Tractography density affects whole-brain structural architecture and resting-state dynamical modeling

Abstract

Dynamical modeling of the resting-state brain dynamics essentially relies on the empirical neuroimaging data utilized for the model derivation and validation. There is however still no standardized data processing for magnetic resonance imaging pipelines and the structural and functional connectomes involved in the models. In this study, we thus address how the parameters of diffusion-weighted data processing for structural connectivity (SC) can influence the validation results of the whole-brain mathematical models and search for the optimal parameter settings. On this way, we simulate the functional connectivity by systems of coupled oscillators, where the underlying network is constructed from the empirical SC and evaluate the performance of the models for varying parameters of data processing. For this, we introduce a set of simulation conditions including the varying number of total streamlines of the whole-brain tractography (WBT) used for extraction of SC, cortical parcellations based on functional and anatomical brain properties and distinct model fitting modalities. We observed that the graph-theoretical network properties of structural connectome can be affected by varying tractography density and strongly relate to the model performance. We explored free parameters of the considered models and found the optimal parameter configurations, where the model dynamics closely replicates the empirical data. We also found that the optimal number of the total streamlines of WBT can vary for different brain atlases. Consequently, we suggest a way how to improve the model performance based on the network properties and the optimal parameter configurations from multiple WBT conditions. Furthermore, the population of subjects can be stratified into subgroups with divergent behaviors induced by the varying number of WBT streamlines such that different recommendations can be made with respect to the data processing for individual subjects and brain parcellations.Author summaryThe human brain connectome at macro level provides an anatomical constitution of inter-regional connections through the white matter in the brain. Understanding the brain dynamics grounded on the structural architecture is one of the most studied and important topics actively debated in the neuroimaging research. However, the ground truth for the adequate processing and reconstruction of the human brain connectome in vivo is absent, which is crucial for evaluation of the results of the data-driven as well as model-based approaches to brain investigation. In this study we thus evaluate the effect of the whole-brain tractography density on the structural brain architecture by varying the number of total axonal fiber streamlines. The obtained results are validated throughout the dynamical modeling of the resting-state brain dynamics. We found that the tractography density may strongly affect the graph-theoretical network properties of the structural connectome. The obtained results also show that a dense whole-brain tractography is not always the best condition for the modeling, which depends on a selected brain parcellation used for the calculation of the structural connectivity and derivation of the model network. Our findings provide suggestions for the optimal data processing for neuroimaging research and brain modeling.

Kuramoto model for populations of quadratic integrate-and-fire neurons with chemical and electrical coupling

“…Geometric relation -determined by Eqs. (20,21)-between the coupling parameters of the QIF model Eqs. ( 1) and the Kuramoto model Eqs.…”

Whole-brain dynamical modeling for classification of Parkinson’s disease

“…Several fitting modalities have been suggested in the literature including the fitting of the grand-averaged empirical and simulated FC matrices, fitting the dynamical FCs, maximization of the metastability, and structure-functional model fitting. 6,13,24,29,30 On that account, it is necessary to investigate which parameter points of a given dynamical mode and which model fitting modalities are the most suitable to answer a given research question by the modeling approach. For example, it was observed that distributions of the optimal model parameters differ when using only functional or structure-functional model fitting and may lead to subject stratifications showing different model fitting values and optimal parameter points.…”

“…For example, it was observed that distributions of the optimal model parameters differ when using only functional or structure-functional model fitting and may lead to subject stratifications showing different model fitting values and optimal parameter points. 30 It is also well known that varying parameters of MRI data processing influence the empirical structural and functional connectomes and their analyses. [31][32][33][34] This subsequently affects model validation.…”

“…[31][32][33][34] This subsequently affects model validation. 6,30,35 Therefore, the impact of data processing on the results of model validation should carefully be considered, especially in clinical applications.…”

“…40 The frequencies of empirical BOLD signals, when included in the whole-brain mathematical models, may influence the optimal model parameters and the quality of the model fitting. 6,30 In this context, investigation of the impact of temporal filtering conditions on the model validation in PD data is important. In the current study we advance the classification of clinical data by application of machine learning to empirical and simulated connectomes.…”