Meeting Banner
Abstract #3804

Accelerating Dynamic MRI Using Patch-Based Spatiotemporal Dictionaries

Yanhua Wang1, Leslie Ying1

1Department of Biomedical Engineering, Department of Electrical Engineering, The State University of New York at Buffalo, Buffalo, NY, United States

We propose a patch-based dictionary learning model for dynamic MRI reconstruction. The image sequence is divided into overlapping patches along both spatial and temporal directions. A set of temporal dependent dictionaries with three-dimensional atoms are adopted to provide sparse representations for compressed sensing reconstruction. This model adapts to specific local spatial-temporal features. Results on cardiac cine dataset demonstrate that the proposed method is capable of preserving both spatial structures and temporal variations.

Keywords

able abrupt accelerate accelerating alternating apply atom atoms available axes becomes biomedical bottom buffalo capture captures cardiac central cine composed compressed consecutive consider constraint contains correlations cosine crucial define degree denote denotes dependent details dictionaries dictionary dimension dimensional discrete divided dynamic eight either electrical employing engineering evaluate expected exploit extracts features field fifth fixed formulated frame frames framework free frequency fully gating grant ground heart in vivo indicate indicates initial initialization learning local located making many minimization model motion noises operation operator optimization original overlapping part patch patches pattern performance position precession preserve preserves problem procedure process profiles promising propose proposed random recently reconstruction reconstructions reduction redundant represent representations represented respectively retrospective sampling scanner seek segment segments selection sensing several simplified simulation sliding solutions solve solved space sparse sparsely sparseness sparsity spatial spatially split steady step steps structures subscript suitable suppose temporal temporally transform transforms truth tuning variation variations view whole window zoomed