Harmonizing connectome-wise statistics across different atlases
Poster No:
1511
Submission Type:
Abstract Submission
Authors:
Qingyuan Liu1, Yongbin Wei1, Koen Helwegen2, Long-Biao Cui3, Yong Liu1, Martijn van den Heuvel2
Institutions:
1School of Artificial Intelligence, Beijing University of Posts and Telecommunications, Beijing, China, 2Vrije Universiteit Amsterdam, Amsterdam, the Netherlands, 3The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an, Shanxi
First Author:
Qingyuan Liu
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
Co-Author(s):
Yongbin Wei
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
Yong Liu
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
School of Artificial Intelligence, Beijing University of Posts and Telecommunications
Beijing, China
Introduction:
Recent studies using multi-site data show that more reproducible connectome-wise alterations (1) can be reported when summary data is combined across multiple studies. This surge in the accumulation of vast amounts of neuroimaging data brings the need for new methods to incorporate summary statistics of connectome studies. This requires particularly new methods to combine data processed using multiple reconstruction pipelines, in particular concerning variations in atlas selection. Here, we introduce a novel computational framework to remap structural network-based t-statistics across brain atlases.
Methods:
T1-weighted MRI and diffusion-weighted imaging (DWI) of 1053 subjects from the Human Connectome Project (HCP) (2) were used. T1-weighted MRI data were processed using FreeSurfer (v7.2) with six atlases used for cortical parcellation, including the Desikan-Killiany (DK) atlas (N = 68) (3), the 114-/219-region subdivisions of the DK atlas (DK-114/DK-219; N = 114/219), the Brainnetome (BN) atlas (N = 210) (4), the HCP-MMP atlas (N = 360) (5), and the Schaefer atlas (N = 200) (6). DWI data were processed using CATO (v3.2.1), with white matter streamlines reconstructed through deterministic fiber tractography.
Next, we propose a framework comprising two consecutive modules to remap network-based t-statistics from a source atlas to a target atlas. The first module uses linear models to map connections from the source to corresponding connections in the target atlas. Coefficients in the linear models are deduced according to the overlap ratios of streamlines between the two atlases, by going through all reconstructed streamlines in each subject from the HCP data. The second module transposes the network-based t-statistic maps by means of derivations of t values incorporating parameters derived from the group-wise overlap ratios generated by the first module and variance maps for the disease and control groups under the source atlas. 369 schizophrenia patients and 418 healthy controls from five data cohorts, including the open-access COBRE (7), UCLA_UNP (8), MCICShare (9), and two in-house datasets (10), were used to examine how the framework could be applied to real network-based statistics and downstream meta-analysis.
Next, we propose a framework comprising two consecutive modules to remap network-based t-statistics from a source atlas to a target atlas. The first module uses linear models to map connections from the source to corresponding connections in the target atlas. Coefficients in the linear models are deduced according to the overlap ratios of streamlines between the two atlases, by going through all reconstructed streamlines in each subject from the HCP data. The second module transposes the network-based t-statistic maps by means of derivations of t values incorporating parameters derived from the group-wise overlap ratios generated by the first module and variance maps for the disease and control groups under the source atlas. 369 schizophrenia patients and 418 healthy controls from five data cohorts, including the open-access COBRE (7), UCLA_UNP (8), MCICShare (9), and two in-house datasets (10), were used to examine how the framework could be applied to real network-based statistics and downstream meta-analysis.
Results:
We started by simulating network-based statistics by randomly generating two groups using the HCP dataset, with a total of 1000 simulations performed in order to evaluate the framework. Taking the DK-114 atlas as a reference, we show the results of remapping originating from the rest of the atlases in Fig. 1A. Empirical t-statistic maps in the DK-114 atlas show high spatial correlations with the predicted t-statistic maps originated from the rest of the five atlases (r=0.53~0.95). Permutation testing further shows that the correlation achieved in each simulation significantly exceeded null distributions derived from correlations to the rest of the simulations (all p<.001). Furthermore, we show our framework's utility in a meta-analysis, with the effect sizes from five schizophrenia cohorts being combined. The real network-based Cohen's d maps in the DK-114 atlas show high spatial correlations with the predicted Cohen's d maps originating from the five referenced atlases (r=0.56~0.94) (Fig. 2A). Cohen's d maps based on different atlases were selected independently and again revealed a high correlation (r=0.74). Furthermore, we set a range of thresholds of effect sizes and showed 66%±16% (mean ± SD) connections with a certain real effect size could be identified according to predicted effect sizes (Fig. 2B).
·Figure 1
·Figure 2
Conclusions:
We present a powerful framework to remap network-based t-statistics across structural connectomes in various brain atlases. This tool harmonizes connectome-wise summary statistics obtained from different atlases, thereby enhancing the ability to uncover disease heterogeneity within the brain connectome by combining connectome summary statistics across studies.
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
Methods Development 2
Keywords:
Other - Connectome; Atlas; Multi-site; Summary statistics; Structural connectivity
1|2Indicates the priority used for review
Provide references using author date format
2. Glasser, M. F. et al. (2016), The human connectome project's neuroimaging approach. Nature neuroscience 19, 1175-1187.
3. Desikan, R. S. et al. (2006), An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 31, 968–980.
4. Fan, L. et al. (2016), The Human Brainnetome Atlas: A New Brain Atlas Based on Connectional Architecture. Cereb Cortex 26, 3508–3526.
5. Glasser, M. F. et al. (2016), A multi-modal parcellation of human cerebral cortex. Nature 536, 171–178.
6. Schaefer, A. et al. (2018), Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cerebral Cortex 28, 3095–3114.
7. Cetin, M. S. et al. (2014), Thalamus and posterior temporal lobe show greater inter-network connectivity at rest and across sensory paradigms in schizophrenia. NeuroImage 97, 117–126.
8. Poldrack, R. A. et al. (2016), A phenome-wide examination of neural and cognitive function. Sci Data 3, 160110.
9. Gollub, R. L. et al. (2013), The MCIC Collection: A Shared Repository of Multi-Modal, Multi-Site Brain Image Data from a Clinical Investigation of Schizophrenia. Neuroinform 11, 367–388.
10. Cui, L.-B. et al. (2019), Connectome-Based Patterns of First-Episode Medication-Naïve Patients With Schizophrenia. Schizophrenia Bulletin 45, 1291–1299.