Community-Aware Transformer for Autism Prediction in fMRI Connectome

Anushree Bannadabhavi,Soojin Lee,Wenlong Deng,Xiaoxiao Li
2023-06-25
Abstract:Autism spectrum disorder(ASD) is a lifelong neurodevelopmental condition that affects social communication and behavior. Investigating functional magnetic resonance imaging (fMRI)-based brain functional connectome can aid in the understanding and diagnosis of ASD, leading to more effective treatments. The brain is modeled as a network of brain Regions of Interest (ROIs), and ROIs form communities and knowledge of these communities is crucial for ASD diagnosis. On the one hand, Transformer-based models have proven to be highly effective across several tasks, including fMRI connectome analysis to learn useful representations of ROIs. On the other hand, existing transformer-based models treat all ROIs equally and overlook the impact of community-specific associations when learning node embeddings. To fill this gap, we propose a novel method, Com-BrainTF, a hierarchical local-global transformer architecture that learns intra and inter-community aware node embeddings for ASD prediction task. Furthermore, we avoid over-parameterization by sharing the local transformer parameters for different communities but optimize unique learnable prompt tokens for each community. Our model outperforms state-of-the-art (SOTA) architecture on ABIDE dataset and has high interpretability, evident from the attention module. Our code is available at <a class="link-external link-https" href="https://github.com/ubc-tea/Com-BrainTF" rel="external noopener nofollow">this https URL</a>.
Neurons and Cognition,Machine Learning,Image and Video Processing
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve This paper aims to predict Autism Spectrum Disorder (ASD) through functional magnetic resonance imaging (fMRI) connectomics. Specifically, the paper proposes a novel method—Community-aware Transformer (Com-BrainTF)—for predicting ASD in fMRI connectomics. #### Main Contributions: 1. **Proposing a novel local-global hierarchical transformer architecture**: By leveraging information at the ROI level and community level, it effectively learns and integrates community-aware ROI embeddings, thereby improving the accuracy of fMRI brain connectomics analysis. 2. **Avoiding over-parameterization**: By sharing local transformer parameters and designing personalized learnable prompt tokens for each community. 3. **Demonstrating the effectiveness of Com-BrainTF**: Quantitative and qualitative experimental results validate the model's effectiveness, with visualizations showing the model's ability to capture functional community patterns, which are crucial for distinguishing ASD from healthy controls (HC). ### Research Background - ASD is a lifelong neurodevelopmental disorder that affects social communication and behavior. - Traditional fMRI data analysis methods construct functional connectivity matrices, but existing deep learning models often overlook functional community information. - The proposed method, Com-BrainTF, combines functional community information and improves ASD prediction accuracy through a multi-level transformer structure. ### Method Overview - **Local-Global Transformer Encoder**: Includes a local transformer and a global transformer. The local transformer learns community-specific embeddings, while the global transformer integrates information from the entire brain. - **Graph Readout Layer**: Aggregates global-level node embeddings to obtain a high-level representation of the brain graph. ### Experimental Results - **Baseline Comparison**: Com-BrainTF outperforms other models across multiple evaluation metrics, particularly excelling in AUROC, accuracy, and sensitivity. - **Interpretability Analysis**: Com-BrainTF correctly identifies functional communities related to ASD, and the results are consistent with findings in neuroscience research. ### Conclusion Com-BrainTF not only performs excellently on the ABIDE dataset but also has high interpretability, capable of detecting significant functional networks related to ASD. Future work will explore different brain parcellation methods.