Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Lesion-aware Edge-based Graph Neural Network for Predicting Language Ability in Patients with Post-stroke Aphasia

Zijian Chen, Maria Varkanitsa, Prakash Ishwar, Janusz Konrad, Margrit Betke, Swathi Kiran, Archana Venkataraman

TL;DR

The paper tackles predicting language ability in post-stroke aphasia from resting-state fMRI by incorporating lesion information into an edge-based graph neural network. It introduces LEGNet with three components—edge-based learning, a lesion encoding module, and subgraph learning—and leverages synthetic lesion augmentation from the Human Connectome Project to pretrain the model. Results on two in-house datasets show LEGNet outperforms baselines and generalizes across protocols, with interpretable subgraphs aligning to language-related regions. The approach offers a data-efficient pathway for lesion-aware network modeling and improved evaluation of language impairment after stroke.

Abstract

We propose a lesion-aware graph neural network (LEGNet) to predict language ability from resting-state fMRI (rs-fMRI) connectivity in patients with post-stroke aphasia. Our model integrates three components: an edge-based learning module that encodes functional connectivity between brain regions, a lesion encoding module, and a subgraph learning module that leverages functional similarities for prediction. We use synthetic data derived from the Human Connectome Project (HCP) for hyperparameter tuning and model pretraining. We then evaluate the performance using repeated 10-fold cross-validation on an in-house neuroimaging dataset of post-stroke aphasia. Our results demonstrate that LEGNet outperforms baseline deep learning methods in predicting language ability. LEGNet also exhibits superior generalization ability when tested on a second in-house dataset that was acquired under a slightly different neuroimaging protocol. Taken together, the results of this study highlight the potential of LEGNet in effectively learning the relationships between rs-fMRI connectivity and language ability in a patient cohort with brain lesions for improved post-stroke aphasia evaluation.

A Lesion-aware Edge-based Graph Neural Network for Predicting Language Ability in Patients with Post-stroke Aphasia

TL;DR

The paper tackles predicting language ability in post-stroke aphasia from resting-state fMRI by incorporating lesion information into an edge-based graph neural network. It introduces LEGNet with three components—edge-based learning, a lesion encoding module, and subgraph learning—and leverages synthetic lesion augmentation from the Human Connectome Project to pretrain the model. Results on two in-house datasets show LEGNet outperforms baselines and generalizes across protocols, with interpretable subgraphs aligning to language-related regions. The approach offers a data-efficient pathway for lesion-aware network modeling and improved evaluation of language impairment after stroke.

Abstract

We propose a lesion-aware graph neural network (LEGNet) to predict language ability from resting-state fMRI (rs-fMRI) connectivity in patients with post-stroke aphasia. Our model integrates three components: an edge-based learning module that encodes functional connectivity between brain regions, a lesion encoding module, and a subgraph learning module that leverages functional similarities for prediction. We use synthetic data derived from the Human Connectome Project (HCP) for hyperparameter tuning and model pretraining. We then evaluate the performance using repeated 10-fold cross-validation on an in-house neuroimaging dataset of post-stroke aphasia. Our results demonstrate that LEGNet outperforms baseline deep learning methods in predicting language ability. LEGNet also exhibits superior generalization ability when tested on a second in-house dataset that was acquired under a slightly different neuroimaging protocol. Taken together, the results of this study highlight the potential of LEGNet in effectively learning the relationships between rs-fMRI connectivity and language ability in a patient cohort with brain lesions for improved post-stroke aphasia evaluation.
Paper Structure (10 sections, 6 equations, 3 figures, 2 tables)

This paper contains 10 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Lesion-Aware GNN with Simulated Training Data
  3. LEGNet Model Architecture
  4. Synthetic Data Generation for Model Pre-Training
  5. Implementation Details
  6. Baseline Models
  7. Experimental Results
  8. Datasets of Post-Stroke Aphasia
  9. Performance Characterization and Model Interpretation
  10. Conclusion

Figures (3)

  • Figure 1: The Lesion-aware Edge-based BrainGNN Model. Top: Edge-to-edge message passing and edge-to-node aggregation. Bottom Left: Patient-specific lesion size and position encoding. Bottom Right: Subgraph updating and language prediction.
  • Figure 2: Synthetic data generation workflow. (a) Artificial lesions are created to lie within a single arterial territory. (b) Language score density is adjusted after the lesion augmentation. (c) Functional connectivity is corrupted in the lesioned area.
  • Figure 3: Left: Top two subgraphs in in DS-1. Right: The $z-$score of each subgraph from Neurosynth yarkoni2011large indicating the association strength with a particular function.