The field of neuroscience is witnessing a significant shift towards graph-based learning, with a focus on developing innovative models that can effectively capture the complex relationships between different brain regions. Recent studies have demonstrated the potential of graph neural networks (GNNs) in analyzing brain connectivity and predicting disease progression. Notably, GNNs have been used to identify key brain circuit abnormalities associated with Alzheimer's disease and Parkinson's disease, offering new insights into the neurobiological mechanisms underlying these conditions. Furthermore, graph-based learning has been applied to various tasks, including EEG-based visual decoding, functional connectome extraction, and brain graph learning. The use of graph-based models has shown promising results, outperforming traditional methods in several cases.

Some noteworthy papers in this area include: Uncovering Alzheimer's Disease Progression via SDE-based Spatio-Temporal Graph Deep Learning on Longitudinal Brain Networks, which proposes a novel framework for predicting AD progression using spatio-temporal graph neural networks. Brain PathoGraph Learning, which introduces a lightweight model for efficient brain graph learning by pathological pattern filtering and pathological feature distillation. PD-Diag-Net, which presents an end-to-end automated diagnostic method for Parkinson's disease using clinical-priors guided network on brain MRI. TRACE, which introduces a new paradigm for learning to compute on graphs, built on an architecturally sound backbone and a principled learning objective. VarCoNet, which employs self-supervised contrastive learning to exploit inherent functional inter-individual variability in brain function, serving as a brain function encoder that generates functional connectome embeddings.