Deep Learning-Based Brain Functional Connectivity Insights into MDD and SZ
Poster No:
1104
Submission Type:
Abstract Submission
Authors:
Jiwon Lee1, Ye-Eun Kim2, Mikhail Votinov3, Sun-Young Kim4, Ute Habel5, Han-Gue Jo6
Institutions:
1Kunsan National University, Kunsan, Korea, Republic of, 2Kunsan National Univeristy, Kunsan, Korea, Republic of, 3Research Center Jülich, Juelich, Germany, 4School of Mechanical Engineering, Kunsan National University, Kunsan, Korea, Republic of, 5RWTH Aachen University, Aachen, Germany, 6Department of AI Convergence, College of Computer and Software, Kunsan National Univeristy, Kunsan, Korea, Republic of
First Author:
Co-Author(s):
Sun-Young Kim
School of Mechanical Engineering, Kunsan National University
Kunsan, Korea, Republic of
School of Mechanical Engineering, Kunsan National University
Kunsan, Korea, Republic of
Han-Gue Jo
Department of AI Convergence, College of Computer and Software, Kunsan National Univeristy
Kunsan, Korea, Republic of
Department of AI Convergence, College of Computer and Software, Kunsan National Univeristy
Kunsan, Korea, Republic of
Introduction:
Classical statistical methods have been employed to identify brain regions and features associated with psychiatric disorders, primarily through group-level comparisons to detect significant differences [1]. While these approaches have provided valuable insights, they struggle to capture the high-dimensional, non-linear, and non-Euclidean relationships inherent in brain imaging data, potentially overlooking subtle yet critical patterns. Graph neural network (GNN) models have emerged as powerful tools for analyzing the complex network structure of brain data, offering robust capabilities for classifying psychiatric disorders [2,3]. Beyond their predictive power, modern deep learning frameworks can also be designed to reveal the key features driving their decisions. Techniques such as feature attribution and layer-wise relevance propagation allow for the extraction of biologically meaningful patterns, providing new insights into the neurobiological underpinnings of psychiatric disorders.
Methods:
We analyzed resting-state fMRI data derived from the SRPBS open dataset, focusing on individuals with major depressive disorder (MDD) and schizophrenia (SZ) (MDD = 153; SZ = 147), along with age-, sex-, and site-matched healthy controls (HC) for each disorder (HC-MDD = 98; HC-SZ = 115) [4]. Whole-brain functional connectivity matrices were generated using Fisher-transformed correlation coefficients of BOLD time series across 100 atlas-based Schaefer cortical brain regions [5]. Graph convolutional networks (GCN), graph attention networks (GAT), and Self-Attention Graph Pooling (SAGPool) models were implemented to classify MDD and SZ from HC groups. The models demonstrating the highest validation accuracy for MDD and SZ were selected for further analysis.
To identify the key features driving model predictions, we employed a perturbation-based feature analysis. Each functional connectivity feature was systematically altered, and the resulting model predictions were compared with those from the original data. Larger deviations in predictions were interpreted as indicating the relative importance of the altered feature in classifying MDD or SZ.
To identify the key features driving model predictions, we employed a perturbation-based feature analysis. Each functional connectivity feature was systematically altered, and the resulting model predictions were compared with those from the original data. Larger deviations in predictions were interpreted as indicating the relative importance of the altered feature in classifying MDD or SZ.
Results:
The SAGPool model demonstrated the highest performance in classifying both MDD and SZ, achieving validation accuracies of 77.04% and 81.35%, respectively. These best-performing models were further analyzed to identify the key functional connectivity features contributing to their classification decisions. For the MDD classification model, the most critical functional connectivities were concentrated within the default mode network (DMN), while for the SZ classification model, the key functional connectivities were primarily located within the salience/ventral attention network (SVAN) and limbic system (Figure 1 bottom panel).
A similar pattern of results was observed with the second-best models, reinforcing the robustness of these findings. Additionally, we compared the significant features derived from the models with those identified through a Student's t-test across all functional connectivity data between groups. This comparison revealed diverse connectivity patterns, with notable distinctions between the features identified by the GNN models and those highlighted by the statistical analysis (Figure 1 top panel).
A similar pattern of results was observed with the second-best models, reinforcing the robustness of these findings. Additionally, we compared the significant features derived from the models with those identified through a Student's t-test across all functional connectivity data between groups. This comparison revealed diverse connectivity patterns, with notable distinctions between the features identified by the GNN models and those highlighted by the statistical analysis (Figure 1 top panel).
·Figure 1. The top 10 most significant functional connections (FCs) identified by the GNN model and Student’s t-test
Conclusions:
This study highlights the potential of GNN models, particularly the SAGPool architecture, in accurately classifying MDD and SZ based on resting-state fMRI data. By leveraging advanced deep learning techniques, we not only achieved high classification performance but also identified critical functional connectivity patterns that underpin these psychiatric conditions. These results underscore the complementary strengths of deep learning approaches and statistical methods in identifying critical functional connectivity features.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia)
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI)
Classification and Predictive Modeling 1
Connectivity (eg. functional, effective, structural) 2
fMRI Connectivity and Network Modeling
Keywords:
Computational Neuroscience
Data analysis
FUNCTIONAL MRI
Machine Learning
Schizophrenia
Other - Major Depressive Disorder, Deep-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?
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] Sunil, G., Gowtham, S., Bose, A., Harish, S., & Srinivasa, G. (2024). Graph neural network and machine learning analysis of functional neuroimaging for understanding schizophrenia. BMC neuroscience, 25(1), 2.
[3] Qin, K., Lei, D., Pinaya, W. H., Pan, N., Li, W., Zhu, Z., ... & Gong, Q. (2022). Using graph convolutional network to characterize individuals with major depressive disorder across multiple imaging sites. EBioMedicine, 78.
[4] Tanaka, S. C., Yamashita, A., Yahata, N., Itahashi, T., Lisi, G., Yamada, T., ... & Imamizu, H. (2021). A multi-site, multi-disorder resting-state magnetic resonance image database. Scientific data, 8(1), 227.
[5] Schaefer, A., Kong, R., Gordon, E. M., Laumann, T. O., Zuo, X. N., Holmes, A. J., ... & Yeo, B. T. (2018). Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI. Cerebral cortex, 28(9), 3095-3114.