Abstract:The human brain is the central hub of the neurobiological system, controlling behavior and cognition in complex ways. Recent advances in neuroscience and neuroimaging analysis have shown a growing interest in the interactions between brain regions of interest (ROIs) and their impact on neural development and disorder diagnosis. As a powerful deep model for analyzing graph-structured data, Graph Neural Networks (GNNs) have been applied for brain network analysis. However, training deep models requires large amounts of labeled data, which is often scarce in brain network datasets due to the complexities of data acquisition and sharing restrictions. To make the most out of available training data, we propose PTGB, a GNN pre-training framework that captures intrinsic brain network structures, regardless of clinical outcomes, and is easily adaptable to various downstream tasks. PTGB comprises two key components: (1) an unsupervised pre-training technique designed specifically for brain networks, which enables learning from large-scale datasets without task-specific labels; (2) a data-driven parcellation atlas mapping pipeline that facilitates knowledge transfer across datasets with different ROI systems. Extensive evaluations using various GNN models have demonstrated the robust and superior performance of PTGB compared to baseline methods.

What problem does this paper attempt to address?

This paper attempts to address the issue of annotated data scarcity in brain network analysis due to the complexity of data acquisition and sharing restrictions. Specifically, the paper proposes a pre-training framework for graph neural networks named PTGB (Pre-Train Graph Neural Networks for Brain Network Analysis), aiming to improve model performance using limited annotated data. ### Main Issues 1. **Annotated Data Scarcity**: Brain network datasets are usually small in scale, and their acquisition and sharing are restricted, leading to overfitting during deep learning model training. 2. **Inconsistency Between Different Datasets**: The brain region parcellation systems (ROI parcellation systems) used by different research institutions are inconsistent, affecting the transferability of pre-trained models across different datasets. ### Solutions 1. **Unsupervised Pre-Training**: PTGB adopts an unsupervised pre-training method, capturing the intrinsic structure of brain networks from large-scale unannotated data through self-supervised learning. 2. **Two-Layer Contrastive Learning Strategy**: Combining data-driven brain atlas mapping technology, PTGB proposes a two-layer contrastive learning strategy to enhance cross-dataset learning ability. 3. **Data-Driven Brain Atlas Mapping**: By using autoencoders and specific regularization terms, features from different datasets are aligned into a unified embedding space, overcoming the inconsistency of different ROI systems. ### Specific Contributions 1. **Unsupervised Pre-Training Method**: Proposes an unsupervised pre-training method suitable for brain networks, addressing the issue of limited training resources. 2. **Two-Layer Contrastive Learning Strategy**: Designs a two-layer contrastive learning strategy specifically for brain networks, combined with data-driven brain atlas mapping technology, enhancing cross-dataset learning ability. 3. **Experimental Validation**: Validates the effectiveness of PTGB through comparative experiments with shallow and deep baseline models and conducts an in-depth analysis of the impact of each component. In summary, this paper effectively addresses the issues of data scarcity and inconsistency between different datasets in brain network analysis by proposing the PTGB framework, providing a new solution for brain network analysis.

PTGB: Pre-Train Graph Neural Networks for Brain Network Analysis

BrainGB: A Benchmark for Brain Network Analysis With Graph Neural Networks

BrainGNN: Interpretable Brain Graph Neural Network for fMRI Analysis

Interpretable Graph Neural Networks for Connectome-Based Brain Disorder Analysis

Deep reinforcement learning guided graph neural networks for brain network analysis

Neural Graph Matching for Pre-training Graph Neural Networks

Learning Task-Aware Effective Brain Connectivity for fMRI Analysis with Graph Neural Networks

Pre-training graph neural networks for link prediction in biomedical networks

PAGANI Toolkit: Parallel Graph-Theoretical Analysis Package for Brain Network Big Data

FBNETGEN: Task-aware GNN-based fMRI Analysis via Functional Brain Network Generation

Graph Neural Networks for Brain Graph Learning: A Survey

PT-KGNN: A framework for pre-training biomedical knowledge graphs with graph neural networks

BrainNPT: Pre-training Transformer Networks for Brain Network Classification

Graph Neural Networks in Network Neuroscience

Hi-GCN: A Hierarchical Graph Convolution Network for Graph Embedding Learning of Brain Network and Brain Disorders Prediction.

Toolkit : Parallel Computing Package for Fast Network Analyses of Brain

BrainNPT: Pre-training of Transformer networks for brain network classification

Strongly Topology-preserving GNNs for Brain Graph Super-resolution

TGNet: tensor-based graph convolutional networks for multimodal brain network analysis

Spatial-Temporal Dynamic Hypergraph Information Bottleneck for Brain Network Classification

Pre-training on Large-Scale Heterogeneous Graph