% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Wischnewski:1018096,
author = {Wischnewski, Kevin and Eickhoff, Simon and Popovych,
Oleksandr},
title = {{V}alidation of dynamical whole-brain models in
high-dimensional parameter spaces},
school = {HHU Düsseldorf},
reportid = {FZJ-2023-04546},
year = {2023},
abstract = {INTRODUCTIONSimulating the resting-state brain dynamics via
mathematical whole-brain models allows for describing a
subject’s brain activity by a set of interpretable model
parameters. In this setting, increased efforts are currently
devoted to personalized simulations, which require a rising
number of optimally selected model parameters [1]. However,
a dense grid search is computationally unfeasible and
constrains the studies of high-dimensional models. In our
work, we apply mathematical optimization algorithms to
explore low- and high-dimensional parameter spaces at
moderate computational costs. We analyze the results in
terms of fit to empirical data and required computation time
as well as the impact of the number of optimized model
parameters on the differentiability between males and
females based on simulated data.METHODSWe used neuroimaging
data of 272 healthy subjects (128 males) of the Human
Connectome Project [2]. Working with an ensemble of coupled
phase oscillators [3] built upon individual empirical
structural connectivity (SC), we aimed at replicating the
empirical functional connectivity (FC) of every subject. We
considered Schaefer’s functional brain atlas (100 regions,
[4]) and the anatomical Harvard-Oxford parcellation (96
regions, [5]) as well as 2 optimization schemes based on
Covariance Matrix Adaptation Evolution Strategy (CMAES, [6])
and Bayesian Optimization (BO, [7]). Both algorithms
represent global stochastic search methods and were applied
to detect the optimal model parameters which maximize the
correlation between empirical FC (eFC) and simulated FC
(sFC). The optimized parameters included the global coupling
and delay (2D parameter space), the noise intensity (3D
parameter space), and the oscillation frequencies of
individual brain regions (99D or 103D parameter space). By
optimizing between 2 and 103 free parameters simultaneously,
we generated whole-brain models of an improved fitting to
empirical data of individual subjects and thus enhanced
model personalization, which we compared across subjects and
brain atlases. Particular focus was set on the optimal model
parameters and sFC as well as on their reliability and
subject-specificity.RESULTSWhen increasing the number of
optimized model parameters, we observed an enhancement of
the quality of the model validation, i.e. the similarity
between eFC and sFC (goodness-of-fit, GoF), for both atlases
and optimization algorithms. While the improvement from 2 to
3 variables ranges around $12\%,$ the transition to the
high-dimensional cases is sharp and can double the GoF. Also
the computational demands increased, but remain within a
tractable extent: For CMAES, the requirements were doubled,
and for BO around 12 times more resources were needed. The
reliability of the optimal parameters drops in higher
dimensions, and several constellations of parameter values
appear to generate comparably high GoF values. However, we
also observed high correlations (positive and negative)
between individual solutions. This hints at the presence of
a manifold in the model parameter space, where the optimal
values may be located. Additionally, we observed higher GoF
values for males than for females, with the differentiation
between these subject groups being enhanced for the model
optimization in high-dimensional parameter
spaces.CONCLUSIONSOur results provide an insight into the
model validation in high-dimensional parameter spaces, which
has been made possible by mathematical optimization schemes.
In particular, we have shown that more optimized parameters
lead to a much higher GoF that can be obtained for several
configurations of model parameters. The fact that
phenotypical differences were more pronounced in the data
derived from high-dimensional simulations implies a great
potential for model-based, personalized studies with
application to the investigation of inter-individual
variability.REFERENCES[1] Hashemi, M., Vattikonda, A. N.,
Sip, V., Diaz-Pier, S., Peyser, A., Wang, H., Guye, M.,
Bartolomei, F., Woodman, M. M., $\&$ Jirsa, V. K. (2021). On
the influence of prior information evaluated by fully
Bayesian criteria in a personalized whole-brain model of
epilepsy spread. PLoS computational biology, 17(7),
e1009129. https://doi.org/10.1371/journal.pcbi.1009129[2]
Van Essen, D. C., Smith, S. M., Barch, D. M., Behrens, T.
E., Yacoub, E., Ugurbil, K., $\&$ WU-Minn HCP Consortium
(2013). The WU-Minn Human Connectome Project: an overview.
NeuroImage, 80, 62–79.
https://doi.org/10.1016/j.neuroimage.2013.05.041[3] Cabral,
J., Hugues, E., Sporns, O., $\&$ Deco, G. (2011). Role of
local network oscillations in resting-state functional
connectivity. NeuroImage, 57(1), 130–139.
https://doi.org/10.1016/j.neuroimage.2011.04.010[4]
Schaefer, A., Kong, R., Gordon, E. M., Laumann, T. O., Zuo,
X. N., Holmes, A. J., Eickhoff, S. B., $\&$ Yeo, B. T. T.
(2018). Local-Global Parcellation of the Human Cerebral
Cortex from Intrinsic Functional Connectivity MRI. Cerebral
cortex (New York, N.Y. : 1991), 28(9), 3095–3114.
https://doi.org/10.1093/cercor/bhx179[5] Desikan, R. S.,
Ségonne, F., Fischl, B., Quinn, B. T., Dickerson, B. C.,
Blacker, D., Buckner, R. L., Dale, A. M., Maguire, R. P.,
Hyman, B. T., Albert, M. S., $\&$ Killiany, R. J. (2006). An
automated labeling system for subdividing the human cerebral
cortex on MRI scans into gyral based regions of interest.
NeuroImage, 31(3), 968–980.
https://doi.org/10.1016/j.neuroimage.2006.01.021[6] Hansen,
N. (2006). The CMA Evolution Strategy: A Comparing Review.
In: Lozano, J.A., Larrañaga, P., Inza, I., Bengoetxea, E.
(eds) Towards a New Evolutionary Computation. Studies in
Fuzziness and Soft Computing, vol 192. Springer, Berlin,
Heidelberg. $https://doi.org/10.1007/3-540-32494-1_4[7]$
Martinez-Cantin, R. (2014). BayesOpt: A Bayesian
Optimization Library for Nonlinear Optimization,
Experimental Design and Bandits. Journal of Machine Learning
Research, 15, 3735-3739.[8] Rosenthal, R., $\&$ Rosnow, R.
L. (1991). Essentials of behavioral research: Methods and
data analysis (2nd ed.). New York: McGraw Hill.},
month = {Jul},
date = {2023-07-22},
organization = {The 29th Annual Meeting of the
Organization for Human Brain Mapping
(OHBM2023), Montréal (Canada), 22 Jul
2023 - 26 Jul 2023},
subtyp = {After Call},
cin = {INM-7},
cid = {I:(DE-Juel1)INM-7-20090406},
pnm = {5232 - Computational Principles (POF4-523) / HBP SGA2 -
Human Brain Project Specific Grant Agreement 2 (785907) /
HBP SGA3 - Human Brain Project Specific Grant Agreement 3
(945539) / VirtualBrainCloud - Personalized Recommendations
for Neurodegenerative Disease (826421)},
pid = {G:(DE-HGF)POF4-5232 / G:(EU-Grant)785907 /
G:(EU-Grant)945539 / G:(EU-Grant)826421},
typ = {PUB:(DE-HGF)24},
doi = {10.34734/FZJ-2023-04546},
url = {https://juser.fz-juelich.de/record/1018096},
}