LUCID: Learned Undersampling-Adaptive Consistency-Guided Inference with Deterministic Flow Matching for Sparse-View CT Reconstruction

2026-06-15 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Medical Imaging & Reconstruction · Depth: Expert, quick

Summary

LUCID is a novel sparsity-adaptive, consistency-guided reconstruction framework designed for sparse-view Computed Tomography (CT). Sparse-view CT aims to reduce radiation dose and scanning time by acquiring fewer projection views, but this angular undersampling leads to severe ill-posed reconstruction problems, manifesting as streak artifacts, structural blurring, and loss of fine details. Unlike existing supervised methods tied to specific sampling settings or generative methods prone to anatomically inconsistent hallucinations under severe undersampling, LUCID utilizes a Flow Matching generative prior. It is trained exclusively on high-quality CT images to learn a continuous transport between a Gaussian distribution and the high-quality CT image distribution, independent of view sampling. During inference, LUCID explicitly incorporates the sampling sparsity level to adapt the generative trajectory of a single pretrained model, constructing an initial state via sparsity-weighted fusion of the sparse-view FBP image and Gaussian noise, performing sparsity-modulated Flow Matching updates, and applying projection-domain data-consistency correction. Experiments confirm LUCID achieves stable reconstruction performance across various sampling densities, enhancing image quality and structural fidelity while mitigating hallucination risks.

Key takeaway

For research scientists developing sparse-view CT reconstruction algorithms, LUCID presents a robust framework to overcome limitations of sampling-specific supervised methods and hallucination-prone generative models. You should consider integrating its sparsity-adaptive Flow Matching generative prior and iterative projection-domain data-consistency correction into your models. This approach can significantly improve image quality and structural fidelity across varying sampling densities, reducing artifacts and the risk of anatomically inconsistent structures in your medical imaging applications.

Key insights

LUCID employs Flow Matching and sparsity-adaptive consistency guidance for robust sparse-view CT reconstruction, reducing artifacts and hallucinations.

Principles

• Generative priors can adapt to varying data sparsity.
• Data-consistency correction enhances generative reconstruction.
• Training on high-quality data enables sampling-independent models.

Method

LUCID constructs a degradation-matched initial state, performs sparsity-modulated Flow Matching updates, and applies projection-domain data-consistency correction after each prior update.

In practice

• Integrate Flow Matching for ill-posed inverse problems.
• Employ sparsity-weighted fusion for initial states.
• Apply iterative data-consistency in generative models.

Topics

• Sparse-View CT
• CT Reconstruction
• Flow Matching
• Generative Models
• Data Consistency
• Medical Imaging

Best for: AI Scientist, Research Scientist, Computer Vision Engineer

Related on AIssential

• RA-CMF: Region-Adaptive Conditional MeanFlow for CT Image Reconstruction — RA-CMF enhances CT image quality and radiomic consistency by adaptively refining heterogeneous regions using conditional MeanFlow and reinforcement learning.
• Contrast-Informed Augmentation and Domain-Adversarial Training for Adult-to-Neonatal MR Reconstruction Generalization — Contrast-informed augmentation and domain-adversarial training enhance deep learning MR reconstruction generalization from adult to neonatal data.
• Margin-Consistent Deep Subtyping of Invasive Lung Adenocarcinoma via Perturbation Fidelity in Whole-Slide Image Analysis — Margin consistency with Perturbation Fidelity enhances deep learning robustness and accuracy for lung adenocarcinoma subtyping.
• GeoRect4D: Geometry-Compatible Generative Rectification for Dynamic Sparse-View 3D Reconstruction — GeoRect4D integrates explicit 3D consistency with generative refinement to reconstruct dynamic 3D scenes from sparse views.
• RelativeFlow: Taming Medical Image Denoising Learning with Noisy Reference — RelativeFlow uses flow matching to learn unified denoising from heterogeneous noisy medical image references.
• Multi-Contrast MRI Motion Correction via Parameter-Informed Disentanglement and Adaptive Experts — A unified MRI motion correction framework disentangles contrast and adapts to severity for robust, generalizable artifact removal.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.