PEAR: PEriodic And fixed Rank separation for fast fMRI

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Signal Processing

Background:

• Faster acquisition in functional MRI (fMRI) using data undersampling enhances spatial-temporal resolution and statistical robustness.
• Accurate reconstruction of undersampled fMRI data necessitates sophisticated data modeling techniques.

Purpose of the Study:

• To introduce PEAR (PEriodic And fixed Rank separation), a novel fMRI reconstruction approach designed for undersampled measurements.
• To model fMRI signals as a combination of periodic and fixed-rank components for improved reconstruction fidelity.

Main Methods:

• The PEAR approach decomposes fMRI signals into a fixed-rank component and a sum of periodic signals, sparse in the temporal Fourier domain.
• Reconstruction is achieved by solving a constrained optimization problem, enforcing specific rank and frequency limitations on the decomposed components.

Main Results:

• PEAR demonstrated superior performance in estimating timecourses and activation maps compared to existing methods in both simulated and real fMRI datasets.
• Significant improvements were observed at high acceleration ratios (R=6.66 to R=10.66).

Conclusions:

• PEAR offers higher fidelity reconstruction of undersampled fMRI data compared to fixed-rank models or conventional Low-rank + Sparse algorithms.
• Decomposing functional information into periodic and fixed-rank components enhances fMRI modeling and outperforms state-of-the-art methods.