跳转至

文章背景与核心概要

医学图像修复在重建病理区域健康组织、提升自动化脑部MRI分析性能方面具有巨大潜力。本文介绍了 RARF（Region-Aware Rectified Flows，区域感知纠正流），这是一个面向掩码数据生成的、任务无关的区域感知纠正流框架，并将其实例化应用于三维脑部MRI修复，作为作者参与 BraTS Inpainting Challenge 2026 的参赛方案。

该技术的核心创新在于将随机插值过程严格限制在目标修复区域内，同时保持观测到的体素固定不变，从而完美保留患者特有的解剖学上下文。模型采用三维神经网络、掩码流匹配（masked flow-matching）以及重建一致性目标，结合掩码感知预处理和数据增强，将初始高斯噪声引导转化为缺失或病变组织的合理且解剖结构一致的重建结果。

RARF: Region-Aware Rectified Flows for 3D Brain MRI Inpainting

RARF: Region-Aware Rectified Flows for 3D Brain MRI Inpainting

Summary

Summary

RARF (Region-Aware Rectified Flows) is a task-agnostic, region-aware rectified flow framework designed for masked data generation. In this work, the authors instantiate RARF for 3D brain MRI inpainting as their submission to the BraTS Inpainting Challenge 2026.

RARF (Region-Aware Rectified Flows) is a task-agnostic, region-aware rectified flow framework designed for masked data generation. In this work, the authors instantiate RARF for 3D brain MRI inpainting as their submission to the BraTS Inpainting Challenge 2026.

By restricting the stochastic interpolation process exclusively to the target inpainting region while keeping the observed voxels fixed, the model preserves patient-specific anatomical context. The framework utilizes a 3D neural network, masked flow-matching, and reconstruction-consistency objectives to transport initial Gaussian noise toward a plausible, anatomically consistent reconstruction of missing or pathological tissue.

By restricting the stochastic interpolation process exclusively to the target inpainting region while keeping the observed voxels fixed, the model preserves patient-specific anatomical context. The framework utilizes a 3D neural network, masked flow-matching, and reconstruction-consistency objectives to transport initial Gaussian noise toward a plausible, anatomically consistent reconstruction of missing or pathological tissue.

Metadata & Reference Information

Metadata & Reference Information

  • arXiv Identifier: arXiv:2609.03956 [cs.CV]
  • Authors:
  • Tomás Guija-Valiente
  • Blanca Rodriguez-Gonzalez
  • Norberto Malpica
  • Angel Torrado-Carvajal
  • Primary Subject: Computer Vision and Pattern Recognition (cs.CV)
  • Secondary Subjects: Artificial Intelligence (cs.AI), Machine Learning (cs.LG)
  • Submission Date: September 3, 2026
  • Source Code: Available on GitHub at github.com/TomasGuija/rarf
  • arXiv Identifier: arXiv:2609.03956 [cs.CV]
  • Authors:
  • Tomás Guija-Valiente
  • Blanca Rodriguez-Gonzalez
  • Norberto Malpica
  • Angel Torrado-Carvajal
  • Primary Subject: Computer Vision and Pattern Recognition (cs.CV)
  • Secondary Subjects: Artificial Intelligence (cs.AI), Machine Learning (cs.LG)
  • Submission Date: September 3, 2026
  • Source Code: Available on GitHub at github.com/TomasGuija/rarf

Abstract

Abstract

Medical image inpainting has the potential to improve automated brain MRI analysis by reconstructing healthy tissue within pathological regions. We introduce RARF, a task-agnostic region-aware rectified flow framework for masked data generation. We instantiate the framework for 3D brain MRI inpainting as our submission to the BraTS Inpainting Challenge 2026.

Medical image inpainting has the potential to improve automated brain MRI analysis by reconstructing healthy tissue within pathological regions. We introduce RARF, a task-agnostic region-aware rectified flow framework for masked data generation. We instantiate the framework for 3D brain MRI inpainting as our submission to the BraTS Inpainting Challenge 2026.

RARF restricts the stochastic interpolation process to the inpainting region, while the observed voxels remain fixed and provide patient-specific anatomical context. A three-dimensional neural network receives the partially voided image, with Gaussian noise filling the missing region, together with the inpainting mask and the corresponding timestep.

RARF restricts the stochastic interpolation process to the inpainting region, while the observed voxels remain fixed and provide patient-specific anatomical context. A three-dimensional neural network receives the partially voided image, with Gaussian noise filling the missing region, together with the inpainting mask and the corresponding timestep.

The model is trained using masked flow-matching and reconstruction-consistency objectives, combined with mask-aware preprocessing and data augmentation. During inference, the learned velocity field transports the initial noise toward a plausible reconstruction of the missing tissue, which is then combined with the unchanged observed anatomy. Experiments under the BraTS evaluation protocol show that the proposed approach produces competitive reconstructions while maintaining anatomical consistency.

The model is trained using masked flow-matching and reconstruction-consistency objectives, combined with mask-aware preprocessing and data augmentation. During inference, the learned velocity field transports the initial noise toward a plausible reconstruction of the missing tissue, which is then combined with the unchanged observed anatomy. Experiments under the BraTS evaluation protocol show that the proposed approach produces competitive reconstructions while maintaining anatomical consistency.