guest ::
| Home > Publications database > Optimizing Current Imaging Pipelines by Whole-Brain Dynamical Models |
| Home > Publications database > Optimizing Current Imaging Pipelines by Whole-Brain Dynamical Models |
| Poster (Invited) | FZJ-2020-05030 |
; ; ; ; ;
2020
Please use a persistent id in citations: http://hdl.handle.net/2128/26406
Abstract: Processing of neuroimaging data and extraction of the structural and functional brain signals relies on plethora of setting and parameters whose values are largely conjectured. This problem is crucial for simulating the whole-brain dynamics by mathematical models derived from and validated against empirical data, which has attracted a great interest during the last decade. For example, considering a certain brain parcellation is essential for defining a model network, but there is practically no empirical evidence for the effect of a particular atlas choice. We address this problem using a computational modeling approach, where the resting-state brain dynamics as reflected by fMRI BOLD is simulated by a dynamical model of coupled oscillators. We considered a system of coupled phase oscillators, where the coupling weights and delays were extracted from empirical structural connectivity (eSC). We compared the modeling outcome for three different brain parcellations as given by the functional Schaefer atlas with 100 and 200 cortical parcels (S100 and S200) and anatomical Harvard-Oxford atlas with 96 parcels (HO96). The natural frequencies of the model oscillators were extracted from non-filtered (NF) BOLD signal as well as filtered in different frequency bands including the broad- (BF), low- (LF) and high- (HF) frequency bands. The model was validated by finding the optimal model parameters of the global coupling and delay, where the strongest similarity as revealed by Pearson’s correlation between simulated and empirical functional connectivity (sFC and eFC) or between sFC and eSC is achieved. We found that the maximal similarity between functional data demonstrates a moderate improvement of the model fitting when comparing S200 to S100, and there is a strong enhancement when comparing HO96 to S100 and S200. For the structure-function relationship between sFC and eSC, the ordering NF < BF < LF < HF can be observed, but the differences between atlases are less consistent. We found that HF condition can lead to distinct distributions of the optimal model parameters than other frequency bands, and the corresponding fitting results less correlate with those obtained for other frequency bands. Finally, the dynamics of the validated model can result in pronounced bimodal distributions of the order parameter R(t) and other quantities characterizing the model fitting for S100 and S200, but not for HO96. We thus showed that a choice of a particular brain parcellation and variation of BOLD frequency band can cause a significant impact on the quality of the model fitting, dynamics of the validated model and their interrelations. The main impact of the brain atlases can be observed for the fitting of sFC to eFC. The frequency bands of the BOLD filtering mostly affect structure-function relationships and can also influence the reliability of the model validation.
Keyword(s): Health and Life (1st) ; Medicine (2nd)
|
The record appears in these collections: |
PDF
Fulltext