The Voxelwise Encoding Model framework: A tutorial introduction to fitting encoding models to fMRI data.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Cognitive Neuroscience

Background:

• Functional brain mapping is crucial for understanding cognition.
• Existing methods for analyzing neuroimaging data can suffer from overfitting and limited feature capacity.
• The Voxelwise Encoding Model (VEM) framework offers a powerful alternative for predicting brain activity from complex stimuli.

Purpose of the Study:

• To demystify the Voxelwise Encoding Model (VEM) framework.
• To provide hands-on tutorials for novice practitioners to implement VEM.
• To facilitate the wider adoption and dissemination of VEM in neuroimaging research.

Main Methods:

• Utilizing a Voxelwise Encoding Model (VEM) approach where features from stimuli predict voxel-wise brain activity.
• Fitting separate encoding models for each spatial sample (voxel).
• Employing free, open-source tools and public datasets for reproducible analysis.

Main Results:

• VEM enables the use of a large number of features, accommodating complex naturalistic stimuli and tasks.
• High-dimensional functional maps are generated, reflecting voxel selectivity to numerous features.
• Model performance evaluation on separate test datasets minimizes overfitting and generalizes results to new subjects and stimuli.

Conclusions:

• VEM is a robust framework for functional brain mapping with significant advantages over traditional methods.
• The provided tutorials aim to lower the barrier to entry for VEM implementation.
• Increased accessibility is expected to promote broader use of VEM in analyzing complex neuroimaging data.