CNN-Based Projected Gradient Descent for Consistent Image Reconstruction

Abstract

We present a new method for image reconstruction which replaces the projector in a projected gradient descent (PGD) with a convolutional neural network (CNN). CNNs trained as high-dimensional (image-to-image) regressors have recently been used to efficiently solve inverse problems in imaging. However, these approaches lack a feedback mechanism to enforce that the reconstructed image is consistent with the measurements. This is crucial for inverse problems, and more so in biomedical imaging, where the reconstructions are used for diagnosis. In our scheme, the gradient descent enforces measurement consistency, while the CNN recursively projects the solution closer to the space of desired reconstruction images. We provide a formal framework to ensure that the classical PGD converges to a local minimizer of a non-convex constrained least-squares problem. When the projector is replaced with a CNN, we propose a relaxed PGD, which always converges. Finally, we propose a simple scheme to train a CNN to act like a projector. Our experiments on sparse view Computed Tomography (CT) reconstruction for both noiseless and noisy measurements show an improvement over the total-variation (TV) method and a recent CNN-based technique.

References in corpus (2)

• Sparsity and Incoherence in Compressive Sampling
• A Review of Convolutional Neural Networks for Inverse Problems in Imaging

Cited by in corpus (37)

• On instabilities of deep learning in image reconstruction - Does AI come at a cost?
• DeepPET: A deep encoder-decoder network for directly solving the PET reconstruction inverse problem
• Computed Tomography Reconstruction Using Deep Image Prior and Learned Reconstruction Methods
• Deep Null Space Learning for Inverse Problems: Convergence Analysis and Rates
• CNN-Based Image Reconstruction Method for Ultrafast Ultrasound Imaging
• High-Resolution Limited-Angle Phase Tomography of Dense Layered Objects Using Deep Neural Networks
• Computationally Efficient Deep Neural Network for Computed Tomography Image Reconstruction
• Robust Phase Unwrapping via Deep Image Prior for Quantitative Phase Imaging
• Deep Learning for space-variant deconvolution in galaxy surveys
• Convolutional Proximal Neural Networks and Plug-and-Play Algorithms
• Low-Dose CT with Deep Learning Regularization via Proximal Forward Backward Splitting
• Image reconstruction algorithms in radio interferometry: from handcrafted to learned regularization denoisers
• Physics-/Model-Based and Data-Driven Methods for Low-Dose Computed Tomography: A survey
• Adaptive 3D descattering with a dynamic synthesis network
• Neural Networks-based Regularization for Large-Scale Medical Image Reconstruction
• CPGD: Cadzow Plug-and-Play Gradient Descent for Generalised FRI
• SGD-Net: Efficient Model-Based Deep Learning with Theoretical Guarantees
• Systematic Review on Learning-based Spectral CT
• Deep Neural Networks with Trainable Activations and Controlled Lipschitz Constant
• Solving Inverse Problems by Joint Posterior Maximization with Autoencoding Prior
• Wasserstein Patch Prior for Image Superresolution
• Fast and High-Quality Blind Multi-Spectral Image Pansharpening
• Two-and-a-half Order Score-based Model for Solving 3D Ill-posed Inverse Problems
• Big in Japan: Regularizing networks for solving inverse problems
• Momentum-Net: Fast and convergent iterative neural network for inverse problems
• Deep Learning for Biomedical Image Reconstruction: A Survey
• WPPNets and WPPFlows: The Power of Wasserstein Patch Priors for Superresolution
• A signal detection model for quantifying over-regularization in non-linear image reconstruction
• A plug-and-play framework for curvilinear structure segmentation based on a learned reconnecting regularization
• A Dataset-free Deep learning Method for Low-Dose CT Image Reconstruction
• Deep learning-based deconvolution for interferometric radio transient reconstruction
• Cross-domain Denoising for Low-dose Multi-frame Spiral Computed Tomography
• Swap-Net: A Memory-Efficient 2.5D Network for Sparse-View 3D Cone Beam CT Reconstruction
• A Statistical Framework to Investigate the Optimality of Signal-Reconstruction Methods
• Joint multiband deconvolution for Euclid and Vera C. Rubin images
• Meta Policy Switching for Secure UAV Deconfliction in Adversarial Airspace
• PlantPose: Universal Plant Skeleton Estimation via Tree-constrained Graph Generation