PEAR: PEriodic And fixed Rank separation for fast fMRI

摘要: In functional MRI (fMRI), faster acquisition via undersampling of data can improve the spatial-temporal resolution trade-off and increase statistical robustness through increased degrees-of-freedom. High quality reconstruction fMRI from undersampled measurements requires proper modeling data. We present an approach based on signal as a sum periodic fixed rank components, for improved measurements. decompose into component which has consisting signals is sparse in temporal Fourier domain. Data performed by solving constrained problem that enforces fixed, moderate one limited number frequencies other. Our coined PEAR - PEriodic And Rank separation fast fMRI. Experimental results include purely synthetic simulation, simulation with real timecourses retrospective dataset. Evaluation was both quantitatively visually versus ground truth, comparing to two additional recent methods Results demonstrate PEAR's improvement estimating activation maps compared against at acceleration ratios R=8,16 (for simulated data) R=6.66,10 data). higher fidelity than when using fixed-rank model or conventional Low-rank+Sparse algorithm. have shown splitting information between components leads better fMRI, over state-of-the-art methods.