Bayesian population receptive field modelling.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Neuroimaging

Background:

• Population receptive field (pRF) mapping is crucial for understanding visual cortex organization.
• Existing methods often lack robust uncertainty quantification and formal model comparison capabilities.

Purpose of the Study:

• To introduce a probabilistic (Bayesian) framework and software toolbox for pRF mapping using fMRI.
• To enable flexible modeling of stimuli across dimensions and account for neurovascular coupling variability.
• To provide formal hypothesis testing for comparing different pRF models.

Main Methods:

• A Bayesian framework with generative (encoding) models for fMRI time series.
• Estimation of neuronal and hemodynamic parameters using variational Laplace.
• Quantification of parameter variance/covariance for uncertainty representation.
• Model comparison using log model evidence (variational free energy).

Main Results:

• Demonstrated a generic approach applicable to stimuli of any dimension, validated with 2D retinotopic mapping.
• Showcased the ability to plot receptive fields with uncertainty quantification for size and location.
• Found strong correlations between pRF size and neuronal scaling parameters, handled by the Bayesian approach.
• Identified a circular Difference of Gaussians (DoG) model as the best fit for 7T fMRI data.

Conclusions:

• The developed Bayesian framework offers a robust and flexible tool for pRF mapping.
• The framework provides improved uncertainty estimation and formal model comparison for receptive field analysis.
• This approach advances the mapping of stimulus spaces onto brain anatomy, applicable to various dimensions.