Anatomically informed bayesian model selection for fMRI group data analysis.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Statistical Modeling

Background:

• Standard voxel-based methods like Statistical Parametric Mapping (SPM) for fMRI group analysis have limitations in controlling both false positives and false negatives.
• Accurate identification of functional brain networks associated with cognitive tasks is crucial for understanding brain function.

Purpose of the Study:

• To introduce a novel Bayesian model selection framework for fMRI group data analysis.
• To overcome the limitations of standard voxel-based testing methods by controlling a Bayesian risk that balances false positives and false negatives.

Main Methods:

• Utilizing a Bayesian model selection framework based on posterior probabilities of mean region activations.
• Employing a pre-defined anatomical parcellation of the brain for network selection.
• Comparing the new approach with SPM-like methods on mental calculation experiment data.

Main Results:

• The Bayesian approach successfully detected the functional network associated with number processing.
• SPM-like approaches either missed or inaccurately identified activation regions in the same dataset.
• The Bayesian risk control demonstrated a superior balance between false positives and false negatives.

Conclusions:

• The proposed Bayesian framework offers a more robust and accurate method for fMRI group data analysis.
• This approach enhances the detection of task-related functional networks compared to traditional voxel-based methods.
• The ability to balance false positives and false negatives provides a more reliable assessment of brain activity in cognitive tasks.