Optimizing Data for Modeling Neuronal Responses.

Area of Science:

• Neuroimaging Statistics
• Computational Neuroscience

Background:

• Determining the best neuroimaging dataset for inferring neuronal responses is a critical challenge.
• Conventional statistical methods are inadequate for comparing datasets due to differing data requirements.
• Choosing appropriate imaging protocols and acquisition methods requires robust data evaluation techniques.

Purpose of the Study:

• To present Bayesian data comparison (BDC) as a principled framework for evaluating functional imaging data quality.
• To establish methods for assessing dataset precision in estimating neuronal connectivity parameters.
• To provide a means for disambiguating competing models using different datasets.

Main Methods:

• Utilizing Bayesian (probabilistic) forward models (e.g., GLMs, DCMs) to model neuronal responses for candidate datasets.
• Summarizing subject-specific model parameters at the group level using a Bayesian GLM.
• Introducing novel measures to evaluate datasets based on parameter estimation precision and model distinguishability.

Main Results:

• BDC provides a framework to evaluate and compare different functional imaging datasets.
• The proposed measures assess the precision of group-level parameter estimates and the ability to distinguish between similar models.
• Demonstrated the approach by comparing four multiband fMRI datasets and validated measures using simulations.

Conclusions:

• Bayesian data comparison (BDC) offers a robust solution for selecting optimal neuroimaging datasets.
• The framework enhances the precision of neuronal response inference and model comparison in functional imaging.
• Provided open-source Matlab code within the SPM software for reproducible analysis.