Abstract
The anatomical structure of the brain can be observed via non-invasive techniques such as diffusion imaging. However, these are imperfect because they miss connections that are actually known to exist, especially long range inter-hemispheric ones. In this paper we formulate the inverse problem of inferring the structural connectivity of brain networks from experimentally observed functional connectivity via functional Magnetic Resonance Imaging (fMRI), by formulating it as a convex optimization problem. We show that structural connectivity can be modeled as an optimal sparse representation derived from the much denser functional connectivity in the human brain. Using only the functional connectivity data as input, we present (a) an optimization problem that models constraints based on known physiological observations, and (b) an ADMM algorithm for solving it. The algorithm not only recovers the known structural connectivity of the brain, but is also able to robustly predict the long range inter-hemispheric connections missed by DSI or DTI, including a very good match with experimentally observed quantitative distributions of the weights/strength of anatomical connections. We demonstrate results on both synthetic model data and a fine-scale 998 node cortical dataset, and discuss applications to other complex network domains where retrieving effective structure from functional signatures are important.
Introduction
Understanding the human connectome requires mapping two distinct yet interrelated networks: structural connectivity (SC), representing physical white-matter tracts, and functional connectivity (FC), representing statistical correlations between regional brain activity. Currently, SC is primarily mapped using Diffusion Spectrum Imaging (DSI) or Diffusion Tensor Imaging (DTI). However, these methods frequently fail to detect long-range connections, particularly those crossing between hemispheres. Conversely, fMRI-derived FC data is significantly denser, capturing correlations that do not necessarily imply direct physical paths. This research addresses the inverse problem: mathematically deriving the sparse physical skeleton of the brain from dense functional data.
Methodology: Structure as a Sparse Basis
The authors propose that structural connectivity can be modeled as an optimal sparse representation of functional connectivity. The approach involves several key steps:
- Spectral Representation: To reduce noise and capture the brain's hierarchical nature, the algorithm operates on the spectral representation (the first $k$ eigenvectors) of the functional matrix.
- Convex Optimization: An objective function is formulated to find a sparse matrix $X$ that maps the functional space back to itself. To account for the debated role of negative correlations in fMRI, the solution is split into a sparse positive component ($X_p$) and a small negative component ($X_n$).
- ADMM Algorithm: The Alternating Direction Method of Multipliers (ADMM) is utilized to solve the optimization problem efficiently at scale.
This framework was validated using both synthetic stochastic block models and a high-resolution 998-node human cortical dataset containing both DSI and fMRI data for the same subjects.
Results and Performance
The algorithm demonstrates a high capacity for recovering known anatomical connections while identifying biologically plausible "missing" links.
- High Recall and Robustness: On real brain data, the algorithm achieved a recall rate between 0.7 and 0.8, successfully identifying the majority of connections present in DSI data. These results remained stable across various values of $k$, indicating robustness to parameter selection.
- The Precision Paradox: While raw precision metrics appeared lower when compared directly to DSI, the "false positives" identified by the algorithm were largely long-range inter-hemispheric links. These links are known to exist biologically but are typically invisible to DSI, suggesting the algorithm's "errors" are actually improvements over existing imaging modalities.
Furthermore, the optimization approach produced a distribution of connection weights that closely matches experimental biological observations, unlike previous methods that often yielded unrealistic density.
Discussion and Future Directions
The ability to infer structural bottlenecks and pathways from functional signatures has implications beyond neuroscience, potentially applying to infrastructure networks like power grids or transportation systems. Future work may incorporate dynamic constraints, such as stability and criticality, to ensure the modeled networks reflect the physiological balance of the living brain. While the interpretation of negative correlations remains an open question, this sparse representation framework provides a powerful tool for bridging the gap between brain structure and function.
Summary updated