Feature characterization in fMRI data: the Information Bottleneck approach.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Analysis

Background:

• Clustering is a flexible technique for analyzing functional Magnetic Resonance Imaging (fMRI) data.
• Traditional clustering faces challenges in determining quantization accuracy and the optimal number of clusters.
• Existing methods often do not account for the full statistical distribution of features.

Purpose of the Study:

• To present the application of the Information Bottleneck (IB) approach for fMRI data analysis.
• To leverage IB for principled vector quantization in neuroimaging.
• To improve the interpretation and comparison of fMRI datasets.

Main Methods:

• Utilized the Information Bottleneck (IB) approach for vector quantization.
• Applied IB clustering to group voxels based on their response magnitudes across experimental conditions.
• Employed an information-theoretic framework to guide the clustering procedure.

Main Results:

• The IB approach provides a consistent representation of fMRI data.
• This method addresses challenges in quantization accuracy and cluster number selection during clustering.
• It considers the complete statistical distribution of features within the feature space.

Conclusions:

• The Information Bottleneck method offers a robust and principled approach to fMRI data clustering.
• IB quantization facilitates easier interpretation and comparison of neuroimaging datasets.
• This technique enhances the analysis of voxel responses in fMRI studies.