| Popis: |
Accelerated resting-state functional magnetic resonance imaging (R-fMRI) can provide higher spatial resolution and improved brain connectivity maps. Current methods for fast R-fMRI rely on either fully-sampled parallel imaging or undersampled reconstruction using signal priors, but not both. We propose a novel Bayesian reconstruction framework that combines simultaneous multislice (SMS) imaging, controlled aliasing, and undersampling in k-space and time to reconstruct high-quality signals and connectivity maps. We use a generative dictionary model on R-fMRI time-series, which is robust to signal fluctuations and artifacts, adapts to inter-subject variations through optimized similarity transforms on its atoms, and uses spatially regularized sparsity. Results on simulated and clinical R-fMRI show that our method gives more accurate reconstructions and connectivity maps than the state of the art, and can enable higher spatial resolution. |
