Individualized prediction of functional connectivity from structural connectivity
Poster No:
1184
Submission Type:
Abstract Submission
Authors:
Yiming Chen1, Weikang Gong2
Institutions:
1Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University, Shanghai, China, 2School of Data Science, Fudan University, Shanghai, China
First Author:
Yiming Chen
Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University
Shanghai, China
Institute of Science and Technology for Brain-Inspired Intelligence, Fudan University
Shanghai, China
Co-Author:
Introduction:
Understanding the relationship between structure and function of the human brain is an essential problem in neuroscience. However, building a reliable model that links individual brain structure and function is challenging, such as the relationship between structural connectivity (SC) and functional connectivity (FC).
Recent advances in deep learning have brought breakthroughs for the field of computational neuroimaging. The advantage of using deep learning models is that complex relations between brain structure and function can be uncovered, without preconditioning on any specific prior hypothesis.
Sarwar and his colleagues (Sarwar et al., 2021) are among the first to propose using deep learning models to predict functional connectivity from structural connectivity. Their work demonstrates that the deep learning model could capture structure-function coupling and surpass biophysical model. However, the proposed model does not guarantee that the predicted functional connectivity is a covariance matrix and they do not explore the limits of deep learning models in learning individual effects.
Although these deep learning models show promising potential in capturing structure-function coupling, the extent to which these models recapitulate individual-specific effects remains unclear. Our work adopts the transformer model (Vaswani et al. 2017) to predict the individual FC using the corresponding SC. By regressing the group mean in SC and FC, we explore the extent to which the model can learn the individual-specific structure-function pattern.
Recent advances in deep learning have brought breakthroughs for the field of computational neuroimaging. The advantage of using deep learning models is that complex relations between brain structure and function can be uncovered, without preconditioning on any specific prior hypothesis.
Sarwar and his colleagues (Sarwar et al., 2021) are among the first to propose using deep learning models to predict functional connectivity from structural connectivity. Their work demonstrates that the deep learning model could capture structure-function coupling and surpass biophysical model. However, the proposed model does not guarantee that the predicted functional connectivity is a covariance matrix and they do not explore the limits of deep learning models in learning individual effects.
Although these deep learning models show promising potential in capturing structure-function coupling, the extent to which these models recapitulate individual-specific effects remains unclear. Our work adopts the transformer model (Vaswani et al. 2017) to predict the individual FC using the corresponding SC. By regressing the group mean in SC and FC, we explore the extent to which the model can learn the individual-specific structure-function pattern.
Methods:
We use the Human Connectome Project (HCP) (Van Essen et al., 2013) dataset which contains 1096 healthy subjects. Each individual was scanned by a customized 3T scanner to obtain dMRI data. We obtained minimally pre-processed diffusion-weighted MRI data and construct SC matrices for each individual. The HCP participants were also scanned for their resting state fMRI, which were collected through a 15-minute scan for each encoding direction (left-to-right and right-to-left). We construct a FC matrix for each individual using PCC, computing for 4 scans respectively and then averaging them. Both SC and FC are constructed using the Schaefer 200 atlas (Schaefer et al., 2018) and the Tian Subcortex atlas (Tian et al., 2020).
The transformer model (Vaswani et al. 2017) has a profound impact on deep learning. The core concept of Transformer is the self-attention (SA) mechanisms, which allows the model to weight the importance of each token in a sequence while taking into account the relationships between tokens across different positions. Although initially designed for natural language processing (NLP), Transformer's strong sequence modeling capabilities have shown considerable potential in analyzing complex multimodal medical imaging data. Therefore, we adopt the transformer model to deal with brain imaging data and learn the mapping between SC and FC.
Aimed at focusing on the individual-specific effects in structure-function coupling, we regress the group mean in SC and FC. Besides, in order to maximize the pearson correlation between the predicted FC and SC, we follow Lu Zhang et al.(2022) and add the Pearson's correlation coefficient (PCC) loss in addition to the normally used mean squared error prediction loss.
The transformer model (Vaswani et al. 2017) has a profound impact on deep learning. The core concept of Transformer is the self-attention (SA) mechanisms, which allows the model to weight the importance of each token in a sequence while taking into account the relationships between tokens across different positions. Although initially designed for natural language processing (NLP), Transformer's strong sequence modeling capabilities have shown considerable potential in analyzing complex multimodal medical imaging data. Therefore, we adopt the transformer model to deal with brain imaging data and learn the mapping between SC and FC.
Aimed at focusing on the individual-specific effects in structure-function coupling, we regress the group mean in SC and FC. Besides, in order to maximize the pearson correlation between the predicted FC and SC, we follow Lu Zhang et al.(2022) and add the Pearson's correlation coefficient (PCC) loss in addition to the normally used mean squared error prediction loss.
·(a) Schema of study design. (b) Architecture of transformer encoder
Results:
We use the Pearson Correlation Coefficient (PCC) between the predicted FC and true FC under the condition of regressing group mean as the metric. Our proposed model reaches r=0.18. The PCC between different subjects' predicted FC is about 0.09, which demonstrates that the individual-specific structure-function coupling surpasses group average structure-function coupling under the condition of regressing group mean. The average regional PCC between predicted FC and true FC reaches 0.21.
·(a) Curve of average between-subject PCC of predicted FC. (b) Curve of average PCC between predicted FC and true FC on the test set. (c) Average regional PCC between predicted FC and true FC.
Conclusions:
Under the condition of regressing group mean of SC and FC, our proposed transformer-based model can recapitulate significant individual-specific structure-function coupling of the human brain.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
fMRI Connectivity and Network Modeling 2
Novel Imaging Acquisition Methods:
Multi-Modal Imaging
Keywords:
FUNCTIONAL MRI
Machine Learning
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.
Vaswani, A. (2017). Attention is all you need. Advances in Neural Information Processing Systems.
Zhang, L., Wang, L., Zhu, D., & Alzheimer's Disease Neuroimaging Initiative. (2022). Predicting brain structural network using functional connectivity. Medical image analysis, 79, 102463.
Van Essen, D. C., Smith, S. M., Barch, D. M., Behrens, T. E., Yacoub, E., Ugurbil, K., & Wu-Minn HCP Consortium. (2013). The WU-Minn human connectome project: an overview. Neuroimage, 80, 62-79.
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.
Tian, Y., Margulies, D. S., Breakspear, M., & Zalesky, A. (2020). Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nature neuroscience, 23(11), 1421-1432.