Basis Expansion Approaches for Regularized Sequential Dictionary Learning Algorithms With Enforced Sparsity for fMRI...

Area of Science:

• Neuroimaging
• Machine Learning
• Signal Processing

Background:

• Functional magnetic resonance imaging (fMRI) data analysis commonly uses sequential dictionary learning.
• Existing methods often overlook the inherent temporal smoothness in fMRI data.
• This temporal structure, if incorporated, can enhance dictionary atom learning.

Purpose of the Study:

• To propose novel sequential dictionary learning algorithms for fMRI data analysis.
• To integrate the prior information of temporal smoothness into dictionary learning for fMRI.
• To improve the regularization of dictionary atoms by enforcing temporal smoothness.

Main Methods:

• Developed two new sequential dictionary learning algorithms tailored for fMRI data.
• Modified the dictionary update stage using a variant of the power method for SVD.
• Enforced temporal smoothness via regularization through basis expansion and sparse basis expansion.

Main Results:

• The proposed algorithms generate regularized dictionary atoms by solving a left regularized rank-one matrix approximation problem.
• Demonstrated the effectiveness of the algorithms on synthetic and real fMRI datasets.
• Showcased improved performance compared to classical dictionary learning approaches.

Conclusions:

• The novel algorithms effectively incorporate temporal smoothness into dictionary learning for fMRI.
• This approach enhances dictionary atom regularization, leading to improved fMRI data analysis.
• The methods offer a promising advancement for neuroimaging data processing and interpretation.