A Cascaded Transformer Network for fMRI Feature Encoding in AD Diagnosis
Poster No:
1127
Submission Type:
Abstract Submission
Authors:
Yanteng Zhang1, Anees Abrol1, Vince Calhoun2
Institutions:
1Center for Translational Research in Neuroimaging and Data Science(GSU, GeorgiaTech, Emory), Atlanta, GA, 2GSU/GATech/Emory, Atlanta, GA
First Author:
Yanteng Zhang
Center for Translational Research in Neuroimaging and Data Science(GSU, GeorgiaTech, Emory)
Atlanta, GA
Center for Translational Research in Neuroimaging and Data Science(GSU, GeorgiaTech, Emory)
Atlanta, GA
Co-Author(s):
Anees Abrol
Center for Translational Research in Neuroimaging and Data Science(GSU, GeorgiaTech, Emory)
Atlanta, GA
Center for Translational Research in Neuroimaging and Data Science(GSU, GeorgiaTech, Emory)
Atlanta, GA
Introduction:
Accurately representing brain imaging features is a core challenge in achieving effective brain disease diagnosis. fMRI, capturing functional connectivity changes between brain regions, which can reveal brain activity critical for the auxiliary diagnosis of cognitive diseases. However, fMRI data faces challenges such as significant physiological noise interference and low spatial resolution, which limits its ability to localize pathological structures. Furthermore, as a four-dimensional spatiotemporal imaging modality, fMRI data is characterized by high dimensionality and complex dynamic signals, making processing and analysis computationally intensive and technically challenging, hindering its widespread application. To address these issues, we propose a Cascaded Transformer network for fMRI feature encoding. The network captures temporal dependencies within functional network connectivity (FNC) and effectively extracts representative dynamic functional features. Compared to traditional deep learning models, the proposed method achieves more accurate fMRI feature representation and demonstrates superior performance in Alzheimer's Disease (AD) diagnosis tasks.
Methods:
For the processing of 4D fMRI data, Independent Component Analysis is first employed to extract features from the 3D fMRI images at each time point, resulting in 53 independent components and forming FNC. Subsequently, a sliding window is applied to construct dynamic FNC matrices. This process generates 448 functional connectivity matrices for each subject, with each matrix having dimensions of 53×53. Next, the upper triangular elements of each FNC matrix are flattened to obtain 448 features, each with 1378 dimensions. To achieve representation of the fMRI features, the 1378-dimensional features are further compressed to 128 dimensions using 1D adaptive GAP. Finally, the serially constructed Transformer modules are employed to encode these features, resulting in a 128-dimensional feature tensor. This processing pipeline effectively captures both the temporal dynamics and spatial functional connectivity of fMRI data, providing an efficient and compact feature representation for subsequent analysis and modeling. The encoded fMRI features are ultimately passed through a two-layer MLP for classification, enabling AD recognition.
Results:
Our model was evaluated on the baseline fMRI scans of 104 subjects from the ADNI, which includes 30 AD and 74 NC subjects. For the AD vs. NC task, our model achieved an ACC of 0.80 and an AUC of 0.76, outperforming the FCN matrix based 2D CNN method with ACC 0.7 and AUC 0.55 and the FNC feature based LSTM method with ACC 0.7 and AUC 0.5. Furthermore, we present the FNC difference matrices between AD and NC subjects after fMRI feature encoding. As shown in figure, cascaded Transformer (CTransformer) demonstrates a stronger capability in dynamic relationship modeling, as evidenced by the more pronounced global distribution in its difference matrix. In contrast, the CNN method primarily focuses on local feature modeling. And the visualization of the difference matrix highlights that the Temporal, Lingual, and Hippocampus are the significant regions distinguishing AD from NC. This result indirectly validates the superior performance of the CTransformer in AD vs. NC task.
Conclusions:
Through the comparison of results and visualizations in the AD task, the superiority and effectiveness of our proposed cascaded Transformer network in fMRI feature encoding have been validated. The cascaded Transformer network efficiently captures the dynamic temporal features of fMRI data and the complex dependencies within functional network connectivity. Moreover, it can provide an intuitive representation of potential associations with AD pathological changes.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 2
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
Keywords:
FUNCTIONAL MRI
Machine Learning
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?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
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. Vaswani Ashish, et al. (2017). Attention Is All You Need. The Proceeding of NIPS17, p6000-6010.
3. Jack Clifford R, et al. (2010). The Alzheimer's Disease Neuroimaging Initiative (ADNI): MRI methods. Journal of Magnetic Resonance Imaging 27.4:685-691.
4. Krizhevsky Alex, et al. (2012). ImageNet Classification with Deep Convolutional Neural Networks. The Proceeding of NIPS12, Curran Associates Inc.
5. Zhang Li, et al. (2020). A survey on deep learning for neuroimaging-based brain disorder analysis. Frontiers in Neuroscience, 14.