FISTA-Net: Learning a Fast Iterative Shrinkage Thresholding Network for Inverse Problems in Imaging
FISTA-Net: обучение сети быстрого итеративного усечения и пороговой обработки для решения обратных задач визуализации
2021-01-25
SCID: 54.1/cn69qk7a
Discuss with AI
Electromagnetic TomographyFISTA-Netalgorithm unfoldinginverse problems in imagingproximal mapping
Figures from the paper
Abstract (AI)
Inverse problems are essential to imaging applications. In this letter, we propose a model-based deep learning network, named FISTA-Net, by combining the merits of interpretability and generality of the model-based Fast Iterative Shrinkage/Thresholding Algorithm (FISTA) and strong regularization and tuning-free advantages of the data-driven neural network. By unfolding the FISTA into a deep network, the architecture of FISTA-Net consists of multiple gradient descent, proximal mapping, and momentum modules in cascade. Different from FISTA, the gradient matrix in FISTA-Net can be updated during iteration and a proximal operator network is developed for nonlinear thresholding which can be learned through end-to-end training. Key parameters of FISTA-Net including the gradient step size, thresholding value and momentum scalar are tuning-free and learned from training data rather than hand-crafted. We further impose positive and monotonous constraints on these parameters to ensure they converge properly. The experimental results, evaluated both visually and quantitatively, show that the FISTA-Net can optimize parameters for different imaging tasks, i.e. Electromagnetic Tomography (EMT) and X-ray Computational Tomography (X-ray CT). It outperforms the state-of-the-art model-based and deep learning methods and exhibits good generalization ability over other competitive learning-based approaches under different noise levels.
Key Findings
1
Experiments on Electromagnetic Tomography and X-ray CT show that FISTA-Net optimizes parameters across imaging tasks and outperforms state-of-the-art model-based and deep learning methods.
2
FISTA-Net maintains good generalization compared with other learning-based approaches under different noise levels.
3
FISTA-Net unfolds the model-based FISTA algorithm into a deep network combining gradient descent, proximal mapping, and momentum modules.
4
Gradient step sizes, thresholding values, and momentum scalars are learned from data without manual tuning, with positive and monotonous constraints imposed to promote convergence.
5
Unlike standard FISTA, FISTA-Net learns iteration-dependent gradient matrices and nonlinear proximal operators through end-to-end training.
Research Object
inverse imaging problems, specifically Electromagnetic Tomography (EMT) and X-ray Computed Tomography (X-ray CT) reconstruction
Research Subject
the reconstruction performance, parameter learning, convergence, and generalization of the FISTA-Net model across imaging tasks and noise levels
Publication Details
Publication Date
2021-01-25
Journal
Publisher
ISSN
Cited by
397
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest