On the intersection between data quality and dynamical modelling of large-scale fMRI signals
Kevin M Aquino1, Ben Fulcher2, Stuart Oldham3
1The Turner Institute for Brain and Mental Health, School of Psychological Sciences, and Monash Biomedical Imaging, Monash University, Victoria 3168, Australia; School of Physics, University of Sydney, New South Wales, 2006 Australia.
Neuroimage
|March 11, 2022
Summary
Brain activity models using resting state functional MRI (fMRI) data are sensitive to preprocessing choices. Complex models may fit global signals, not neural dynamics, highlighting the need for data quality assurance in brain modeling.
Area of Science:
- Neuroscience
- Computational Neuroscience
- Systems Neuroscience
Background:
- Large-scale brain dynamics are modeled using nonlinear dynamical equations coupled by structural connectivity.
- Resting state functional MRI (fMRI) is used to study spontaneous brain dynamics.
- fMRI data preprocessing choices significantly impact functional connectivity estimates.
Purpose of the Study:
- To investigate the influence of fMRI preprocessing variations on whole-brain dynamical model fits and interpretations.
- To determine if complex models capture genuine neural dynamics or artifactual global signals.
- To propose benchmarks for evaluating model fit and validity in neuroimaging studies.
Main Methods:
- Analysis of three popular whole-brain dynamical models using different fMRI preprocessing pipelines.
- Development and testing of a simple two-parameter model to capture global signal fluctuations.
- Simulation and data analysis to evaluate model performance and resilience to denoising.
Main Results:
- fMRI preprocessing choices dramatically affect model fits and interpretations.
- Model accuracy is often driven by fitting global signals, not complex neural dynamics.
- A simple two-parameter model performs comparably to complex biophysical models in capturing fluctuations.
Conclusions:
- Many complex biophysical models may inadvertently fit trivial global signal properties.
- Relaxing assumptions of homogeneous neural populations can improve accuracy but increases complexity.
- Tighter integration of data quality assurance and model development is crucial for valid brain modeling.
