Detail publikace

Low-rank model for dynamic MRI: joint solving and debiasing

Originální název

Low-rank model for dynamic MRI: joint solving and debiasing

Anglický název

Low-rank model for dynamic MRI: joint solving and debiasing

Jazyk

en

Originální abstrakt

Reconstruction procedures from compressed-sensed MRI data are often treated as optimization problems. The most popular approach is to solve convex problems including the l1-norm. It is known that this type of regularization seeks for sparse solutions, however it gives biased estimates. Debiasing is a postprocessing procedure commonly used in many applications, especially where the optimization criterion is penalized least squares. In LASSO-type problems, debiasing is performed such that an additional least squares estimate is run while the non-sparse support is fixed. In low-rank modelling, l1-norm is applied on the singular values of a matrix. The debiasing procedure is more complicated, and especially, it turns out that it can amplify noise in the estimates. This abstract shows a method which debiases the estimates within a single procedure.

Anglický abstrakt

Reconstruction procedures from compressed-sensed MRI data are often treated as optimization problems. The most popular approach is to solve convex problems including the l1-norm. It is known that this type of regularization seeks for sparse solutions, however it gives biased estimates. Debiasing is a postprocessing procedure commonly used in many applications, especially where the optimization criterion is penalized least squares. In LASSO-type problems, debiasing is performed such that an additional least squares estimate is run while the non-sparse support is fixed. In low-rank modelling, l1-norm is applied on the singular values of a matrix. The debiasing procedure is more complicated, and especially, it turns out that it can amplify noise in the estimates. This abstract shows a method which debiases the estimates within a single procedure.

Dokumenty

BibTex


@inproceedings{BUT128720,
  author="Marie {Mangová} and Pavel {Rajmic}",
  title="Low-rank model for dynamic MRI: joint solving and debiasing",
  annote="Reconstruction procedures from compressed-sensed MRI data are often treated as optimization problems. The most popular approach is to solve convex problems including the l1-norm. It is known that this type of regularization seeks for sparse solutions, however it gives biased estimates. Debiasing is a postprocessing procedure commonly used in many applications, especially where the optimization criterion is penalized least squares. In LASSO-type problems, debiasing is performed such that an additional least squares estimate is run while the non-sparse support is fixed. In low-rank modelling, l1-norm is applied on the singular values of a matrix. The debiasing procedure is more complicated, and especially, it turns out that it can amplify noise in the estimates. This abstract shows a method which debiases the estimates within a single procedure.",
  address="Springer",
  booktitle="ESMRMB 2016, 33rd Annual Scientific Meeting, Vienna, AT, September 29--October 1: Abstracts, Friday",
  chapter="128720",
  doi="10.1007/s10334-016-0569-9",
  howpublished="online",
  institution="Springer",
  number="Supplement 1",
  year="2016",
  month="september",
  pages="200--201",
  publisher="Springer",
  type="conference paper"
}