Solving ill-posed inverse problems using iterative deep neural networks

Abstract

We propose a partially learned approach for the solution of ill posed inverse problems with not necessarily linear forward operators. The method builds on ideas from classical regularization theory and recent advances in deep learning to perform learning while making use of prior information about the inverse problem encoded in the forward operator, noise model and a regularizing functional. The method results in a gradient-like iterative scheme, where the "gradient" component is learned using a convolutional network that includes the gradients of the data discrepancy and regularizer as input in each iteration. We present results of such a partially learned gradient scheme on a non-linear tomographic inversion problem with simulated data from both the Sheep-Logan phantom as well as a head CT. The outcome is compared against FBP and TV reconstruction and the proposed method provides a 5.4 dB PSNR improvement over the TV reconstruction while being significantly faster, giving reconstructions of 512 x 512 volumes in about 0.4 seconds using a single GPU.

Cited by in corpus (70)

• Physics-Constrained Deep Learning for High-dimensional Surrogate Modeling and Uncertainty Quantification without Labeled Data
• On instabilities of deep learning in image reconstruction - Does AI come at a cost?
• Learned Primal-dual Reconstruction
• A physics-informed variational DeepONet for predicting the crack path in brittle materials
• Internet of Underwater Things and Big Marine Data Analytics -- A Comprehensive Survey
• Deep D-bar: Real time Electrical Impedance Tomography Imaging with Deep Neural Networks
• Model based learning for accelerated, limited-view 3D photoacoustic tomography
• FISTA-Net: Learning A Fast Iterative Shrinkage Thresholding Network for Inverse Problems in Imaging
• Shallow Neural Networks for Fluid Flow Reconstruction with Limited Sensors
• A Review on Deep Learning in Medical Image Reconstruction
• Computed Tomography Reconstruction Using Deep Image Prior and Learned Reconstruction Methods
• Semi-supervised Sequence Modeling for Elastic Impedance Inversion
• Deep Learning in Photoacoustic Tomography: Current approaches and future directions
• Solving Inverse Problems With Deep Neural Networks -- Robustness Included?
• Learning with Known Operators reduces Maximum Training Error Bounds
• Learning The Invisible: A Hybrid Deep Learning-Shearlet Framework for Limited Angle Computed Tomography
• Learning Deep Implicit Fourier Neural Operators (IFNOs) with Applications to Heterogeneous Material Modeling
• The LoDoPaB-CT Dataset: A Benchmark Dataset for Low-Dose CT Reconstruction Methods
• Limited View and Sparse Photoacoustic Tomography for Neuroimaging with Deep Learning
• Deep-Learning based Inverse Modeling Approaches: A Subsurface Flow Example
• Deep Null Space Learning for Inverse Problems: Convergence Analysis and Rates
• Spatio-Temporal Deep Learning-Based Undersampling Artefact Reduction for 2D Radial Cine MRI with Limited Data
• Solving inverse problems using conditional invertible neural networks
• Numerical Solution of Inverse Problems by Weak Adversarial Networks
• A Partially Learned Algorithm for Joint Photoacoustic Reconstruction and Segmentation
• Structural engineering from an inverse problems perspective
• SOM-Net: Unrolling the Subspace-based Optimization for Solving Full-wave Inverse Scattering Problems
• Deep Learning for space-variant deconvolution in galaxy surveys
• Nonlocal Kernel Network (NKN): a Stable and Resolution-Independent Deep Neural Network
• A Helmholtz equation solver using unsupervised learning: Application to transcranial ultrasound
• Task adapted reconstruction for inverse problems
• Neural Networks-based Regularization for Large-Scale Medical Image Reconstruction
• Rethinking the ill-posedness of the spectral function reconstruction -- why is it fundamentally hard and how Artificial Neural Networks can help
• SGD-Net: Efficient Model-Based Deep Learning with Theoretical Guarantees
• Solving Inverse Problems with Hybrid Deep Image Priors: the challenge of preventing overfitting
• Infinite-dimensional inverse problems with finite measurements
• Limitations of Deep Learning for Inverse Problems on Digital Hardware
• A data-driven iteratively regularized Landweber iteration
• Rotation Equivariant Proximal Operator for Deep Unfolding Methods in Image Restoration
• Supervised learning in Hamiltonian reconstruction from local measurements on eigenstates
• Data driven regularization by projection
• A Neural Network Warm-Start Approach for the Inverse Acoustic Obstacle Scattering Problem
• Unified Supervised-Unsupervised (SUPER) Learning for X-ray CT Image Reconstruction
• Data-driven nonsmooth optimization
• Fourier Neural Operator Networks: A Fast and General Solver for the Photoacoustic Wave Equation
• Learning and correcting non-Gaussian model errors
• Big in Japan: Regularizing networks for solving inverse problems
• RAGO: Recurrent Graph Optimizer For Multiple Rotation Averaging
• Approximate k-space models and Deep Learning for fast photoacoustic reconstruction
• Unsupervised Knowledge-Transfer for Learned Image Reconstruction
• Application of machine learning regression models to inverse eigenvalue problems
• Inverse problems on low-dimensional manifolds
• Efficient Physics-Based Learned Reconstruction Methods for Real-Time 3D Near-Field MIMO Radar Imaging
• The Deep Arbitrary Polynomial Chaos Neural Network or how Deep Artificial Neural Networks could benefit from Data-Driven Homogeneous Chaos Theory
• Semi-Blind Source Separation with Learned Constraints
• Unrolled Optimization with Deep Learning-based Priors for Phaseless Inverse Scattering Problems
• Spatiotemporal PET reconstruction using ML-EM with learned diffeomorphic deformation
• Learned Interferometric Imaging for the SPIDER Instrument
• A data-assisted two-stage method for the inverse random source problem
• Learning phase field mean curvature flows with neural networks
• Accelerating innovation with software abstractions for scalable computational geophysics
• A Fast and Adaptive Algorithm to Compute the X-ray Transform
• Electron-boson spectral density functions of cuprates obtained from optical spectra via machine learning
• Equivariant neural networks for inverse problems
• Stability Bounds for the Unfolded Forward-Backward Algorithm
• Regularization of Inverse Problems: Deep Equilibrium Models versus Bilevel Learning
• Assessment of Deep-Learning Methods for the Enhancement of Experimental Low Dose Dental CBCT Volumes
• Fast Inexact Bilevel Optimization for Analytical Deep Image Priors
• TomoSelfDEQ: Self-Supervised Deep Equilibrium Learning for Sparse-Angle CT Reconstruction
• DAWN-FM: Data-Aware and Noise-Informed Flow Matching for Solving Inverse Problems