FMRI Data Analysis Preserving Map Variability Via Unsupervised Object-Centric Learning.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) generates complex volumetric data to map brain activity.
• Analyzing spatial neural activation maps in fMRI is crucial for understanding functional brain interconnections.
• Existing methods, like matrix factorization, have limitations in capturing detailed variabilities.

Purpose of the Study:

• To introduce a novel data-driven method for analyzing fMRI data.
• To accurately estimate variabilities in spatial neural activation maps from fMRI volumes.
• To demonstrate the superiority of the proposed method over conventional approaches.

Main Methods:

• A deep object-centric learning paradigm is employed.
• Individual fMRI volume components are treated as "objects."
• A set of autoencoders learns latent representations for these objects.

Main Results:

• The proposed method faithfully estimates variabilities in spatial neural activation maps.
• Numerical tests on synthetic and real fMRI data confirm the method's effectiveness.
• The approach shows advantages over existing matrix factorization-based methods.

Conclusions:

• The novel deep object-centric learning method offers a powerful tool for fMRI data analysis.
• This approach enhances the understanding of functional brain connectivity by accurately capturing neural activation patterns.
• The method provides a significant advancement over traditional analysis techniques.