guest ::
| Home > Publications database > Effects of structural connectivity for the whole-brain resting-state dynamical models |
| Home > Publications database > Effects of structural connectivity for the whole-brain resting-state dynamical models |
| Poster (Other) | FZJ-2020-04942 |
; ; ;
2020
Please use a persistent id in citations: http://hdl.handle.net/2128/26372
Abstract: [INTRODUCTION]The primary goal of this study is to investigate the impact of MRI data processing on the modeling of the resting-state brain dynamics, in particular, the number of streamlines for the whole-brain tractography (WBT) that affects structural connectivity (the number of streamlines between two regions; SC) and path-length (mean length of streamlines; PL). We consider a system of phase oscillators (as brain regions) coupled with delay, where SC is used for coupling between two oscillators and PL is used as proxy for the delay of the signal propagation. We systematically explore the parameter space (coupling and delay) of the considered model and find optimal parameter configurations, where the model dynamics closely replicates the empirical data. Finally, we discuss the optimal number of streamlines and the difference between two distinct brain parcellations for modeling.[METHODS]T1-weighted images and diffusion-weighted images from 105 subjects from the Human Connectome Project (HCP) dataset were preprocessed for the WBT. The developed pipeline has three parts of preprocessing, tractography and reconstructing, where FSL, FreeSurfer, MRtrix, and ANTs were used. After preprocessing, the WBT was extracted for six different number of streamlines (10M, 2M, 500K, 100K, 50K, and 10K), and SC and PL were reconstructed for Schaefer's and Harvard-Oxford atlases with 100 and 96 cortical areas, respectively, by MRtrix. Furthermore, BOLD signals were extracted from the minimally preprocessed fMRI with these two atlases, and functional connectivity (FC) was calculated as Pearson cross-correlation. Finally, the whole-brain mathematical model of coupled phase oscillators, where the coupling weights, delays and natural frequencies were extracted from the empirical SC, PL and BOLD, respectively, generated simulated FC for 12 conditions (6 WBTs and 2 atlases).[RESULTS]Simulated FC was compared to the empirical FC and SC, and similarity between simulated and empirical data was calculated by Pearson correlation for varying parameters of the global coupling and delay. We found that the shapes of the corresponding model parameter planes were consistent across all WBTs. However, the distributions of the maximal similarity between simulated and empirical FCs and between simulated FC and empirical SC showed differences through varying WBTs. On the one hand, the case of 10M streamlines showed the highest similarity for Schaefer's atlas. On the other hand, WBT of 100K streamlines can be optimal for fitting simulated and empirical FC for Harvard-Oxford atlas. The optimal parameters for the strongest correlation between simulated and empirical FCs are unimodaly distributed for Harvard-Oxford atlas, whereas those for Schaefer's atlas can exhibit clustering. In contrast, the optimal parameter sets for the correlation between simulated FC and empirical SC showed a similar two-cluster pattern for both atlases. Figure 2 illustrates that clusters undergo a reshaping, where the optimal parameters redistribute between clusters as WBT varies for the Schaefer's atlas, whereas the clusters for Harvard-Oxford atlas remain stable for any considered number of streamlines.[CONCLUSIONS]In this study, we described the impact of WBT streamlines used for whole-brain models on the results of model validation. We showed that variation of WBT has an impact on the maximal similarity between simulated and empirical data as well as on the distribution of the optimal model parameters for the considered Schaefer's and Harvard-Oxford brain atlases. Still there are unclear structural-functional relationships for modeling, e.g., an effect of different types of atlases on the modeling outcome. Further systematic modeling studies on the effect of data-processing parameters are necessary for designing an optimal approach for dynamical models and neuroimaging data analytics.
|
The record appears in these collections: |
PDF
Fulltext by OpenAccess repository