Decoding short- and long-range structural connectivity using manifold learning techniques

Poster No:


Submission Type:

Abstract Submission 


Seulki Yoo1, Sunghyoung Hong2,3, Seok-Jun Hong2,3,4, Bo-yong Park5,6,3


1Convergence Research Institute, Sungkyunkwan University, Suwon, Republic of Korea, 2Department of Biomedical Engineering, Sungkyunkwan university, Suwon, Republic of Korea, 3Center for Neurosciene and Imaging Research, Sungkyunkwan University, Suwon, Republic of Korea, 4Center for the Developing Brain, Child Mind Institute, New York, NY, United States, 5Department of Data Science, Inha University, Incheon, Republic of Korea, 6Department of Statistics and Data Science, Inha University, Incheon, Republic of Korea

First Author:

Seulki Yoo  
Convergence Research Institute, Sungkyunkwan University
Suwon, Republic of Korea


Sunghyoung Hong  
Department of Biomedical Engineering, Sungkyunkwan university|Center for Neurosciene and Imaging Research, Sungkyunkwan University
Suwon, Republic of Korea|Suwon, Republic of Korea
Seok-Jun Hong  
Department of Biomedical Engineering, Sungkyunkwan university|Center for Neurosciene and Imaging Research, Sungkyunkwan University|Center for the Developing Brain, Child Mind Institute
Suwon, Republic of Korea|Suwon, Republic of Korea|New York, NY, United States
Bo-yong Park  
Department of Data Science, Inha University|Department of Statistics and Data Science, Inha University|Center for Neurosciene and Imaging Research, Sungkyunkwan University
Incheon, Republic of Korea|Incheon, Republic of Korea|Suwon, Republic of Korea


The human brain consists of billions of neurons and organizes networks through complex physical connections between brain regions. The structural connectivity estimated via diffusion magnetic resonance imaging (MRI) tractography represents the physical architecture of the brain. Investigation of short- and long-range connectivity is intriguing because these patterns reflect brain hierarchy, where the unimodal regions are mainly characterized by short-range connections, while the heteromodal association areas by a large number of long-range connections.[1] However, quantitative differences between short- and long-range structural connections and their role in explaining brain function have not been investigated. In this study, we aim to unravel the underlying characteristics of short- and long-range structural connectivity by employing manifold learning techniques.


We obtained diffusion MRI data of 86 healthy young adults (male/female = 35/51; mean ± standard deviation [SD] age = 28.77 ± 3.33 [range = 22-36]) from the Q3 release of the Human Connectome Project database.[2] Diffusion MRI data were preprocessed using MRtrix3 [3], and structural connectome was constructed based on the sub-parcellation of the Desikan-Killiany atlas with 400 parcels [4] and log-transformed to adjust for the scale.[5] For each individual, the streamlines with a connection length below 30% of the maximum length were defined as short-range connections, while those above 30% were considered long-range connections.[6] We then estimated low-dimensional representations of the structural connectivity (i.e., gradients) for both short- and long-range connections using nonlinear dimensionality reduction techniques.[7] Furthermore, we associated the short- and long-range connectivity-based gradients with multiple task activation maps obtained from the BrainMap database to assess distinct structure-function coupling strategies according to the connection lengths.


We generated five gradients (G1-G5) for short- and long-range structural connectivity, as well as the whole connectivity (Figure 1). Although the order of the gradients was slightly different (G3 and G4), their spatial patterns based on the whole connectivity were largely explained by those based on the short-range connectivity (spatial correlations: FDR-corrected p-values < 0.001). For the long-range connectivity, the gradients of the anterior-posterior axis (G2) and sensorimotor-transmodal axis (G4) showed similar spatial patterns with those based on the whole connectivity. However, we could not find the left-right and dorsal-ventral patterns seen in the whole connectivity-based gradients. When we associated the gradients with task-evoked functional activation patterns, the anterior-posterior and sensorimotor-transmodal axes were closely associated with emotions and actions, regardless of connection lengths (Figure 2). On the other hand, the interoception was dominantly related to the short-range gradients, while perception was associated with the long-range gradients.
Supporting Image: Figure1.png
Supporting Image: Figure2.png


By estimating principal gradients of short- and long-range structural connectomes [8, 9], we found that short-range connections may dominate the whole-brain structural connectome organization. Associations with task maps revealed a distinct relationship between short- and long-range connections. Although defining the threshold level should be explored further, our work offered a novel insight for the principle of structural connectome organization with respect to physical connection lengths.

Modeling and Analysis Methods:

Connectivity (eg. functional, effective, structural) 1
Diffusion MRI Modeling and Analysis 2


Computational Neuroscience
Other - Gradient; Structure-function coupling; Diffusion tractography

1|2Indicates the priority used for review

Provide references using author date format

[1]. Oligschläger, S. (2017), 'Gradients of connectivity distance are anchored in primary cortex', Brain Structure and Function, vol. 222, pp. 2173-2182
[2]. WU-Minn, H. C. P. (2017), '1200 subjects data release reference manual', (https://www. humanconnectome. org)
[3]. Tournier J-D. (2019), 'MRtrix3: A fast, flexible and open software framework for medical image processing and visualisation', Neuroimage, vol. 202, pp. 116137
[4]. Desikan, R. S. (2006), 'An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest', Neuroimage, vol. 31, no.3, pp. 968-980
[5]. Fornito, A. (2016), 'Fundamentals of brain network analysis', Academic press.
[6]. Betzel, RF. (2019), 'Distance-dependent consensus thresholds for generating group-representative structural brain networks', Network neuroscience, vol.3, pp.475-496
[7]. Vos de Wael, R (2020), 'BrainSpace: a toolbox for the analysis of macroscale gradients in neuroimaging and connectomics datasets', Communications biology, vol.3, no.1, pp.103
[8]. Margulies, D. S. (2016), 'Situating the default-mode network along a principal gradient of macroscale cortical organization' Proceedings of the National Academy of Sciences, vol.113, no.44, pp.12574-12579
[9]. Huntenburg, J. M. (2018), 'Large-scale gradients in human cortical organization', Trends in cognitive sciences, vol.22, no.1, pp.21-31

National Research Foundation of Korea (NRF-2021R1F1A1052303; NRF-2022R1A5A7033499), Institute for Information and Communications Technology Planning and Evaluation (IITP) funded by the Korea Government (MSIT) (No. 2022-0-00448, Deep Total Recall: Continual Learning for Human-Like Recall of Artificial Neural Networks; No. RS-2022-00155915, Artificial Intelligence Convergence Innovation Human Resources Development (Inha University); No. 2021-0-02068, Artificial Intelligence Innovation Hub), Institute for Basic Science (IBS-R015-D1).