Multi-Scale Analysis Framework Integrating Brain Functional Connectivity and Activity
Poster No:
1766
Submission Type:
Abstract Submission
Authors:
Xiangxiang Cui1, Min Zhao2, Dongmei Zhi1, Weizheng Yan3, Vince Calhoun4, Chuanjun Zhuo5, Jing Sui6
Institutions:
1Beijing Normal University, Beijing, Beijing, 2Institute of Automation, Chinese Academy of Sciences, Beijing, China, Beijing, Beijing, 3Lab of Neuroimaging, National Institutes of Health, Bethesda, MD, 4GSU/GATech/Emory, Decatur, GA, 5Tianjin Mental Health Center, Nankai University Affiliated Anding Hospital, Beijing, Tianjin, 6Beijing Normal University, Beijing, China
First Author:
Co-Author(s):
Chuanjun Zhuo
Tianjin Mental Health Center, Nankai University Affiliated Anding Hospital
Beijing, Tianjin
Tianjin Mental Health Center, Nankai University Affiliated Anding Hospital
Beijing, Tianjin
Introduction:
In the field of psychiatric diagnosis, despite some research advancements, challenges persist in accurately classifying and understanding mental illnesses. Notably, current methodologies have limitations in capturing and analyzing the complexities of brain functional networks, especially in handling the multiscale spatiotemporal features of brain activity. Our study aims to overcome these limitations by utilizing multiscale information to enhance the accuracy of psychiatric disorder classification and to delve deeper into the role of spatiotemporal characteristics of brain functional networks in disease identification.
Methods:
Our research methodology integrates strategies of both multiscale modeling and multiscale features. 1) Firstly, in terms of feature multiscale, we conducted variable step-length analysis on time series (TC) to extract functional network connectivity (FNC) and dynamic functional network connectivity (dFNC), revealing the brain network's activity characteristics across different temporal scales. Additionally, through multiscale entropy analysis of TC features, we further uncovered the hierarchical organization patterns of brain network states. 2) Secondly, in terms of model multiscale, we first proposed Neural Connection Search (NCS) to optimize network connections, and applied multiscale dilated convolution to further enhance network architecture. This approach not only mimics the brain's long and short connection mechanisms but also significantly improves the performance of psychiatric disorder classification.
Results:
In in-house dataset, our method achieved an accuracy rate (ACC) of 87.9%, surpassing advanced algorithms. Our method also realized a specificity (SPE) of 88.7%, indicating its robust capability in accurately identifying negative samples. Moreover, our method excelled in sensitivity (SEN) and F1 score, reaching 87.3% and 88.0%, respectively, further proving its ability to maintain high sensitivity for positive samples while also ensuring high precision. These results consistently indicate that our method provides more accurate and reliable predictions when dealing with complex datasets, which is essential for practical applications. Overall, by integrating traditional features (FNC) with time series characteristics (TCs) and their derived features (dFNC), our multi-scale analysis method not only stands out in individual metrics but also exhibits balanced and superior performance across all evaluated metrics.
Conclusions:
By combining multiscale feature analysis and multiscale modeling, our study effectively analyzes the spatiotemporal features of brain functional networks, thereby enhancing the accuracy of psychiatric disorder classification. These findings represent a important technical advancement and offer a new perspective for a deeper understanding of the complex dynamics of brain networks, holding substantial scientific significance and practical value for the diagnosis of mental illnesses.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling 1
Methods Development
Keywords:
Design and Analysis
FUNCTIONAL MRI
Modeling
Psychiatric Disorders
Other - deep learning, Brain connectivity and activity, Transformer
1|2Indicates the priority used for review
Provide references using author date format
2. Yan, Weizheng, et al. "Discriminating schizophrenia using recurrent neural network applied on time courses of multi-site FMRI data." EBioMedicine 47 (2019): 543-552.
3. Yan, Weizheng, et al. "Deep chronnectome learning via full bidirectional long short-term memory networks for MCI diagnosis." Medical Image Computing and Computer Assisted Intervention–MICCAI 2018: 21st International Conference, Granada, Spain, September 16-20, 2018, Proceedings, Part III 11. Springer International Publishing, 2018.
4. Zhao, Min, et al. "An attention-based hybrid deep learning framework integrating brain connectivity and activity of resting-state functional MRI data." Medical image analysis 78 (2022): 102413.
5. Zoph, Barret, , et al."Neural architecture search with reinforcement learning." arXiv preprint arXiv:1611.01578 (2016).
6. Zhong, Zhao, et al. "Practical block-wise neural network architecture generation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.