Joint cmICA: auto-linking of structural and functional connectivity

Lei Wu, Vince D. CalhounbioRxiv 2022 · DOI

TL;DR

This paper introduces "joint cmICA," a novel data-driven method that simultaneously analyzes structural and functional brain connectivity from fMRI and dMRI data. It automatically links brain regions and their connections, providing a more comprehensive understanding of brain networks than previous approaches.

What Problem Does This Paper Address?

Understanding how the brain works relies heavily on studying its connectivity – how different regions are structurally linked (e.g., by white matter tracts) and how they functionally interact (e.g., through synchronized activity). While both structural connectivity (SC) and functional connectivity (FC) offer crucial insights, integrating them into a single framework has been challenging. This is due to fundamental differences in how these connectivities are measured and represented, making it difficult to directly link them. Existing methods often require pre-defined regions of interest (ROIs) or process modalities separately, which can introduce bias and miss important cross-modal relationships. This paper aims to bridge this gap by providing a unified, data-driven approach to simultaneously analyze and link SC and FC.

Methods

The core of this work is the joint cmICA (connectivity matrix ICA) method. This approach extends the existing connectivity matrix ICA (cmICA) by integrating data from both diffusion MRI (dMRI) for structural connectivity and functional MRI (fMRI) for functional connectivity.

Here's a breakdown of the key methodological aspects:

  • Data-Driven Parcellation: Unlike methods that rely on predefined anatomical atlases, joint cmICA automatically identifies distinct brain regions (parcels) and their connectivity patterns directly from the data.
  • Simultaneous Analysis: The method processes SC and FC data together, allowing for the identification of "shared" sources that are present in both modalities.
  • Connectivity Matrix ICA (cmICA): This is the foundational technique. It takes a connectivity matrix (representing connections between all pairs of brain regions) as input and decomposes it into independent spatial sources and their corresponding connectivity maps.
  • Joint Integration: For joint cmICA, the SC and FC matrices are concatenated. This allows the ICA algorithm to find independent sources that are common to both structural and functional connectivity.
  • Enhanced Pipeline: The authors incorporated GPU computation and fast PCA for efficient processing of the large covariance matrices involved in whole-brain tractography.
  • Model Order Selection: They used ICASSO with multiple runs to ensure the stability and robustness of the chosen number of components (model order).
  • Feature Selection: A contribution ratio was calculated to ensure that selected components were genuinely "shared" between SC and FC, with contributions falling within a defined range (20-80%).

Datasets

The study utilized fMRI and diffusion MRI (dMRI) data from 60 healthy control subjects. This data was previously collected as part of the COINS platform and used in prior unimodal cmICA analyses. The authors specifically excluded patients to focus on general connectivity patterns.

Key Insights and Ideas

  • Unified View of Brain Networks: Joint cmICA offers a powerful way to see how structural pathways and functional interactions align (or misalign) within the brain, providing a more holistic picture than analyzing each separately.
  • Data-Driven Discovery: The method's ability to automatically identify brain regions and their connections without relying on pre-existing atlases is crucial. This can reveal novel organizational principles and avoid biases inherent in atlas-based approaches.
  • Linking Structure and Function: The core innovation is the direct linking of structural white matter tracts to functional connectivity networks. This allows researchers to understand how structural pathways support functional communication.
  • Common Cortical Segregation: The study found that both SC and FC modalities reveal similar patterns of brain region segregation, though with varying degrees of similarity. This suggests an underlying organizational principle shared by both types of connectivity.
  • Complementary Information: The research highlights that SC and FC provide complementary information, and integrating them can reveal insights that might be missed by analyzing either modality alone, particularly in detecting subtle differences in diseases.

Key Results

  • 28 Shared Sources Identified: After artifact removal and feature selection, 28 components (representing shared sources) were chosen for further analysis.
  • Matching of SC and FC Networks: The study demonstrated that the connectivity maps derived from SC (white matter tracts) and FC (functional networks) were well-matched for these shared sources.
  • Categorization of Tracts: The identified connections were grouped into five major categories based on white matter tracts: commissural fibers, inferior longitudinal/inferior frontal-occipital/uncinate fasciculus, superior longitudinal fasciculus, cingulate, and corticospinal tracts.
  • Detailed Mapping: Figures and tables provided detailed visualizations and MNI coordinates of these matched SC and FC networks, illustrating how specific white matter tracts connect specific functional brain regions. For example, commissural fibers were shown to connect cortical areas across hemispheres.
  • Contribution Imbalance: The study noted that while efforts were made to balance contributions from SC and FC, there was an observed bias, with some sources being more heavily influenced by SC and others by FC, likely due to differences in their spatial distributions.

Limitations and Future Work

  • Contribution Bias: The authors acknowledge that despite balancing PCA outputs, the ICA decomposition was not always equally contributed by SC and FC due to their inherent spatial distribution differences. This imbalance can lead to some sources being more influenced by one modality.
  • Computational Cost: Implementing joint cmICA, especially for SC tractography at a whole-brain voxel scale, is computationally intensive, requiring significant CPU/GPU resources.
  • Individual Subject Matching: While the study focused on group-level analysis, the authors suggest that investigating the specific correspondence and differences in source matching at the individual subject level is a promising future direction, potentially using dual model fitting.
  • Sophisticated Fusion Strategies: To further balance intrinsic data differences and maintain component quality, the authors propose exploring more sophisticated fusion strategies, such as constrained decomposition or leveraging non-linear relationships.
  • Model-Based Approaches: The paper mentions the potential for extending existing model-based work between SC and FC into joint cmICA.

Summary updated

Abstract

Abstract The study of human brain connectivity, including structural connectivity (SC) and functional connectivity (FC), provides insights into the neurophysiological mechanism of brain function and its relationship to human behavior and cognition. Both types of connectivity measurements provide crucial yet complementary information. However, integrating these two modalities into a single framework remains a challenge, because of the differences in their quantitative interdependencies as well as their anatomical representations due to distinctive imaging mechanisms. In this study, we introduced a new method, joint cmICA (connectivity matrix ICA), which provides a data-driven parcellation and automated-linking of SC and FC information simultaneously using a joint analysis of functional MRI and diffusion-weighted MRI data. We showed that these two connectivity modalities produce common cortical segregation, though with various degrees of (dis)similarity. Moreover, we show conjoint functional connectivity networks and structural white matter tracts that directly link these cortical parcellations/sources, within one analysis. Overall, data driven joint cmICA provides a new approach for integrating or fusing structural connectivity and functional connectivity systematically and conveniently, and provides an effective tool for connectivity-based multimodal data fusion in brain.
BibTeX
@misc{wu2022joint,
  author = {Lei Wu and Vince D. Calhoun},
  title = {{Joint cmICA: auto-linking of structural and functional connectivity}},
  year = {2022},
  doi = {10.1101/2022.09.12.507415},
  howpublished = {bioRxiv},
}
Figure