Open AccessA Modified GAN for Compressed Sensing MRI
Author(s) -
Zelong Geng,
Jinxu Tao,
Jinxin Xu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1642/1/012001
Subject(s) - compressed sensing , computer science , residual , sampling (signal processing) , artificial intelligence , generator (circuit theory) , artificial neural network , iterative reconstruction , deep learning , image (mathematics) , computer vision , pattern recognition (psychology) , algorithm , power (physics) , physics , filter (signal processing) , quantum mechanics
Magnetic resonance imaging is a commonly used diagnosis method in medicine. Most reconstruction methods are based on the compressed sensing theory, while it is inefficient and time-consuming. In recent years, deep neural networks have developed rapidly. GAN architecture has been widely used in various image tasks after published. This paper proposes a new MRI reconstruction method RISEGAN. The generator uses U-Net structure to extract multi-scale features in the down sampling and up sampling modules. Combined with the residual learning and squeeze excitation blocks, the mapping between the under sampled and the fully sampled image is established. Experiment results show that our method can finish reconstruction with premium quality. Evaluation indicators have also improved.