Optimal HRF and smoothing parameters for fMRI time series within an autoregressive modeling framework

Area of Science:

• Neuroimaging analysis
• Statistical modeling in neuroscience
• Functional magnetic resonance imaging (fMRI)

Background:

• Functional magnetic resonance imaging (fMRI) data analysis often involves predictive parametric models like nearest-neighbor autoregressive models with exogenous input (NNARX).
• Preprocessing steps such as spatial smoothing are common but their optimal application requires careful consideration.
• Constraining autoregressive parameters to mimic the canonical hemodynamic response function (HRF) is a key aspect of fMRI modeling.

Purpose of the Study:

• To present a novel algorithm for estimating parameters of linear transformations and the HRF within a maximum-likelihood framework for fMRI data.
• To enable simultaneous estimation of the optimal amount of spatial smoothing and HRF parameters for a given fMRI dataset.
• To demonstrate the capability of this framework for estimating activated brain regions.

Main Methods:

• Development of a maximum-likelihood algorithm to estimate parameters of linear transformations and the HRF.
• Integration of spatial smoothing as an instantaneous linear transformation within the modeling procedure.
• Constraining autoregressive parameters to match the canonical hemodynamic response function (HRF).

Main Results:

• The proposed method allows for the simultaneous estimation of optimal spatial smoothing and HRF parameters.
• For a motor-task fMRI dataset, weak but non-zero spatial smoothing was found to be optimal.
• The framework successfully demonstrated the estimation of activated brain regions.

Conclusions:

• The developed maximum-likelihood approach provides a rigorous method for optimizing fMRI preprocessing and modeling.
• Simultaneous estimation of spatial smoothing and HRF parameters enhances the accuracy of identifying brain activity.
• This methodology offers a robust framework for analyzing fMRI time series and detecting activated regions.