PURPOSE: To improve the graph model of our previous work GOOSE for fat-water decomposition with higher computational efficiency and quantitative accuracy. METHODS: A modification of the GOOSE fat water decomposition algorithm is introduced while the global convergence guarantees of GOOSE are still inherited to minimize fat-water swaps and phase wraps. In this paper, two non-equidistant graph optimization frameworks are proposed as a single-step framework termed as rapid GOOSE (R-GOOSE), and a multi-step framework termed as multi-scale R-GOOSE (mR-GOOSE). Both frameworks contain considerably less graph connectivity than GOOSE, resulting in a great computation reduction thus making it readily applicable to multidimensional fat water applications. The quantitative accuracy and computational time of the novel frameworks are compared with GOOSE on the 2012 ISMRM Challenge datasets to demonstrate the improvement in performance. RESULTS: Both frameworks accomplish the same level of high accuracy as GOOSE among all datasets. Compared to 100 layers in GOOSE, only 8 layers were used in the new graph model. Computational time is lowered by an order of magnitude to around 5 s for each dataset in (mR-GOOSE), R-GOOSE achieves an average run-time of 8 s. CONCLUSION: The proposed method provides fat-water decomposition results with a lower run-time and higher accuracy compared to the previously proposed GOOSE algorithm. Magn Reson Med 79:2401-2407, 2018.
PURPOSE: To improve the graph model of our previous work GOOSE for fat-water decomposition with higher computational efficiency and quantitative accuracy. METHODS: A modification of the GOOSEfatwater decomposition algorithm is introduced while the global convergence guarantees of GOOSE are still inherited to minimize fat-water swaps and phase wraps. In this paper, two non-equidistant graph optimization frameworks are proposed as a single-step framework termed as rapid GOOSE (R-GOOSE), and a multi-step framework termed as multi-scale R-GOOSE (mR-GOOSE). Both frameworks contain considerably less graph connectivity than GOOSE, resulting in a great computation reduction thus making it readily applicable to multidimensional fatwater applications. The quantitative accuracy and computational time of the novel frameworks are compared with GOOSE on the 2012 ISMRM Challenge datasets to demonstrate the improvement in performance. RESULTS: Both frameworks accomplish the same level of high accuracy as GOOSE among all datasets. Compared to 100 layers in GOOSE, only 8 layers were used in the new graph model. Computational time is lowered by an order of magnitude to around 5 s for each dataset in (mR-GOOSE), R-GOOSE achieves an average run-time of 8 s. CONCLUSION: The proposed method provides fat-water decomposition results with a lower run-time and higher accuracy compared to the previously proposed GOOSE algorithm. Magn Reson Med 79:2401-2407, 2018.
Authors: Abraam S Soliman; Jing Yuan; Karl K Vigen; James A White; Terry M Peters; Charles A McKenzie Journal: Magn Reson Med Date: 2013-09-04 Impact factor: 4.668
Authors: Thorsten A Bley; Oliver Wieben; Christopher J François; Jean H Brittain; Scott B Reeder Journal: J Magn Reson Imaging Date: 2010-01 Impact factor: 4.813