On the Limits of Applying Graph Transformers for Brain Connectome Classification
Jose Lara-Rangel, Clare Heinbaugh
TL;DR
The paper addresses whether graph transformers offer advantages over traditional GNNs for static brain connectome classification. It compares Exphormer against ResidualGCN on HCP-Activity, HCP-Gender, and HCP-Age, including synthetic edge-removal variants. Contrary to expectations, Exphormer does not surpass ResidualGCN, both models are robust to edge drop, and overfitting persists. The authors argue that current NeuroGraph graph structures may provide little predictive signal and call for better preprocessing pipelines and more curated benchmarks to extract meaningful brain-edge information.
Abstract
Brain connectomes offer detailed maps of neural connections within the brain. Recent studies have proposed novel connectome graph datasets and attempted to improve connectome classification by using graph deep learning. With recent advances demonstrating transformers' ability to model intricate relationships and outperform in various domains, this work explores their performance on the novel NeuroGraph benchmark datasets and synthetic variants derived from probabilistically removing edges to simulate noisy data. Our findings suggest that graph transformers offer no major advantage over traditional GNNs on this dataset. Furthermore, both traditional and transformer GNN models maintain accuracy even with all edges removed, suggesting that the dataset's graph structures may not significantly impact predictions. We propose further assessing NeuroGraph as a brain connectome benchmark, emphasizing the need for well-curated datasets and improved preprocessing strategies to obtain meaningful edge connections.