Neurodegenerative Brain Network Classification via Adaptive Diffusion with Temporal Regularization

Hyuna Cho1, Jaeyoon Sim1, Guorong Wu2, Won Hwa Kim1
1POSTECH, 2UNC-Chapel Hill

ICML 2024

:greece: Vienna, Austria

Paper Poster

Overview of architecture for neurodegenerative brain network classification (AGT).

Abstract

Analysis of neurodegenerative diseases on brain connectomes is important in facilitating early diagnosis and predicting its onset. However, investigation of the progressive and irreversible dynamics of these diseases remains underexplored in cross-sectional studies as its diagnostic groups are considered independent. Also, as in many real-world graphs, brain networks exhibit intricate structures with both homophily and heterophily. To address these challenges, we propose Adaptive Graph diffusion network with Temporal regularization (AGT). AGT introduces node-wise convolution to adaptively capture low (i.e., homophily) and high-frequency (i.e., heterophily) characteristics within an optimally tailored range for each node. Moreover, AGT captures sequential variations within progressive diagnostic groups with a novel temporal regularization, considering the relative feature distance between the groups in the latent space. As a result, our proposed model yields interpretable results at both node-level and group-level. The superiority of our method is validated on two neurodegenerative disease benchmarks for graph classification: Alzheimer’s Disease Neuroimaging Initiative (ADNI) and Parkinson’s Progression Markers Initiative (PPMI) datasets.

Node-level Scale Adaptation

Figure: Comparison between global filters and an adaptive local filter. The global filter refers to the conventional filters whose range of feature aggregation is identical across the whole nodes. (a) A low-pass global filter smooths out features among neighboring nodes, (b) while a high-pass global filter accentuates the difference between a target node and its neighboring nodes. (c) Unlike these global filters, the adaptive local filter captures optimal bandwidths for each node with trainable node-wise scaling parameters.

Quantitative Results

Table: Classification performance on the ADNI and PPMI datasets with 5-fold cross-validation. The best results are marked in bold and the second-best results are indicated by an underline.

ROI Analysis

Table: Five brain regions with the smallest trained scales for the AD group in the ADNI dataset and the PD group in the PPMI dataset. The regional scales were averaged across 5 trained models from 5 folds.

Temporal Analysis

Figure: Five brain regions with the smallest trained scales for the AD group in the ADNI dataset and the PD group in the PPMI dataset. The regional scales were averaged across 5 trained models from 5 folds.

Figure: Visualization of the trained scales on the ADNI-FDG experiment. Top: Trained scales of right cortex ROIs. Bottom: Trained scales of left cortex ROIs. All node-wise scales are averaged across 5 folds.

ROI Analysis

Figure: (a) and (c) are from a brain network (i.e., X and A) of subject #3130. Edges in (b) and (d) are the adaptive wavelet basis derived from a trained model. The node features in (b) become more discriminative from (a), while those in (d) become more similar compared to the nodes in (c).

Conclusion

We presented a novel wavelet-based GNN addressing challenges in analyzing the evolving dynamics of neurodegenerative diseases on brain connectomes. Our method captures the sequential variations within diagnostic groups with a group-wise temporal regularization. Also, it adaptively captures both homophily and heterophily within a graph by adjusting node-wise spectral bandwidth. As a result, our method outperformed various GNNs for brain connectome classification on two representative neurodegenerative disease benchmarks. Our framework offers clinical interpretability, showing significant potential to be applied in the analysis of various neurodegenerative diseases.

BibTeX

@inproceedings{choneurodegenerative,
      title={Neurodegenerative Brain Network Classification via Adaptive Diffusion with Temporal Regularization},
      author={Cho, Hyuna and Sim, Jaeyoon and Wu, Guorong and Kim, Won Hwa},
      booktitle={Forty-first International Conference on Machine Learning}
    }