Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Causal Inference with Graph Neural Networks for Alzheimer's Disease Analysis

Pranay Kumar Peddi, Dhrubajyoti Ghosh

TL;DR

The paper addresses the challenge of distinguishing causal brain-region influences on Alzheimer's disease from confounding factors by presenting Causal-GCN, a graph-convolutional network augmented with do-calculus back-door adjustment. It defines and estimates average causal effects for each ROI via do-interventions, incorporating PCA-based adjustment for covariates such as age, sex, and APOE4, and uses bootstrap uncertainty to rank regions by their causal impact on AD probability. On the ADNI-derived dataset (484 subjects, 62 ROIs), Causal-GCN achieves performance comparable to baseline GNNs (AUC ≈ 0.535) while offering interpretable causal ROI rankings that highlight posterior, cingulate, and insular hubs, consistent with known AD pathology. This approach provides mechanistic insights and a principled framework to identify regions whose perturbation would influence disease progression, advancing interpretability and potential therapeutic targeting in neurodegenerative research.

Abstract

Deep graph learning has advanced Alzheimer's (AD) disease classification from MRI, but most models remain correlational, confounding demographic and genetic factors with disease specific features. We present Causal-GCN, an interventional graph convolutional framework that integrates do-calculus-based back-door adjustment to identify brain regions exerting stable causal influence on AD progression. Each subject's MRI is represented as a structural connectome where nodes denote cortical and subcortical regions and edges encode anatomical connectivity. Confounders such as age, sec, and APOE4 genotype are summarized via principal components and included in the causal adjustment set. After training, interventions on individual regions are simulated by serving their incoming edges and altering node features to estimate average causal effects on disease probability. Applied to 484 subjects from the ADNI cohort, Causal-GCN achieves performance comparable to baseline GNNs while providing interpretable causal effect rankings that highlight posterior, cingulate, and insular hubs consistent with established AD neuropathology.

Integrating Causal Inference with Graph Neural Networks for Alzheimer's Disease Analysis

TL;DR

The paper addresses the challenge of distinguishing causal brain-region influences on Alzheimer's disease from confounding factors by presenting Causal-GCN, a graph-convolutional network augmented with do-calculus back-door adjustment. It defines and estimates average causal effects for each ROI via do-interventions, incorporating PCA-based adjustment for covariates such as age, sex, and APOE4, and uses bootstrap uncertainty to rank regions by their causal impact on AD probability. On the ADNI-derived dataset (484 subjects, 62 ROIs), Causal-GCN achieves performance comparable to baseline GNNs (AUC ≈ 0.535) while offering interpretable causal ROI rankings that highlight posterior, cingulate, and insular hubs, consistent with known AD pathology. This approach provides mechanistic insights and a principled framework to identify regions whose perturbation would influence disease progression, advancing interpretability and potential therapeutic targeting in neurodegenerative research.

Abstract

Deep graph learning has advanced Alzheimer's (AD) disease classification from MRI, but most models remain correlational, confounding demographic and genetic factors with disease specific features. We present Causal-GCN, an interventional graph convolutional framework that integrates do-calculus-based back-door adjustment to identify brain regions exerting stable causal influence on AD progression. Each subject's MRI is represented as a structural connectome where nodes denote cortical and subcortical regions and edges encode anatomical connectivity. Confounders such as age, sec, and APOE4 genotype are summarized via principal components and included in the causal adjustment set. After training, interventions on individual regions are simulated by serving their incoming edges and altering node features to estimate average causal effects on disease probability. Applied to 484 subjects from the ADNI cohort, Causal-GCN achieves performance comparable to baseline GNNs while providing interpretable causal effect rankings that highlight posterior, cingulate, and insular hubs consistent with established AD neuropathology.

Paper Structure

This paper contains 14 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Method
  3. Overview
  4. Data Representation
  5. Graph Convolutional Network with Covariates
  6. Back-Door Adjustment using Principal Components
  7. Implementation of the Intervention Operator
  8. Estimation of Causal Effects and Statistical Inference
  9. Results
  10. Dataset and Preprocessing
  11. Model Performance
  12. Causal ROI Identification
  13. Uncertainty and Robustness
  14. Discussion

Figures (3)

  • Figure 1: Representative T1-weighted MRI slices showing cortical and subcortical segmentations for cognitively normal (CN), mild cognitive impairment (MCI), and Alzheimer’s dementia (AD) subjects. Progressive cortical thinning and ventricular enlargement are visible from CN to AD.
  • Figure 2: Causal-GCN framework illustrating the data input, model architecture, and causal inference steps including do-intervention and bootstrap-based ROI ranking.
  • Figure 3: Fold-wise causal effects $\Delta_j^{(\mathrm{AD})}$ for each ROI across five cross-validation folds. Error bars denote 95% bootstrap confidence intervals.