Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NeuroKoop: Neural Koopman Fusion of Structural-Functional Connectomes for Identifying Prenatal Drug Exposure in Adolescents

Badhan Mazumder, Aline Kotoski, Vince D. Calhoun, Dong Hye Ye

TL;DR

NeuroKoop is introduced, a novel graph neural network-based framework that integrates structural and functional brain networks utilizing neural Koopman operator-driven latent space fusion and unifies node embeddings derived from source-based morphometry and functional network connectivity based brain graphs, resulting in enhanced representation learning and more robust classification of prenatal drug exposure (PDE) status.

Abstract

Understanding how prenatal exposure to psychoactive substances such as cannabis shapes adolescent brain organization remains a critical challenge, complicated by the complexity of multimodal neuroimaging data and the limitations of conventional analytic methods. Existing approaches often fail to fully capture the complementary features embedded within structural and functional connectomes, constraining both biological insight and predictive performance. To address this, we introduced NeuroKoop, a novel graph neural network-based framework that integrates structural and functional brain networks utilizing neural Koopman operator-driven latent space fusion. By leveraging Koopman theory, NeuroKoop unifies node embeddings derived from source-based morphometry (SBM) and functional network connectivity (FNC) based brain graphs, resulting in enhanced representation learning and more robust classification of prenatal drug exposure (PDE) status. Applied to a large adolescent cohort from the ABCD dataset, NeuroKoop outperformed relevant baselines and revealed salient structural-functional connections, advancing our understanding of the neurodevelopmental impact of PDE.

NeuroKoop: Neural Koopman Fusion of Structural-Functional Connectomes for Identifying Prenatal Drug Exposure in Adolescents

TL;DR

NeuroKoop is introduced, a novel graph neural network-based framework that integrates structural and functional brain networks utilizing neural Koopman operator-driven latent space fusion and unifies node embeddings derived from source-based morphometry and functional network connectivity based brain graphs, resulting in enhanced representation learning and more robust classification of prenatal drug exposure (PDE) status.

Abstract

Understanding how prenatal exposure to psychoactive substances such as cannabis shapes adolescent brain organization remains a critical challenge, complicated by the complexity of multimodal neuroimaging data and the limitations of conventional analytic methods. Existing approaches often fail to fully capture the complementary features embedded within structural and functional connectomes, constraining both biological insight and predictive performance. To address this, we introduced NeuroKoop, a novel graph neural network-based framework that integrates structural and functional brain networks utilizing neural Koopman operator-driven latent space fusion. By leveraging Koopman theory, NeuroKoop unifies node embeddings derived from source-based morphometry (SBM) and functional network connectivity (FNC) based brain graphs, resulting in enhanced representation learning and more robust classification of prenatal drug exposure (PDE) status. Applied to a large adolescent cohort from the ABCD dataset, NeuroKoop outperformed relevant baselines and revealed salient structural-functional connections, advancing our understanding of the neurodevelopmental impact of PDE.

Paper Structure

This paper contains 14 sections, 5 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Modality-Wise Graph Encoding
  4. Neural Koopman-Driven Multimodal Fusion
  5. Dynamic Latent Fusion via Neural Koopman Operator
  6. Joint Training Objective and PDE Classification
  7. Results and Discussion
  8. Dataset and Data Pre-processing
  9. Experimental Settings
  10. Quantitative Evaluation
  11. Comparative Evaluation with SOTA Methods
  12. Ablation Experiments
  13. Qualitative Evaluation
  14. Conclusions

Figures (2)

  • Figure 1: NeuroKoop overview: After obtaining SBM form sMRI and FNC from rs-fMRI, modality-specific GNN encoders generated node embeddings from corresponding structural and functional brain graphs, which were fused through cross-modal attention layer (CAL). The neural Koopman layer then dynamically refined the fused latent representation based on each individual’s working memory (WM) score, followed by global average pooling (GAP) to yield a subject-level vector for classifying prenatal drug exposure (PDE) status, while adversarial regularization encourages alignment with functional organization.
  • Figure 2: Axial depiction of group-averaged cross-modal attention, showing the top 3% strongest connections for both unexposed and exposed groups across seven brain networks from NeuroMark—subcortical (SCN), auditory (ADN), sensorimotor (SMN), visual (VSN), cognitive control (CON), default mode (DMN), and cerebellar (CBN) network. Edges within the same network are color-coded, while those linking different networks are shown in gray. Edge thickness is proportional to attention weight, highlighting the relative importance of each connection in the fused representation.