Structurally Constrained Effective Brain Connectivity

Alessandro Crimi, Luca Dodero, Fabio Sambataro, Vittorio Murino, Diego Sona

TL;DR

This paper introduces a novel "Structurally Constrained Autoregressive Model" (CMAR) that fuses structural and functional brain connectivity data to derive a more accurate representation of "effective" connectivity. This method offers a data-driven approach to understanding how brain structure influences function, outperforming existing methods like Dynamic Causal Modeling (DCM) in certain benchmarks and showing promise for identifying biomarkers.

What Problem Does This Paper Address?

Understanding the intricate relationship between the brain's physical structure (how regions are wired together) and its function (how these regions communicate and process information) is a fundamental challenge in neuroscience. While structural connectivity (e.g., from diffusion MRI) and functional connectivity (e.g., from fMRI time series correlations) are well-studied, bridging the gap to "effective connectivity" – which describes the causal influence one neural system exerts over another – remains difficult. Existing methods often struggle with the complexity of whole-brain analysis, require extensive manual tuning, or produce results that don't fully align with known ground-truth. This paper aims to develop a more robust and interpretable method for inferring effective connectivity by leveraging structural information as a prior.

Methods

The core of this work is the Structurally Constrained Autoregressive Model (CMAR). It builds upon the framework of Multivariate Autoregressive (MAR) models, which describe how time series data (like fMRI signals) depend on their past values. The key innovation is the constraint: the MAR model's parameters are forced to only consider connections that exist in the initial structural connectivity matrix. This is achieved by using the structural connectivity as a bias, effectively "sparsifying" the functional model to only fit existing structural pathways.

The model can also incorporate indirect connections, extending beyond direct physical links to capture multi-step influences, inspired by neural field theory. For Granger causality analysis, the authors propose a method that uses the CMAR to estimate effective connectivity, which is then used to quantify causal influences. They also explore using deconvoluted BOLD signals (representing neural activity more directly) to improve the model's performance.

Datasets

The authors utilized several publicly available datasets for their experiments:

  • Nathan Kline Institute-Rockland dataset: This large dataset (200 healthy subjects) was used for simulations, Default Mode Network (DMN) analysis, and community detection tasks. It includes resting-state fMRI (rs-fMRI), diffusion-weighted imaging (DWI), and T1-weighted MRI data.
  • Autism Brain Imaging Data Exchange II (ABIDE-II) dataset: This dataset (26 ASD and 21 TD subjects) was used for the case-control classification task to differentiate individuals with autism spectrum disorder from typically developing individuals. It also includes rs-fMRI, DWI, and T1-weighted MRI data.

Key Insights and Ideas

  • Structure as a Prior for Function: The paper demonstrates that using structural connectivity as a prior constraint significantly improves the estimation of effective connectivity, leading to a sparser and more biologically plausible representation.
  • Beyond Direct Connections: Incorporating indirect connections, while computationally more intensive, further refines the effective connectivity estimation and reduces reconstruction error, suggesting that multi-step influences are important.
  • Data-Driven vs. Model-Driven: The CMAR approach is data-driven, allowing it to adapt to individual brain data, whereas methods like DCM often require extensive prior specification and tuning, making them less flexible for exploratory analysis.
  • Effective Connectivity for Biomarkers: The improved discrimination in the ASD vs. TD classification task suggests that effective connectivity derived from CMAR can capture subtle differences relevant for identifying biomarkers.
  • Community Detection Enhancement: Using effective connectivity for community detection resulted in clusters that better reflected functional interactions while preserving the underlying structural organization.

Key Results

  • Improved Ground-Truth Matching: The CMAR model showed a smaller difference from ground-truth causal inferences in simulations compared to DCM.
  • DMN Analysis: The model successfully identified relevant DMN connections without requiring extensive prior specification, unlike DCM.
  • ASD Classification: The classification accuracy for distinguishing ASD from TD subjects was higher when using effective connectivity features (up to 72.34% with MAR(2)) compared to structural (60.57%) or functional (69.1%) connectivity alone.
  • Community Detection: Effective connectivity-based clustering resulted in sparser and more functionally meaningful communities compared to structural clustering.
  • Reconstruction Error: Using indirect connections generally led to lower reconstruction errors in predicting functional signals. Deconvolving BOLD signals also further improved reconstruction.

Limitations and Future Work

  • Temporal Resolution: The authors acknowledge that the limited temporal resolution of BOLD signals is a fundamental constraint, likely limiting the optimal autoregressive order to low values (e.g., 2).
  • Computational Complexity: While the CMAR is more efficient than some alternatives for whole-brain analysis, incorporating indirect connections and extensive simulations still requires significant computational resources.
  • Biophysical Interpretation: The paper focuses on effective connectivity as capturing "biophysical transfer of activity," but acknowledges that this might not encompass all aspects of functional connectivity, especially those arising from more complex emergent properties.
  • Sample Size: The authors suggest that further experiments with larger datasets are needed to fully validate the findings, particularly for the classification tasks.
  • Future Work: The authors suggest extending the model to other data types (e.g., EEG, MEG) and exploring more sophisticated propagators for indirect connections. They also highlight the potential for using the derived effective connectivity for biomarker discovery and understanding neurological disorders.

Summary updated

Abstract

The relationship between structure and function is of interest in many research fields involving the study of complex biological processes. In neuroscience in particular, the fusion of structural and functional data can help understanding the underlying principles of the operational networks in the brain. To address this issue, this paper proposes a constrained autoregressive model leading to a representation of “effective” connectivity that can be used to better understand how the structure modulates the function. Or simply, it can be used to find novel biomarkers characterizing groups of subjects. In practice, an initial structural connectivity representation is re-weighted to explain the functional co-activations. This is obtained by minimizing the reconstruction error of an autoregressive model constrained by the structural connectivity prior. The model has been designed to also include indirect connections, allowing to split direct and indirect components in the functional connectivity, and it can be used with raw and deconvoluted BOLD signal. The derived representation of dependencies was compared to the well known dynamic causal model, giving results closer to known ground-truth. Further evaluation of the proposed effective network was performed on two typical tasks. In a first experiment the direct functional dependencies were tested on a community detection problem, where the brain was partitioned using the effective networks across multiple subjects. In a second experiment the model was validated in a case-control task, which aimed at differentiating healthy subjects from individuals with autism spectrum disorder. Results showed that using effective connectivity leads to clusters better describing the functional interactions in the community detection task, while maintaining the original structural organization, and obtaining a better discrimination in the case-control classification task. Highlights A method to combine structural and functional connectivity by using autoregressive model is proposed. The autoregressive model is constrained by structural connectivity defining coefficients for Granger causality. The usefulness of the generated effective connections is tested on simulations, ground-truth default mode network experiments, a classification and clustering task. The method can be used for direct and indirect connections, and with raw and deconvoluted BOLD signal.
BibTeX
@article{crimi2021structurally,
  author = {Alessandro Crimi and Luca Dodero and Fabio Sambataro and Vittorio Murino and Diego Sona},
  title = {{Structurally Constrained Effective Brain Connectivity}},
  journal = {NeuroImage},
  year = {2021},
  doi = {10.1016/j.neuroimage.2021.118288},
}
Figure