Poster No:
1422
Submission Type:
Abstract Submission
Authors:
Jiwon Lee1, Ye-Eun Kim2, Mikhail Votinov3, Lisa Wagels4, Ute Habel5, Sun-Young Kim6, Munseob Lee7, Han-Gue Jo6
Institutions:
1Kunsan National University, Gunsan-si, Jeollabuk-do, 2Kunsan National University, Kunsan, Jeollabuk do, 3Institute of Neuroscience and Medicine (INM-10), Research Centre Jülich, Jülich, Germany, 4RWTH Aachen University Hospital, Aachen, Germany, 5Department of Psychiatry, Psychotherapy and Psychosomatics, RWTH Aachen University Hospital, Aachen, North Rhine-Westphalia, 6Kunsan National University, Gunsan, Jeollabuk-do, 7Electronics and Communication Research Institute (ETRI), Deajeon, Deajeon
First Author:
Jiwon Lee
Kunsan National University
Gunsan-si, Jeollabuk-do
Co-Author(s):
Ye-Eun Kim
Kunsan National University
Kunsan, Jeollabuk do
Mikhail Votinov
Institute of Neuroscience and Medicine (INM-10), Research Centre Jülich
Jülich, Germany
Lisa Wagels
RWTH Aachen University Hospital
Aachen, Germany
Ute Habel
Department of Psychiatry, Psychotherapy and Psychosomatics, RWTH Aachen University Hospital
Aachen, North Rhine-Westphalia
Munseob Lee
Electronics and Communication Research Institute (ETRI)
Deajeon, Deajeon
Han-Gue Jo
Kunsan National University
Gunsan, Jeollabuk-do
Introduction:
Graph neural network (GNN), a deep learning-based method known for their effectiveness in analyzing graph-structured data, is inherently suitable for the examination of the complex brain network[1]. GNN can operate on non-Euclidean domains, employing convolutions on graphs by leveraging input node features and the relationships between nodes to produce novel features. Recent studies showed that GNN is suitable for analyzing the connectome of functional magnetic resonance imaging (MRI) and can identify the functional network alteration of major depressive disorder (MDD) patients[2,3]. However, previous GNN studies mainly focus on the brain network derived from one modality, either functional imaging or structural imaging[4-6], potentially leading to the observed shortcomings in model accuracy. In this current study, both functional and structural MRIs are combined and analyzed for MDD classification. Specifically, GNN models were formulated to leverage various types of node attributes derived from functional MRI, structural MRI, or their combination. These distinct node features were performed using graph convolutional network (GCN) and graph attention network (GAT) models.
Methods:
GCN and GAT underwent training for classification using resting-state functional and structural MRI scans from a cohort of 775 participants (153 with MDD and 622 healthy controls)[7]. For constructing the graph, a whole-brain functional connectivity matrix was utilized, representing Fisher transformed correlation coefficients among BOLD time series across 100 atlas-based Schaefer's cortical brain regions[8]. Node features were characterized in three ways: i) a nodal functional connectivity vector comprising 100 node features derived from functional MRI, ii) six structural parameters encompassing grey matter volume, fractal dimension, gyrification, Toro GI20, sulcal depth, and thickness extracted from structural MRI, or iii) a combination of both resulting in 106 node features. The fitting procedure and evaluation of predictive models were carried out using a 5-fold cross-validation technique.
Results:
The performance of GNN models was highly depend on the node feature type. GCN and GAT achieved the highest accuracies (73.23%±0.04 and 77.65%±0.05, respectively) when the node features were combined with both functional and structural MRI data. Node features with functional MRI alone showed accuracies of 64.84%±0.02 and 75.27%±0.03 for GCN and GAT, respectively. Node features only with structural MRI exhibited the lowest performance, with accuracies of 62.26%±0.06 and 52.35%±0.07 for GCN and GAT, respectively. The impact of the number of GNN layers and hyperparameters was also tested, but it did not change the effect of the node feature type.
Conclusions:
This study explored the application value of using both brain functional and structural information in modeling GNN and differentiating patients with MDD from healthy controls. Our proposed multi-modal method achieved prediction results of 77% on a large-sample dataset. These findings highlighted the benefits of incorporating structural information into functional MRI in the development of GNN model for classifying MDD brain networks. Notably, GAT demonstrated better performance compared to the GCN, revealing that within GAT – a model that operates attention mechanism to address important nodes in a neighborhood – the impact of node feature types was relatively diminished. This observation suggests for the integration of functional and structural information, not solely in node features but also about designing the graph architecture in a way that facilitates the fusion of both types of information.
Disorders of the Nervous System:
Psychiatric (eg. Depression, Anxiety, Schizophrenia) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Keywords:
FUNCTIONAL MRI
Machine Learning
Psychiatric Disorders
STRUCTURAL MRI
1|2Indicates the priority used for review
Provide references using author date format
1. Zhou, Jie, et al. "Graph neural networks: A review of methods and applications." AI open 1 (2020): 57-81.
2. Qin, Kun, et al. "Using graph convolutional network to characterize individuals with major depressive disorder across multiple imaging sites." EBioMedicine 78 (2022).
3. Venkatapathy, Sujitha, et al. "Ensemble graph neural network model for classification of major depressive disorder using whole-brain functional connectivity." Frontiers in Psychiatry 14 (2023): 1125339.
4. Zhang, Hao, et al. "Classification of brain disorders in rs-fMRI via local-to-global graph neural networks." IEEE Transactions on Medical Imaging 42.2 (2022): 444-455.
5. Gao, Jingjing, et al. "Brain age prediction using the graph neural network based on resting-state functional MRI in Alzheimer's disease." Frontiers in Neuroscience 17 (2023).
6. Subaramya, S., et al. "Graph Neural Network based Alzheimer’s Disease Classification using Structural Brain Network." 2022 22nd International Conference on Advances in ICT for Emerging Regions (ICTer). IEEE, 2022.
7. Tanaka, Saori C., et al. "A multi-site, multi-disorder resting-state magnetic resonance image database." Scientific data 8.1 (2021): 227.
8. Schaefer, Alexander, et al. "Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity MRI." Cerebral cortex 28.9 (2018): 3095-3114.