Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

D-CoRP: Differentiable Connectivity Refinement for Functional Brain Networks

Haoyu Hu, Hongrun Zhang, Chao Li

TL;DR

D-CoRP tackles edge-noise in functional brain networks derived from fMRI by introducing a differentiable connectivity-refinement plugin grounded in information-bottleneck theory. It learns a learnable edge mask $A_{re}=\mathbf{m}\odot\mathbf{A}$ via a distribution-estimator (GAT) and a differentiable sampling mechanism with a temperature-controlled Bernoulli-like relaxation. The framework offers two optimization targets, including an efficient variant, and demonstrates improved accuracy and stability when integrated with multiple GNN backbones across three brain datasets for brain age prediction. The approach is generalizable to other graph domains and emphasizes preserving informative modular connectivity while denoising spurious edges.

Abstract

Brain network is an important tool for understanding the brain, offering insights for scientific research and clinical diagnosis. Existing models for brain networks typically primarily focus on brain regions or overlook the complexity of brain connectivities. MRI-derived brain network data is commonly susceptible to connectivity noise, underscoring the necessity of incorporating connectivities into the modeling of brain networks. To address this gap, we introduce a differentiable module for refining brain connectivity. We develop the multivariate optimization based on information bottleneck theory to address the complexity of the brain network and filter noisy or redundant connections. Also, our method functions as a flexible plugin that is adaptable to most graph neural networks. Our extensive experimental results show that the proposed method can significantly improve the performance of various baseline models and outperform other state-of-the-art methods, indicating the effectiveness and generalizability of the proposed method in refining brain network connectivity. The code will be released for public availability.

D-CoRP: Differentiable Connectivity Refinement for Functional Brain Networks

TL;DR

D-CoRP tackles edge-noise in functional brain networks derived from fMRI by introducing a differentiable connectivity-refinement plugin grounded in information-bottleneck theory. It learns a learnable edge mask via a distribution-estimator (GAT) and a differentiable sampling mechanism with a temperature-controlled Bernoulli-like relaxation. The framework offers two optimization targets, including an efficient variant, and demonstrates improved accuracy and stability when integrated with multiple GNN backbones across three brain datasets for brain age prediction. The approach is generalizable to other graph domains and emphasizes preserving informative modular connectivity while denoising spurious edges.

Abstract

Brain network is an important tool for understanding the brain, offering insights for scientific research and clinical diagnosis. Existing models for brain networks typically primarily focus on brain regions or overlook the complexity of brain connectivities. MRI-derived brain network data is commonly susceptible to connectivity noise, underscoring the necessity of incorporating connectivities into the modeling of brain networks. To address this gap, we introduce a differentiable module for refining brain connectivity. We develop the multivariate optimization based on information bottleneck theory to address the complexity of the brain network and filter noisy or redundant connections. Also, our method functions as a flexible plugin that is adaptable to most graph neural networks. Our extensive experimental results show that the proposed method can significantly improve the performance of various baseline models and outperform other state-of-the-art methods, indicating the effectiveness and generalizability of the proposed method in refining brain network connectivity. The code will be released for public availability.
Paper Structure (12 sections, 7 equations, 3 figures, 2 tables)

This paper contains 12 sections, 7 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Overview
  4. Differentiable Connectivity Refinement
  5. Optimization with Information Bottleneck
  6. Experiments and Results
  7. Task and Datasets
  8. Experiment Settings and Evaluation Metrics
  9. Performance of D-CoRP on Baselines
  10. Comparison with Other Denoising Methods
  11. Hyperparameter Tuning on D-CoRP
  12. Conclusion

Figures (3)

  • Figure 1: Model architecture. D-CoRP refines the graph connection in two steps: 1) Differentiable sampling for the masking matrix (contains Distribution Estimator and Differentiable Sampling); 2) Information bottleneck optimization for the refined results as shown in IB Optimizer part. D-CoRP can be directly plugged into the graph model as a plugin.
  • Figure 2: Visualization of Refined Brain Network. One example of the brain networks refined by D-CoRP, compared to other denoising methods all based on GCN. (a): original brain network; (b): refined brain networks and their differences from the original network with various connectivity refinement methods.
  • Figure 3: Results of Variant Temperature.