A Simple Patch-Based Blind Source Identification Method for fMRI Analysis
Poster No:
1430
Submission Type:
Abstract Submission
Authors:
YAORONG XIAO1, Rogers Silva2, Oktay Agcaoglu3, Vince Calhoun4, Sergey Plis1
Institutions:
1Georgia State University, Atlanta, GA, 2TReNDS Center, Atlanta, GA, 3Tri-Institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS), Atlanta, GA, 4GSU/GATech/Emory, Atlanta, GA
First Author:
Co-Author(s):
Oktay Agcaoglu, PhD
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:
Infomax (Bell & Sejnowski, 1995) is commonly applied to functional magnetic resonance imaging(fMRI) data sets to detect spatial components, enabling researchers to uncover underlying brain activity patterns, functional connectivity networks(FNC), and sources of noise. We propose a novel method, JDX, which leverages joint diagonalization and shifts the focus from independence properties to the structural features of the sources. Experimental results demonstrate that JDX produces robust results in a faster manner compared to infomax. The components captured by JDX provide a complementary analytical approach for exploring brain activity patterns based on experiments.
Methods:
A goal of JDX is to find the demixing matrix in \(\mathbf{X} = \mathbf{B} \mathbf{S}\), where \( \mathbf{X} \) represents the observed mixtures, \( \mathbf{S} \) is the estimated components. We utilized joint diagonalization on covariance matrices computed from sub-batches of the mixtures. These sub-batches are selected in a structured manner. For example, in our experiment, the mixture has dimensions \( 53 \times 63 \times 52 \times 100 \), and one covariance matrix can be computed using a \( 10 \times 10 \times 10 \times 100 \) sub-batch of the mixture. A transformation matrix that can approximately joint diagonalize these covariance matrices \( \mathbf{C} \) is considered as the final demixing matrix \( \mathbf{B} \).
The joint diagonalization problem can be formulated as:
\[
\mathbf{B} = \underset{\mathbf{B}}{\arg \min} \sum_{i=1}^n \left\| \text{off}(\mathbf{B C}_i \mathbf{B}^T) \right\|_F^2,
\]
where \( \mathbf{C}_i \) are the series of covariance matrices, \( \text{off}(A) = A - \text{diag}(A) \), \( \| \cdot \|_F \) denotes the Frobenius norm.
fMRI features, appearing as blobs within the sub-batch area, are effectively captured. This approach avoids the limitation of forcing components into uncorrelated conditions and ensures stable results when the same sub-batch is used consistently.
The joint diagonalization problem can be formulated as:
\[
\mathbf{B} = \underset{\mathbf{B}}{\arg \min} \sum_{i=1}^n \left\| \text{off}(\mathbf{B C}_i \mathbf{B}^T) \right\|_F^2,
\]
where \( \mathbf{C}_i \) are the series of covariance matrices, \( \text{off}(A) = A - \text{diag}(A) \), \( \| \cdot \|_F \) denotes the Frobenius norm.
fMRI features, appearing as blobs within the sub-batch area, are effectively captured. This approach avoids the limitation of forcing components into uncorrelated conditions and ensures stable results when the same sub-batch is used consistently.
Results:
We utilize the Functional Biomedical Informatics Research Network (FBIRN) dataset for this study which contains health control(HC) and schizophrenia(SZ) patients. ICA is performed at the group level (Calhoun et al., 2001). PCA is initially applied to each subject along the time dimension, and the data from all subjects are then concatenated along the time dimension. A second PCA is performed on the concatenated dataset, reducing the dimensionality to 100 components. 53 components were selected based on their correlation with the Neuromark template (Du et al., 2020). The corresponding spatial maps were visualized using Brainbow (https://pypi.org/project/brainbow/). JDX, used 200 sub-patches, achieves these results in an average of 18 seconds, whereas Infomax requires an average of 34 seconds based on 10 independent runs.
The HC-SZ matrices are computed by averaging the FNC values across all HC subjects and subtracting the corresponding values from individuals with SZ. The spatial maps and FNC shows that JDX captured different features compared to Infomax, nevertheless, the classification experiment shows that features extracted by JDX achieve better performance. The classification test applied by using Polyssifier (https://pypi.org/project/polyssifier/). It shows AUC socres on multiple classic machine learning models trained by using the back-reconstructed FNCs from each subject. The results for each model demonstrate that JDX outperforms the Infomax approach in classifying schizophrenia.
The HC-SZ matrices are computed by averaging the FNC values across all HC subjects and subtracting the corresponding values from individuals with SZ. The spatial maps and FNC shows that JDX captured different features compared to Infomax, nevertheless, the classification experiment shows that features extracted by JDX achieve better performance. The classification test applied by using Polyssifier (https://pypi.org/project/polyssifier/). It shows AUC socres on multiple classic machine learning models trained by using the back-reconstructed FNCs from each subject. The results for each model demonstrate that JDX outperforms the Infomax approach in classifying schizophrenia.
·Spatial maps of Infomax and JDX aligned by correlation of Neuromark template.
·HC-SZ FNC matrices and classification results based on polysifer test across 8-Fold Cross-Validations.
Conclusions:
We developed a new blind source identification method, JDX. It can capture components without independence assumption, making it more generalized. JDX can achieve faster and robust performance in a simple way by joint diagonalize a list of covariance matrices of sub-pathces. The spatial maps generated by JDX enable better performance in classification tasks, indicating that JDX captures more meaningful results.
Modeling and Analysis Methods:
Classification and Predictive Modeling
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Methods Development 2
Keywords:
Data analysis
FUNCTIONAL MRI
Machine Learning
Schizophrenia
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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?
Provide references using APA citation style.
maximization approach to blind separation and blind
deconvolution. Neural computation, 7(6), 1129–
1159.
Calhoun, V. D., Adali, T., Pearlson, G. D., & Pekar, J. J.
(2001). A method for making group inferences from
functional mri data using independent component
analysis. Human brain mapping, 14(3), 140–151.
Du, Y., Fu, Z., Sui, J., Gao, S., Xing, Y., Lin, D., . . . others
(2020). Neuromark: An automated and adaptive
ica based pipeline to identify reproducible fmri
markers of brain disorders. NeuroImage: Clinical,
28, 102375.