Integrating Graph Features for Dynamic Functional Connectivity Prediction: A Deep Learning Approach
Poster No:
1476
Submission Type:
Abstract Submission
Authors:
Kristen Wingert1, Divesh Thaploo1, Adebiyi Sobitan1, Atsuko Kurosu1, Ninet Sinaii2, Nadia Biassou2
Institutions:
1National Institutes on Deafness and Other Communication Disorders, Bethesda, MD, 2National Institutes of Health, Bethesda, MD
First Author:
Co-Author(s):
Introduction:
This study introduces an innovative graph-based deep learning framework for predicting jackknife dynamic functional connectivity (DFC) patterns using static connectivity and graph-based features. DFC is a vital metric in neuroscience, providing insights into evolving neural interactions over time. The proposed framework addresses the challenge of modeling both temporal and structural dynamics by learning complex relationships between static connectivity matrices, graph features, and jackknife correlation values during training.
Methods:
Data Collection and Preprocessing:
Task-based fMRI data from 172 healthy adults (86 females, 86 males) were queried from the Human Connectome Project (HCP) (Barch et al., 2013). Participants performed an auditory language comprehension task and a baseline mathematical control task. Imaging data were acquired on a 3T scanner (TR=720 ms, TE=33.1 ms, voxel size=2 mm). MRI preprocessing included FreeSurfer-based surface parcellation (Fischl, 2012) and AFNI pipelines (Cox, 1996), with motion correction, bandpass filtering (0.01–0.1 Hz), and spatial smoothing (6 mm FWHM). Static functional connectivity (FC) matrices were generated using Pearson correlation across 68 Desikan-Killiany atlas ROIs (Desikan et al., 2006), and Bonferroni-corrected p-values ensured significance. Networks were filtered to retain connections present in at least 85% of participants, resulting in robust graph structures spanning intra- and inter-hemispheric links.
Framework and Architecture:
The proposed model comprises three components: (1) static connectivity matrices, (2) node-level graph features, and (3) dynamic adjacency matrices. A custom Graph Convolutional Network (GCN) layer processes node features while incorporating edge attributes to aggregate neighborhood information. Static embeddings, initialized as identity matrices, are tiled across temporal dimensions to facilitate sequence-level modeling. Temporal dependencies are captured using a Long Short-Term Memory (LSTM) layer, with residual connections enhancing information preservation and convergence. Specifically, residual connections combine LSTM outputs with mean-transformed static embeddings to stabilize predictions. A dense layer refines the temporal context vector, culminating in the prediction of jackknife correlation values.
Task-based fMRI data from 172 healthy adults (86 females, 86 males) were queried from the Human Connectome Project (HCP) (Barch et al., 2013). Participants performed an auditory language comprehension task and a baseline mathematical control task. Imaging data were acquired on a 3T scanner (TR=720 ms, TE=33.1 ms, voxel size=2 mm). MRI preprocessing included FreeSurfer-based surface parcellation (Fischl, 2012) and AFNI pipelines (Cox, 1996), with motion correction, bandpass filtering (0.01–0.1 Hz), and spatial smoothing (6 mm FWHM). Static functional connectivity (FC) matrices were generated using Pearson correlation across 68 Desikan-Killiany atlas ROIs (Desikan et al., 2006), and Bonferroni-corrected p-values ensured significance. Networks were filtered to retain connections present in at least 85% of participants, resulting in robust graph structures spanning intra- and inter-hemispheric links.
Framework and Architecture:
The proposed model comprises three components: (1) static connectivity matrices, (2) node-level graph features, and (3) dynamic adjacency matrices. A custom Graph Convolutional Network (GCN) layer processes node features while incorporating edge attributes to aggregate neighborhood information. Static embeddings, initialized as identity matrices, are tiled across temporal dimensions to facilitate sequence-level modeling. Temporal dependencies are captured using a Long Short-Term Memory (LSTM) layer, with residual connections enhancing information preservation and convergence. Specifically, residual connections combine LSTM outputs with mean-transformed static embeddings to stabilize predictions. A dense layer refines the temporal context vector, culminating in the prediction of jackknife correlation values.
Results:
Results:
The model was evaluated via leave-one-subject-out cross-validation. Across held-out subjects, the test R² scores ranged from 0.727 to 0.911, with corresponding mean squared errors (MSE) between 0.0043 and 0.0234 and mean absolute errors (MAE) as low as 0.052. Cross-validation further demonstrated the model's robustness, with validation R² scores averaging 0.55 to 0.71 across folds. Loss stabilized at 0.008041, confirming consistent generalization to unseen data.
The model was evaluated via leave-one-subject-out cross-validation. Across held-out subjects, the test R² scores ranged from 0.727 to 0.911, with corresponding mean squared errors (MSE) between 0.0043 and 0.0234 and mean absolute errors (MAE) as low as 0.052. Cross-validation further demonstrated the model's robustness, with validation R² scores averaging 0.55 to 0.71 across folds. Loss stabilized at 0.008041, confirming consistent generalization to unseen data.
Conclusions:
Model Novelty and Contribution:
The proposed framework integrates static connectivity, node-level graph features, and temporal dependencies into a unified pipeline for predicting dynamic functional interactions. Notably, it introduces a custom GCN layer to learn node embeddings while accounting for edge attributes, enhancing structural feature representation. By decoupling static inputs from temporal predictions during testing, the model offers an efficient solution adaptable to real-world applications.
Conclusion and Applications:
By combining graph-based features with temporal modeling, this framework advances the state-of-the-art in functional connectivity analysis. Its ability to generalize across subjects underscores its potential in decoding and predicting dynamic neural interactions, paving the way for personalized neuroscience. Applications include cognitive research, clinical diagnostics for neurological disorders, and the development of biomarkers for personalized medicine. Future work will explore integrating multi-modal neuroimaging data and extending the model to larger, more diverse datasets.
The proposed framework integrates static connectivity, node-level graph features, and temporal dependencies into a unified pipeline for predicting dynamic functional interactions. Notably, it introduces a custom GCN layer to learn node embeddings while accounting for edge attributes, enhancing structural feature representation. By decoupling static inputs from temporal predictions during testing, the model offers an efficient solution adaptable to real-world applications.
Conclusion and Applications:
By combining graph-based features with temporal modeling, this framework advances the state-of-the-art in functional connectivity analysis. Its ability to generalize across subjects underscores its potential in decoding and predicting dynamic neural interactions, paving the way for personalized neuroscience. Applications include cognitive research, clinical diagnostics for neurological disorders, and the development of biomarkers for personalized medicine. Future work will explore integrating multi-modal neuroimaging data and extending the model to larger, more diverse datasets.
Language:
Language Comprehension and Semantics
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling 1
Task-Independent and Resting-State Analysis
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
ADULTS
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
Language
Machine Learning
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?
Which processing packages did you use for your study?
Provide references using APA citation style.
1. Barch DM, Burgess GC, Harms MP, Petersen SE, Schlaggar BL, Corbetta M, et al. (2013). Function in the human connectome: Task-fMRI and individual differences in behavior. NeuroImage, 80:169-89.
2. Cox RW. (1996). AFNI: software for analysis and visualization of functional magnetic resonance neuroimages, 29(3):162-73.
3. Desikan RS, Ségonne F, Fischl B, Quinn BT, Dickerson BC, Blacker D, et al. (2006). An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage, 31(3):968-80.
4. Fischl B. (2012). FreeSurfer. Neuroimage, 62(2):774-81