RA-CMF: Region-Adaptive Conditional MeanFlow for CT Image Reconstruction

2026-04-25 · Source: cs.CV updates on arXiv.org · Field: Science & Research — Mathematics & Computational Sciences, Health & Medical Research · Depth: Expert, extended

Summary

A novel framework, Region-Adaptive Conditional MeanFlow (RA-CMF), has been developed for enhancing CT image reconstruction, particularly for lung cancer screening and diagnosis. This method addresses variability in CT images caused by different imaging protocols and scanner models, which can affect noise, contrast, and texture. RA-CMF integrates a conditional MeanFlow network, which models enhancement trajectories by predicting image-conditioned flow fields, with a regional reinforcement learning-driven policy network. The policy network adaptively allocates refinement budgets and stopping criteria tile-wise, focusing computational effort on complex structural regions like lung parenchyma and tumor ROIs. Evaluated on the National Lung Screening Trial (NLST) dataset, RA-CMF achieved an average radiomic feature Concordance Correlation Coefficient (CCC) of 0.96, an average PSNR of 31.30 ± 4.16, and an average SSIM of 0.94 ± 0.07 within tumor ROIs. Overall image quality improved to an average PSNR of 34.23 ± 1.71 and SSIM of 0.95 ± 0.01, demonstrating superior performance over baseline, STAN-CT, and CMF models.

Key takeaway

For Computer Vision Engineers developing medical image reconstruction pipelines, RA-CMF offers a robust approach to improve CT image quality and radiomic feature consistency. You should consider implementing region-adaptive enhancement strategies, especially for heterogeneous medical images, to focus computational resources on clinically relevant areas and achieve higher accuracy in quantitative analysis.

Key insights

RA-CMF enhances CT image quality and radiomic consistency by adaptively refining heterogeneous regions using conditional MeanFlow and reinforcement learning.

Principles

• Image degradation is often spatially non-uniform.
• Adaptive refinement improves local image quality.
• MeanFlow models learn average velocity fields for efficient generation.

Method

RA-CMF combines a conditional MeanFlow backbone for global enhancement with a reinforcement learning policy network that adaptively selects and refines image tiles based on local complexity, optimizing for quality, stability, and computational cost.

In practice

• Use conditional MeanFlow for image-to-image enhancement.
• Integrate reinforcement learning for adaptive spatial refinement.
• Prioritize refinement in complex anatomical regions.

Topics

• CT Image Reconstruction
• Conditional MeanFlow
• Reinforcement Learning
• Radiomic Feature Consistency
• Adaptive Spatial Refinement

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

Related on AIssential

• LUCID: Learned Undersampling-Adaptive Consistency-Guided Inference with Deterministic Flow Matching for Sparse-View CT Reconstruction — LUCID employs Flow Matching and sparsity-adaptive consistency guidance for robust sparse-view CT reconstruction, reducing artifacts and hallucinations.
• Parameter-Efficient Adaptation of SAM 3 for Automated ITV Generation from 4DCT Images — Parameter-efficient fine-tuning of SAM 3 with temporal coherence effectively automates ITV generation from 4DCT images using minimal data.
• RelativeFlow: Taming Medical Image Denoising Learning with Noisy Reference — RelativeFlow uses flow matching to learn unified denoising from heterogeneous noisy medical image references.
• Scaling Generative Foundation Models for Chest Radiography with Rectified Flow Transformers — A 1.3B-parameter generative foundation model for chest radiographs achieves expert-indistinguishable synthesis and controllable editing.
• CAHAL: Clinically Applicable resolution enHAncement for Low-resolution MRI scans — CAHAL enhances clinical MRI resolution by using physics-informed, native-space processing and a bivariate Mixture of Experts to prevent anatomical distortions.
• Physics-Driven Zero-Shot MRI Reconstruction with Non-local Image Priors — Combining physics-driven constraints with non-local image priors enhances zero-shot MRI reconstruction stability and performance.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.