Enhancing Group-Level Discriminant fMRI Analysis with Independent Filter Analysis (IFA)
Poster No:
1457
Submission Type:
Abstract Submission
Authors:
Zain Souweidane1, Stephen Smith2, Christian Beckmann1
Institutions:
1Donders Institute for Brain, Cognition, and Behaviour, Nijmegen, Netherlands, 2University of Oxford, Oxford, Oxfordshire
First Author:
Co-Author(s):
Introduction:
Group-level fMRI analysis, particularly group Independent Component Analysis (ICA) with dual regression (Nickerson et al., 2017), effectively captures biologically relevant spatial networks (RSNs). However, group ICA is limited by shared component bias, prioritizing commonalities over discriminative features. This limitation arises from the PCA-based projection of the data into a lower-dimensional space before ICA unmixing, which preserves bulk variance rather than group-discriminative features. This hinders downstream analyses, like classifying network matrices (netmats) from dual regression (Dadi et al., 2019). Independent Filter Analysis (IFA) addresses this by preserving group-discriminative information while maintaining biologically interpretable spatial networks.
Methods:
IFA identifies a spatial basis that preserves biologically meaningful networks and group-level differences in functional connectivity. To achieve this, MIGP (Smith et al., 2014) and Probabilistic Principal Component Analysis (PPCA) (Beckmann & Smith, 2004) are applied to extract a basis representing functional spatial networks. TSSF (Xu et al., 2020) or SPADE (Llera et al., 2020) is used to identify discriminative components that maximize group-level differences in functional connectivity. TSSF is ideal for parcellated data, providing robust and flexible classification of group differences, while SPADE can generalize this approach to voxel data, making it suitable for higher-resolution analyses. If discrimination is done in a parcellated space, the discriminant filters are projected to voxel space via dual regression (Nickerson et al., 2017). The PPCA and discriminative bases are then combined and unmixed using ICA. Following ICA, dual regression is applied again to generate individual-level spatial maps and netmats. The full IFA pipeline is visualized in Figure 1.
To assess IFA's ability to preserve group differences, the subject-wise netmats are mapped into a tangent space, centered on the mean netmats from the training group (last step in Figure 1). Group separability is quantified using various machine learning classifiers, with classification accuracy providing an empirical measure of discriminability (Pervaiz et al., 2020). A tangent-space t-test is performed on the netmats to identify statistically significant group differences in connectivity (Varoquaux et al., 2010).
To assess IFA's ability to preserve group differences, the subject-wise netmats are mapped into a tangent space, centered on the mean netmats from the training group (last step in Figure 1). Group separability is quantified using various machine learning classifiers, with classification accuracy providing an empirical measure of discriminability (Pervaiz et al., 2020). A tangent-space t-test is performed on the netmats to identify statistically significant group differences in connectivity (Varoquaux et al., 2010).
Results:
The Human Connectome Project (HCP) dataset (Glasser et al., 2013) was used to evaluate IFA against group ICA across three scenarios: simulated two-group differences embedded into resting-state data, between-task differences (e.g., working memory vs. motor) in the HCP task FMRI data, and within-task differences (e.g., 0-back vs. 2-back working memory). IFA consistently enhanced group-level discrimination while preserving biologically interpretable networks. The degree of improvement depended on the inherent separability of the groups and the total number of PCA components used in the analysis. For large between-task differences, IFA achieved modest gains in netmat classification accuracy. However, for more subtle within-task differences, such as 0-back faces vs. 2-back faces, IFA substantially improved classification accuracy, increasing performance by 9% (Figure 2a)-from 87% (ICA with 10 PCA components) to 96% (IFA with 6 PCA components and 4 discriminant components). Additionally, IFA identified 35 statistically significant connectivity differences in tangent-space t-tests, compared to 26 for ICA (Figure 2b), highlighting its improved ability to detect interpretable group-level differences in functional connectivity.
Conclusions:
IFA addresses shared component bias in group ICA by retaining group-discriminative features while preserving the ability to extract biologically interpretable networks (RSNs). By improving classification accuracy and enabling the detection of group differences in netmats, IFA offers a robust approach for identifying meaningful group-level variations in functional connectivity.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Methods Development
Task-Independent and Resting-State Analysis
Keywords:
Data analysis
FUNCTIONAL MRI
Machine Learning
Multivariate
Statistical Methods
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. Dadi, K. (2019). Benchmarking functional connectome-based predictive models for resting-state fMRI. NeuroImage, 192, 115–134. https://doi.org/10.1016/j.neuroimage.2019.02.062
3. Glasser, M. F. (2013). The minimal preprocessing pipelines for the Human Connectome Project. NeuroImage, 80, 105–124. https://doi.org/10.1016/j.neuroimage.2013.04.127
4. Llera, A. (2020). Spatial Patterns for Discriminative Estimation (p. 746891). bioRxiv. https://doi.org/10.1101/746891
5. Nickerson, L. D. (2017). Using Dual Regression to Investigate Network Shape and Amplitude in Functional Connectivity Analyses. Frontiers in Neuroscience, 11. https://doi.org/10.3389/fnins.2017.00115
6. Pervaiz, U. (2020). Optimising network modelling methods for fMRI. NeuroImage, 211, 116604. https://doi.org/10.1016/j.neuroimage.2020.116604
7. Smith, S. M. (2014). Group-PCA for very large fMRI datasets. NeuroImage, 101, 738–749. https://doi.org/10.1016/j.neuroimage.2014.07.051
8. Varoquaux, G. (2010). Detection of Brain Functional-Connectivity Difference in Post-stroke Patients Using Group-Level Covariance Modeling. In T. Jiang, N. Navab, J. P. W. Pluim, & M. A. Viergever (Eds.), Medical Image Computing and Computer-Assisted Intervention – MICCAI 2010 (pp. 200–208). Springer. https://doi.org/10.1007/978-3-642-15705-9_25
9. Xu, J. (2020). Tangent space spatial filters for interpretable and efficient Riemannian classification. Journal of Neural Engineering, 17(2), 026043. https://doi.org/10.1088/1741-2552/ab839e