Low-Rank Tucker-2 Model for Multi-Subject fMRI Data Decomposition With Spatial Sparsity Constraint

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Tucker decomposition is used for analyzing multi-subject fMRI data to understand brain function.
• Existing methods struggle to extract common spatial and temporal patterns due to noise and data characteristics.

Purpose of the Study:

• To propose a novel low-rank Tucker-2 model with spatial sparsity for enhanced analysis of multi-subject fMRI data.
• To improve the extraction of common spatial and temporal brain activity patterns across subjects.

Main Methods:

• Developed a low-rank Tucker-2 model incorporating spatial sparsity constraints (l_p norm) and low-rank constraints (Frobenius norm).
• Utilized the alternating direction method of multipliers (ADMM) and half quadratic splitting for model optimization.
• Extracted spatial, temporal, and subject-specific features from the core tensor for subject classification.

Main Results:

• The proposed method demonstrated superior performance in extracting common spatial and temporal components compared to four established algorithms.
• Simulated and experimental fMRI data validated the effectiveness of the new approach.
• Extracted features showed promise in characterizing subjects within patient or healthy control groups.

Conclusions:

• The proposed sparse, low-rank Tucker decomposition model offers an improved approach for analyzing multi-subject fMRI data.
• This method enhances the extraction of common brain activity patterns, aiding in subject characterization and classification.