Shift-Invariant Canonical Polyadic Decomposition of Complex-Valued Multi-Subject fMRI Data With a Phase Sparsity...

Area of Science:

• Neuroimaging
• Signal Processing
• Computational Neuroscience

Background:

• Canonical polyadic decomposition (CPD) analyzes multi-subject complex-valued fMRI data, capturing shared spatial and temporal components using magnitude and phase.
• Standard CPD struggles with high noise and subject variability in fMRI data.
• Shift-invariant CPD addresses temporal variability, but spatial variability and denoising remain challenges.

Purpose of the Study:

• To develop an improved CPD method for complex-valued fMRI data by incorporating phase sparsity.
• To effectively denoise complex-valued components and model inter-subject spatial variability.
• To enhance the detection of task-related activations in fMRI studies.

Main Methods:

• Proposed a novel method combining shift-invariant CPD with a phase sparsity constraint on shared spatial maps.
• Estimated subject-specific time delays for complex-valued time courses using real-valued shift-invariant CPD.
• Applied a smoothed l0 norm for phase de-ambiguity and sparsity on complex-valued spatial maps.

Main Results:

• The proposed method demonstrated superior performance compared to existing complex-valued algorithms (tensor-based spatial ICA, shift-invariant CPD, unconstrained CPD).
• Significantly improved denoising and modeling of spatial variability in complex-valued fMRI components.
• Detected 178.7% more contiguous task-related activations compared to a real-valued shift-invariant CPD and ICA combination.

Conclusions:

• The phase sparsity constrained CPD effectively denoises complex-valued fMRI data and models spatial variability.
• This approach offers a significant advancement for analyzing multi-subject fMRI data, particularly for identifying task-related activations.
• The method shows promise for enhancing the sensitivity and robustness of neuroimaging analyses.