MMM-BNF: multi-atlas,multi-modal,multi-site dynamic brain network fusion for schizophrenia diagnosis
Poster No:
1111
Submission Type:
Abstract Submission
Authors:
Yixin Ji1, Jin Zhang2, Qi Zhu1, Rongtao Jiang3, Vince Calhoun4, Daoqiang Zhang1, Shile Qi1
Institutions:
1Nanjing University of Aeronautics and Astronautics, Nanjing, Jiangsu, 2Northwestern Polytechnical University, Xian, Shanxi, 3Yale School of Medicine, New Haven, CT, 4GSU/GATech/Emory, Atlanta, GA
First Author:
Co-Author(s):
Introduction:
Dynamic brain networks (DBNs), derived from functional magnetic resonance imaging (fMRI), outperform static brain networks (SBNs) in capturing variations in functional connectivity and have become a useful tool in diagnosing psychiatric disorders1, like schizophrenia (SZ). However, most existing DBN methods rely on a single brain atlas, limiting their ability to capture the multi-scale organization of the brain. While multi-atlas methods offer a comprehensive perspective, they face challenges in the complementarity between anatomical topology and functional activity. In addition, the heterogeneity introduced by multi-site data is rarely considered in multi-atlas fusion2, limiting the effectiveness of DBN identification.
Methods:
fMRI and structural MRI (sMRI) were collected from the Function Biomedical Informatics Research Network (FBIRN) phase III3 and the Bipolar-Schizophrenia Network for Intermediate Phenotypes II (B-SNIP II)4. The FBIRN consisted of 161 SZ patients and 157 normal controls (NCs) from 7 sites, while the B-SNIP II included 133 SZ patients and 160 NCs from 4 sites. The brain was parcellated into regions using three atlases: AAL (90 regions)5, Brainnetome (246 regions)6, and Yeo (130 regions)7. BOLD time series from fMRI were used to construct DBNs via sliding window and Pearson's correlation, while gray matter volumes derived from sMRI were used to build brain structural networks (BSNs) using Jensen divergence8 (Fig. 1a). Each subject's BSN was represented as an adjacency matrix, while brain functional networks from each window were used as the node features. These inputs were fed into a spatial-temporal feature extraction module to extract intrinsic temporal dynamics and spatial topological features (Fig. 1b). Then a multi-site similarity graph was constructed, with edges adaptively determined by feature similarity and site-feature dependencies. This graph served as input to a graph convolutional network (GCN) for feature extraction, which was followed by two fully connected layers and a rectified linear unit for classification (Fig. 1c).
Results:
(1) Fusing all brain atlases outperformed using a single or two atlases, achieving the best classification performance in distinguishing SZ from controls. The use of spatial-temporal feature extraction module improved classification on both datasets. In addition, incorporating the multi-site GCN enhanced the performance across all combination of brain atlases, demonstrating its effectiveness in leveraging multi-site data (Fig. 2-I); (2) Comparing with the other 5 multi-atlas fusion methods, the proposed method achieved superior performance across most metrics, with ACC improvements of at least 7.26% and 6.13% on two datasets (Fig. 2-II); (3) The identified discriminative brain regions across different atlases on the two datasets were primarily located in the frontal and temporal lobes, which were consistent with exiting studies. The overlapping information provided by different atlases further enhances confidence in identifying these regions (Fig. 2-III).
Conclusions:
This study proposed a multi-atlas, multi-modal and multi-site DBN fusion method for SZ diagnosis. A spatial-temporal feature extraction module was developed for each atlas to integrate DBNs and BSNs, capturing the temporal dynamics and spatial topological features. The multi-site GCN modeled feature similarity and dependencies across sites, and the extracted features from all atlases were concatenated for classification. Experimental results showed that our method outperformed the other multi-atlas fusion methods. Ablation studies highlighted the significant contributions of atlas quantity, spatial-temporal feature extraction, and the multi-site GCN to the improved performance. Furthermore, the identified discriminative brain regions provided insights into the functional and structural frontal-temporal dysfunction of SZ. In summary, our method has a prospect to identify multi-site SZ patients in clinical applications.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
fMRI Connectivity and Network Modeling 2
Neuroinformatics and Data Sharing:
Brain Atlases
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Data analysis
FUNCTIONAL MRI
Machine Learning
Psychiatric Disorders
Schizophrenia
STRUCTURAL MRI
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. Wang, M. (2019). Identifying autism spectrum disorder with multi-site fMRI via low-rank domain adaptation. IEEE Transactions on Medical Imaging, 39(3), 644-655.
3. Glover, G.H. (2012). Function biomedical informatics research network recommendations for prospective multicenter functional MRI studies. Journal of Magnetic Resonance Imaging, 36(1), 39-54.
4. Tamminga, C.A. (2013). Clinical phenotypes of psychosis in the Bipolar-Schizophrenia Network on Intermediate Phenotypes (B-SNIP). American Journal of Psychiatry, 170(11), 1263-1274.
5. Tzourio-Mazoyer, N. (2002). Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. NeuroImage, 15(1), 273-289.
6. Fan, L. (2016). The human brainnetome atlas: a new brain atlas based on connectional architecture. Cerebral Cortex, 26(8), 3508-3526.
7. Yeo, B.T. (2011). The organization of the human cerebral cortex estimated by intrinsic functional connectivity. Journal of Neurophysiology, 106, 1125-1165.
8. Li, Y. (2021). Surface-based single-subject morphological brain networks: effects of morphological index, brain parcellation and similarity measure, sample size-varying stability and test-retest reliability. NeuroImage, 235, 118018.