Trait Anxiety Prediction Using Structural Connectivity with GNN - Pooling Layers and Vectorization
Poster No:
1190
Submission Type:
Abstract Submission
Authors:
Sung-Chul Jung1, Hyun-Joo Song2, Dong-Hyun Kim1
Institutions:
1Department of Electrical and Electronic Engineering, Yonsei University, Seoul, Korea, Republic of, 2Department of Psychology, Yonsei University, Seoul, Korea, Republic of
First Author:
Sung-Chul Jung
Department of Electrical and Electronic Engineering, Yonsei University
Seoul, Korea, Republic of
Department of Electrical and Electronic Engineering, Yonsei University
Seoul, Korea, Republic of
Co-Author(s):
Dong-Hyun Kim
Department of Electrical and Electronic Engineering, Yonsei University
Seoul, Korea, Republic of
Department of Electrical and Electronic Engineering, Yonsei University
Seoul, Korea, Republic of
Introduction:
Trait anxiety refers to enduring psychological characteristic reflecting a general tendency of individuals to experience anxiety across various situations. Accurately predicting trait anxiety levels is crucial as it reflects psychological state and informs effective mental health treatments. However, few studies have focused on this purpose using MR images and this study addresses this gap. This study uses structural connectivity matrices derived from DWI and T1 data as input and STAI-T score representing trait anxiety levels. Instead of Connectome Based Predictive Modeling (CPM) [1], which uses linear model, we propose a Graph Neural Network (GNN) to capture nonlinear pattern and enhance predictive performance. Key components of GNN, e.g. pooling layers and vectorization methods, are critical for aggregating node features and representing graph-level information. Their optimization is vital for accurate predictions and meaningful structural connectivity analysis. This study focused on prediction accuracy and comparison of various pooling layers and vectorization methods.
Methods:
This study utilized the LEMON dataset for DWI, T1 data and phenotype data [2]. DWI data was preprocessed using MRtrix3 [3], FSL, [4], ANTs [5], Freesurfer [6]. T1-weighted image was parcellated to 84 ROIs defined based on Desikan-Killiany atlas via Freesurfer [7]. The 84 ROIs served as nodes in the structural connectivity matrix and the edge weights consist of the summation of mean Fractional Anisotropy (FA) values per each streamline for a given node pair, derived from tractography. The 84 x 84 symmetric structural connectivity matrix per subject was input into the GNN model. The GNN model incorporated a Graph Attention layer to reflect properties of structural connectivity and was modified from a reference model [8]. We evaluated Top-k pooling and soft pooling under single-layer and double-layer configurations, as shown in Figure 1. For vectorization, max and mean pooling, attention-weighted summation, and a combination of all methods were tested to assess their impact on predictive performance . This study included 47 subjects and employed Leave-One-Out-Cross-Validation (LOOCV) for evaluation. Performance metrics included Mean Absolute Error, Pearson correlation and Spearman correlation coefficient.
Results:
Figure 2 summarizes the research results, highlighting the effectiveness of the proposed method across various pooling layers and vectorization methods. For instance, while the Pearson correlation coefficient in previously reported study was 0.32, our best combination achieved 0.73 in similar datasets [9]. Among pooling layers, prediction using Top-k pooling outperformed those using soft pooling [10] by effectively reducing noise, demonstrating that noise reduction is more critical for prediction accuracy than variability in key nodes. For vectorization methods, prediction using attention weighted summation showed superior performance by dynamically assigning importance to nodes, compared to mean and max pooling, which risk overlooking critical information. Combining all these methods played a complementary role, leading to improved performance. However, with Top-k pooling, attention-weighted summation was infeasible, leaving Mean and Max pooling as the only options. Nevertheless, this configuration achieved the best performance by effectively reducing noise.
Conclusions:
This study explored the impact of pooling layers and vectorization methods on predicting trait anxiety levels using structural connectivity data. It investigated the best combination of pooling layers and vectorization methods in GNN model for predicting trait anxiety levels using structural connectivity matrices extracted from DWI data. Moreover, our approach showed superior performance on trait anxiety levels compared to previously reported CPM results in similar datasets. This study aims to further advance towards accurately predicting trait anxiety levels using only MR images without the need for surveys.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia)
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural) 1
Keywords:
Anxiety
Machine Learning
Modeling
MRI
Psychiatric
Psychiatric Disorders
STRUCTURAL MRI
Other - Connectivity
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] Babayan, A., Erbey, M., Kumral, D., Reinelt, R. D., Reiter, A. M. F., Röbbig, J. … Villringer, A. (2019). A Mind-Brain-Body dataset of MRI, EEG, cognition, emotion, and peripheral physiology in young and old adults. Scientific Data, 6, Article 180308.
[3] Tournier, J. D., Smith, R., Raffelt, D., Tabbara, R., Dhollander, T., Pietsch, M. … Connelly, A. (2019). MRtrix3: A fast, flexible and open software framework for medical image processing and visu alization. Neuroimage, 202, Article 116137.
[4] Smith, S. M., Jenkinson, M., Woolrich, M. W., Beckmann, C. F., Behrens, T. E., Johansen-Berg, H. … Matthews, P. M. (2004). Advances in functional and structural MR image analysis and implementation as FSL. NeuroImage, 23(Suppl 1), S208-219.
[5] Avants, B. B., Tustison, N. J., Song, G., Cook, P. A., Klein, A., & Gee, J. C. (2011). A reproducible evaluation of ANTs similarity metric performance in brain image registration. NeuroImage, 54(3), 2033–2044.
[6] Fischl, B., van der Kouwe, A., Destrieux, C., Halgren, E., Segonne, F., Salat, D. H. … Dale, A. M. (2004). Automatically parcellating the human cerebral cortex. Cerebral Cortex, 14(1), 11–22.
[7] Desikan, R. S., Segonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C., Blacker, D. … Killiany, R. J. (2006). An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. Neuroimage, 31(3), 968–980.
[8] Kwon, H.; Kim, J.; Son, S.; Jang, Y.; Kim, B.; Lee, H.; Lee, J. (2022). Sparse Hierarchical Representation Learning on Functional Brain Networks for Prediction of Autism Severity Levels. Frontiers in Neuroscience, 16, Article 935431.
[9] Yoo, C., Park, S. & Kim, M.J. (2022) Structural connectome-based prediction of trait anxiety. Brain Imaging and Behavior, 16(6), 2467–2476.
[10] Stergiou, A., Poppe, R., & Kalliatakis, G. (2021). Refining activation downsampling with SoftPool. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 10357–10366.