Parameter clustering in Bayesian functional principal component analysis of neuroscientific data

Area of Science:

• Neuroscience
• Statistics
• Data Science

Background:

• Neuroscientific technology has generated complex spatiotemporal datasets.
• Existing models may oversimplify intricate patterns in brain recordings.
• Advanced analytical methods are needed for meaningful pattern identification.

Purpose of the Study:

• To propose a novel model, parameter clustering functional principal component analysis (PCl-fPCA), for exploring complex spatiotemporal neuroscientific data.
• To develop a computationally feasible approach for signal reconstruction and pattern discovery.
• To enhance the understanding of brain time series data.

Main Methods:

• Merging functional data analysis and Bayesian nonparametrics.
• Utilizing a Dirichlet process Gaussian mixture model to cluster functional principal component scores within a Bayesian functional PCA framework.
• Capturing spatial dependence and temporal interactions without prior spatial assumptions.

Main Results:

• Demonstrated improved curve and correlation reconstruction in simulation studies compared to existing Bayesian and frequentist functional PCA models.
• Successfully applied the PCl-fPCA method to functional magnetic resonance imaging (fMRI) and electroencephalogram (EEG) data.
• Provided a rich exploration of spatiotemporal dependence in brain time series.

Conclusions:

• PCl-fPCA offers a flexible and computationally feasible method for analyzing complex neuroscientific data.
• The model enhances insight into spatiotemporal dynamics by clustering functional principal component scores.
• This approach advances the analysis of fMRI and EEG data, revealing intricate patterns in brain activity.