Integrated Effective Connectivity Reveals Mesulam’s Cortical Hierarchy in the Human Brain

1761 

Abstract Submission 

Younghyun Oh^1,2,3, Yejin Ann^1,4, Takuya Ito⁵, Sean Froudist-Walsh⁶, Boris Bernhardt⁷, Casey Paquola⁸, Michael Milham^9,10, Choong-Wan Woo^1,2,3, Seok-Jun Hong^1,2,4,3,10

¹IBS Center for Neuroscience Imaging Research, Sungkyunkwan University, Suwon, South Korea, ²Department of Biomedical Engineering, Sungkyunkwan University, Suwon, South Korea, ³Life-inspired Neural networks for Prediction and Optimization (LNPO) Group, Suwon, South Korea, ⁴Department of Intelligence Precision Health Care, Sungkyunkwan University, Suwon, South Korea, ⁵T.J. Watson Research Center, IBM Research,, Yorktown, United States,, ⁶Bristol Computational Neuroscience Unit, Faculty of Engineering, University of Bristol,, Bristol, United Kingdom,, ⁷McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital, McGill University,, Montreal, Canada,, ⁸Institute of Neuroscience and Medicine, Forschungszentrum Jülich,, Jülich, Germany, ⁹Nathan S. Kline INstitute for Psychiatric Research,, New York, United States, ¹⁰Center for the Developing Brain, Child Mind Institute,, New York, United States

Younghyun Oh  
IBS Center for Neuroscience Imaging Research, Sungkyunkwan University|Department of Biomedical Engineering, Sungkyunkwan University|Life-inspired Neural networks for Prediction and Optimization (LNPO) Group
Suwon, South Korea|Suwon, South Korea|Suwon, South Korea

Yejin Ann  
IBS Center for Neuroscience Imaging Research, Sungkyunkwan University|Department of Intelligence Precision Health Care, Sungkyunkwan University
Suwon, South Korea|Suwon, South Korea

Takuya Ito, PhD  
T.J. Watson Research Center, IBM Research,
Yorktown, United States,

Sean Froudist-Walsh  
Bristol Computational Neuroscience Unit, Faculty of Engineering, University of Bristol,
Bristol, United Kingdom,

Boris Bernhardt  
McConnell Brain Imaging Centre, Montreal Neurological Institute and Hospital, McGill University,
Montreal, Canada,

Casey Paquola  
Institute of Neuroscience and Medicine, Forschungszentrum Jülich,
Jülich, Germany

Michael Milham  
Nathan S. Kline INstitute for Psychiatric Research,|Center for the Developing Brain, Child Mind Institute,
New York, United States|New York, United States

Choong-Wan Woo  
IBS Center for Neuroscience Imaging Research, Sungkyunkwan University|Department of Biomedical Engineering, Sungkyunkwan University|Life-inspired Neural networks for Prediction and Optimization (LNPO) Group
Suwon, South Korea|Suwon, South Korea|Suwon, South Korea

Seok-Jun Hong  
IBS Center for Neuroscience Imaging Research, Sungkyunkwan University|Department of Biomedical Engineering, Sungkyunkwan University|Department of Intelligence Precision Health Care, Sungkyunkwan University|Life-inspired Neural networks for Prediction and Optimization (LNPO) Group|Center for the Developing Brain, Child Mind Institute,
Suwon, South Korea|Suwon, South Korea|Suwon, South Korea|Suwon, South Korea|New York, United States

Identifying macro-scale directed functional flows between the brain regions, i.e. effective connectivity (EC), is key to understanding the emergence of our complex behaviors. However, reliable EC mapping remains as a daunting task, due to a lack of consensus across existing EC algorithms. This absence of methodological integration hinders a deeper investigation of neural dynamics along the brain architectures. Here, we sought to address these issues by introducing a new analytical framework, "integrated EC" (iEC), which combines the strengths of individual EC algorithms to create synergistic effects on sensitivity and reliability. Our framework revealed distinct connectome profiles across large-scale cortical hierarchy, demonstrating the utility of iEC as a robust tool for network neuroscience.

To implement the iEC framework, we first selected the major EC algorithms based on their mathematical uniqueness (Fig 1a). The iEC was then obtained by linear combination of the individual algorithms via Bayesian optimization (Fig 1b). We validated the inferred EC by assessing whether it could recover a FC pattern from rs-fMRI (Fig 1e). Next, we investigated the connectome profile, signal flows, and functional hierarchy using iEC to elucidate the organization of the human brain. For connectome profile, we assessed in-/out-degree and computed the ratio of their positive/negative connections. We further mapped the signal flow along the iEC by solving an equation 'x(t)=e^At x_0', where x_0 is a binary vector initialized with ones at seed brain regions and A is a the iEC matrix. Finally. the cortical hierarchy was determined based on the established method1, which involved conducting a general linear model in the form g(E[EC_ij ])=θ_i-θ_j, with θ_i representing the hierarchical level of brain region i.

Prior to implementing the iEC framework, we first assessed face validity of EC algorithms using Hopf model2 based on the ground-truth directed connectivity (Fig 1c). Across different network resolutions, iEC showed a superior performance over individual algorithms, showing mean correlation values of 0.82 and 0.41 with ground-truth networks of 50 and 180 nodes, respectively (Fig 1d). This validity was confirmed by a subsequent analysis based on experimental rs-fMRI data, in which we recovered FC from EC and tested the model fit between simulated and empirical FCs (Fig 1e). Again, the iEC showed better FC recovery compared to individual algorithms (Fig 1f; model fit r=0.83, Fig 1h), with a notable improvement by the inclusion of negative connections (Fig 1g). The connectome profile of iEC showed two distinct network characteristics (Fig 2a, b): i) generally higher in/out-degree in the default mode network (DMN) and ii) a clear positive-negative connection ratio along the sensory-fugal axis. This distinct topology informs the observed signal propagation pattern (Fig 2c, d): Positive signals predominantly originated from lower sensory regions, whereas negative signals propagated mostly from the higher association areas. Building on these findings, we posited that the positive/negative connections might correspond with feedforward/feedback connections, respectively, enabling us to deduce a cortical hierarchy from the discerned iEC. (Fig 2e). Our hierarchy map diverged from the established principal gradient4, particularly by positioning interoceptive regions at the top of hierarchy, instead of DMN (Fig 2f-g), which fully reflects the original concept of cortical diagram proposed by Mesulam4 (Fig 2h).

The iEC approach we introduce aims to overcome the current challenges in neuroimaging related to inferring the directionality of connectomes. By integrating existing methods, we have enhanced both accuracy and validity of EC mapping, which we believe the two key factors in uncovering novel principles of the human brain organization.

fMRI Connectivity and Network Modeling ¹

Methods Development

Anatomy and Functional Systems

Cortical Anatomy and Brain Mapping ²

Cortical Cyto- and Myeloarchitecture

FUNCTIONAL MRI

Modeling

Other - Effective Connectivity