Bayesian spatiotemporal modeling on complex-valued fMRI signals via kernel convolutions.

Area of Science:

• Neuroimaging
• Statistical modeling
• Signal processing

Background:

• Functional magnetic resonance imaging (fMRI) is crucial for understanding brain function.
• Complex-valued fMRI (CV-fMRI) offers richer information than magnitude-only approaches.
• Accurate voxel-level brain activation detection remains a challenge, especially with complex spatial and temporal dependencies.

Purpose of the Study:

• To develop a novel model-based approach for enhanced brain activation detection in CV-fMRI data.
• To improve the accuracy and reduce false positives in identifying brain activity at the voxel level.
• To compare the proposed model's performance against existing spatial and complex-valued fMRI analysis methods.

Main Methods:

• A model-based approach combining Bayesian variable selection, a novel spatial kernel convolution, and autoregressive processes.
• Development of a computationally efficient Markov chain Monte Carlo algorithm for posterior inference.
• Application of the model to simulated and human task-related CV-fMRI data.

Main Results:

• The proposed spatiotemporal model yields more accurate posterior probability activation maps with fewer false positives compared to alternative methods.
• Complex-valued approaches significantly outperform magnitude-only fMRI data analysis.
• The novel kernel structure enhances sensitivity rates for voxel-level activation detection.

Conclusions:

• The developed model provides a superior method for brain activation detection in CV-fMRI.
• Incorporating spatial and temporal structures is critical for accurate CV-fMRI analysis.
• The approach offers improved sensitivity and specificity for neuroimaging research.