for fingerprint extraction.
Fingerprint extraction: FIMRIB's Linked ICA (FLICA)[4], a Bayesian approach that finds consensus independent components across RSNs, was used to identify continuous axes of subject variability (i.e., fingerprints). Additionally, subject-ICA (i.e., treating 'subjects' as variables) was applied on FLICA's output to further identify axes of subject variability. 2 x 500 = 1,000 fingerprints were extracted from these two methods.
Subgroup Modelling: We next used Gaussian Mixture Modelling (across subject weights) on each fingerprint, while optimising the number of subgroups using Bayesian Information Criteria and Davies Bouldin algorithm. 330 + 459 (=789) of FLICA + subject-ICA fingerprints yielded more than 1 subgroup.
Evaluation: We evaluated subgroup definitions using: a) reproducibility analysis; b) distinctiveness; c) statistical differences with respect to phenotypes; d) spatial maps of subgroup variability.
Fingerprint extraction: FIMRIB's Linked ICA (FLICA)[4], a Bayesian approach that finds consensus independent components across RSNs, was used to identify continuous axes of subject variability (i.e., fingerprints). Additionally, subject-ICA (i.e., treating 'subjects' as variables) was applied on FLICA's output to further identify axes of subject variability. 2 x 500 = 1,000 fingerprints were extracted from these two methods.
Subgroup Modelling: We next used Gaussian Mixture Modelling (across subject weights) on each fingerprint, while optimising the number of subgroups using Bayesian Information Criteria and Davies Bouldin algorithm. 330 + 459 (=789) of FLICA + subject-ICA fingerprints yielded more than 1 subgroup.
Evaluation: We evaluated subgroup definitions using: a) reproducibility analysis; b) distinctiveness; c) statistical differences with respect to phenotypes; d) spatial maps of subgroup variability.
·Figure1
Results:
First (Fig. 1b), split-half reproducibility analysis showed that both the number of subgroups derived from each fingerprint, and subject assignments to each subgroup were highly reproducible for 600 (of the 789) of the subgroup definitions. Additionally, we found the 789 subgroup divisions to be highly distinct from each other.
Second (Fig. 1c), statistical comparison of subgroups with respect to a range of over 1000 phenotypes across Cognitive, Alcohol, and Tobacco measures, and Bone, Cardiovascular and Mental health, showed many significant differences. This is despite subgroups being identified using phenotype-unaware, unsupervised methods.
Third, we conducted voxel-wise comparison of subgroups for 11 RSNs in two categories: 5 Sensory-motor and 6 Cognitive RSNs. Interestingly, we found (Fig. 2a) that spatial maps of subgroup variability within Sensory-motor and Cognitive RSNs were highly similar to each other, and distinct from the other category, highlighting two distinct modes of subgroup variability across the brain related to sensory-motor vs cognitive processing. Additionally, we found (Fig. 2b) distinct maps of subgroup variability related to phenotype categories, with pairs of Cognitive/Tobacco, Alcohol/Mental health, Cardiovascular/Bone health showing highest similarities to each other.
Second (Fig. 1c), statistical comparison of subgroups with respect to a range of over 1000 phenotypes across Cognitive, Alcohol, and Tobacco measures, and Bone, Cardiovascular and Mental health, showed many significant differences. This is despite subgroups being identified using phenotype-unaware, unsupervised methods.
Third, we conducted voxel-wise comparison of subgroups for 11 RSNs in two categories: 5 Sensory-motor and 6 Cognitive RSNs. Interestingly, we found (Fig. 2a) that spatial maps of subgroup variability within Sensory-motor and Cognitive RSNs were highly similar to each other, and distinct from the other category, highlighting two distinct modes of subgroup variability across the brain related to sensory-motor vs cognitive processing. Additionally, we found (Fig. 2b) distinct maps of subgroup variability related to phenotype categories, with pairs of Cognitive/Tobacco, Alcohol/Mental health, Cardiovascular/Bone health showing highest similarities to each other.
·Figure2
Conclusions:
We propose new techniques to model subgroup variability in big fMRI data based on spatial characteristics of the brain networks. We identify several interesting patterns of how these subgroups are significantly different in the brain and with respect to phenotype categories.
Modeling and Analysis Methods:
Bayesian Modeling
fMRI Connectivity and Network Modeling 1
Methods Development
Multivariate Approaches 2
Task-Independent and Resting-State Analysis
Keywords:
FUNCTIONAL MRI
Machine Learning
Modeling
Other - functional connectivity, big data fMRI, subpopulation modelling, PROFUMO, phenotype prediction
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
[2] N. Filippini et al., “Distinct patterns of brain activity in young carriers of the APOE-ε4 allele,” Proc Natl Acad Sci U S A, vol. 106, no. 17, pp. 7209–7214, 2009, doi: 10.1073/pnas.0811879106.
[3] S.-R. Farahibozorg et al., “Hierarchical modelling of functional brain networks in population and individuals from big fMRI data,” Neuroimage, 2021, doi: 10.1016/j.neuroimage.2021.118513.
[4] W. Gong, C. Beckmann, and S. Smith, “Phenotype Discovery from Population Brain Imaging,” bioRxiv, pp. 1–35, 2020, doi: 10.1101/2020.03.05.973172.