Premium
Sci‐Fri PM Imaging‐01: Comparison of POCS and cTERA Image Reconstruction Algorithms Applied to 3D Sparsely Sampled k‐Space Data
Author(s) -
Peng H,
Sabati M,
Lauzon M,
Frayne R
Publication year - 2006
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2244676
Subject(s) - k space , algorithm , iterative reconstruction , sampling (signal processing) , image quality , computer science , projection (relational algebra) , compressed sensing , focus (optics) , artificial intelligence , computer vision , fourier transform , reconstruction algorithm , image (mathematics) , mathematics , physics , optics , mathematical analysis , filter (signal processing)
Real‐time 3D magnetic resonance (MR) imaging sometimes focus on rapid data acquisition by full sampling the central zone of phase‐encoding plane while sparse sampling its periphery, which results in the formation of sparsely sampled MR raw data. This research compares the performance of image reconstruction from sparsely sampled MR data by the often‐used zero filling (ZF) algorithm, and by two new methods, i.e. , projection‐onto‐convex sets (POCS) and constrained transient error reconstruction approach (cTERA) algorithms. It is found that both the POCS and cTERA algorithms reconstruct high‐quality images that greatly improve the depiction of high‐frequency‐containing structures compared to ZF. However, POCS and cTERA take significantly increased computational times. Due to their intrinsic complexities compared to conventional Fourier transform and ZF reconstruction, POCS and cTERA algorithms are currently not implemented on commercial clinical MR scanners. It is also demonstrated that for a given scan time, it is possible to find an optimal specific sparse sampling strategies by both POCS and cTERA. Under pseudo optimal conditions, the corresponding POCS and cTERA images demonstrated better high‐resolution image quality compared to other, less optimal, sampling strategies. Therefore, it is desirable to acquire some high‐frequency data, as opposed to spending time only collecting an enlarged central region.