Multi-Modal Graph Neural Network with Transformer-Guided Adaptive Diffusion for Preclinical Alzheimer Classification

Jaeyoon Sim1, Minjae Lee1, Guorong Wu2, Won Hwa Kim1
1POSTECH, 2UNC-Chapel Hill

MICCAI 2024

:canada: Marrakesh, Morocco

Paper Poster

Illustration of overall framework.

Abstract

The graphical representation of the brain offers critical insights into diagnosing and prognosing neurodegenerative disease via relationships between regions of interest (ROIs). Despite recent emergence of various Graph Neural Networks (GNNs) to effectively capture the relational information, there remain inherent limitations in interpreting the brain networks. Specifically, convolutional approaches ineffectively aggregate information from distant neighborhoods, while attention-based methods exhibit deficiencies in capturing node-centric information, particularly in retaining critical characteristics from pivotal nodes. These shortcomings reveal challenges for identifying disease-specific variation from diverse features from different modalities. In this regards, we propose an integrated framework guiding diffusion process at each node by a downstream transformer where both short- and long-range properties of graphs are aggregated via diffusion-kernel and multi-head attention respectively. We demonstrate the superiority of our model by improving performance of pre-clinical Alzheimer’s disease (AD) classification with various modalities. Also, our model adeptly identifies key ROIs that are closely associated with the preclinical stages of AD, marking a significant potential for early diagnosis and prevision of the disease.

Quantitative Results

Table: Preclinical AD classification performance (CN/SMC/EMCI) on ADNI data.

Discussion on the Scales

Figure: Visualization of learned scales on the cortical regions of left (top) and right (bottom) hemispheres.

Table: 8 Localized ROIs with the smallest trained scales for classification. (L) and (R) denote left and right hemisphere, respectively.

Figure: Visualization of learned scales on the cortical regions of a brain using three biomarkers such as cortical thickness (top), β-Amyloid (middle) and FDG (bottom).

Pre-clinical AD via ROI Attention

Figure: Distribution of attention scores across all brain regions with cortical thickness (left), β-Amyloid (center) and FDG (right).

Table: Corresponding ROIs with the 5 highest attention scores for classification. Importance Rate (IR) indicates how many ROIs pay attention. (L) and (R) denote left and right hemisphere, respectively.

Ablation Study on the Blocks

Table: Performance comparisons of different blocks. For attention block, our multimodal (MM) attention and existing position-wise attention are compared.

Conclusion

In this work, we proposed a novel end-to-end framework GTAD to dynamically define node-centric ranges per imaging modality via diffusion kernel, guided by a subsequent transformer. Our framework captures local characteristics on graphs by flexibly optimizing node-wise scales separately on imaging modalities, and obtains a global representation by employing multi-modal self-attention, which guides the model to better prediction. Leveraging multiple imaging measures, GTAD demonstrates superiority as evidenced by improved performance in preclinical AD classification, and the results identifies disease-specific variation through AD-specific key ROIs in the brain.

BibTeX

@article{sim2024multi,
  author    = {Jaeyoon Sim and Minjae Lee and Guorong Wu and Won Hwa Kim},
  title     = {Multi-Modal Graph Neural Network with Transformer-Guided Adaptive Diffusion for Preclinical Alzheimer Classification},
  journal   = {MICCAI},
  year      = {2024},
}