Technology Category
Computer Vision: Learning & Optimization for CVSearch and Optimization: Non-convex OptimizationMachine Learning: Multi-instance/Multi-view Learning
LAMA: Stable Dual-Domain Deep Reconstruction For Sparse-View CT
🤖 AI Summary
This work addresses the ill-posed inverse problem of sparse-view CT reconstruction. We propose a Learned Alternating Minimization Algorithm (LAMA), introducing the first dual-domain (image + sinogram) variational modeling framework that incorporates a learnable nonconvex, nonsmooth regularizer, optimized via Nesterov smoothing to ensure convergence. LAMA deeply integrates deep residual networks with classical variational priors, thereby enhancing interpretability and stability without compromising theoretical rigor. Extensive experiments on multiple CT benchmark datasets demonstrate that LAMA significantly outperforms state-of-the-art methods in reconstruction accuracy and robustness, while simultaneously reducing model parameter count and GPU memory consumption—achieving an optimal balance between performance and computational efficiency.
Technology Category
Application Category
📝 Abstract
Inverse problems arise in many applications, especially tomographic imaging. We develop a Learned Alternating Minimization Algorithm (LAMA) to solve such problems via two-block optimization by synergizing data-driven and classical techniques with proven convergence. LAMA is naturally induced by a variational model with learnable regularizers in both data and image domains, parameterized as composite functions of neural networks trained with domain-specific data. We allow these regularizers to be nonconvex and nonsmooth to extract features from data effectively. We minimize the overall objective function using Nesterov's smoothing technique and residual learning architecture. It is demonstrated that LAMA reduces network complexity, improves memory efficiency, and enhances reconstruction accuracy, stability, and interpretability. Extensive experiments show that LAMA significantly outperforms state-of-the-art methods on popular benchmark datasets for Computed Tomography.
Problem
Research questions and friction points this paper is trying to address.
Solves sparse-view CT reconstruction using dual-domain learning
Combines data-driven and classical techniques for convergence
Enhances accuracy, stability, and interpretability in CT imaging
Innovation
Methods, ideas, or system contributions that make the work stand out.
Learned Alternating Minimization Algorithm (LAMA)
Dual-domain learnable regularizers with neural networks
Nesterov's smoothing and residual learning optimization
🔎 Similar Papers
