用于可控皮肠溃疡病情合成的 Disentangled Anatomy-Disease Diffusion (DADD)
摘要
在保持患者特定解剖的同时，合成可控疾病阶段的纵向医学图像受阻于病理纹理和结构特征的缠合。我们针对皮肠溃疡 (UC) 内窥镜图像提出了此挑战，其中病情严重程度沿 Mayo 内窥镜评分 (MES) 的连续有序进展。我们的框架 Disentangled Anatomy-Disease Diffusion (DADD) 条件化地使用两个互补嵌入：用于患者解剖的预训练图像编码器和单独训练的累积病情严重程度的序数嵌入器。由于图像嵌入必然捕获疾病信息，我们引入了 Feature Purifier，即一种基于跨注意力的抹除机制，用于识别和抑制与疾病相关的通道，从而产生干净的解剖表示。这些清理后的解剖标记和目标疾病标记通过具备分辨率依赖路由门的 Triple-Pathway Cross-Attention 机制注入到去噪网络中。该架构利用 U-Net 层级结构，其中不同网络深度编码全局结构与细粒度病理纹理。Furthermore, we introduce Delta Steering, a training-free directional signal derived from the ordinal embeddings that enables explicit, single-pass control over disease transitions at inference without requiring additional forward passes. Validated on the LIMUC dataset, our approach produces high-fidelity images across all severity levels and effectively rebalances skewed class distributions, enhancing performance for downstream classification tasks. The dataset is available at zenodo.org/records/5827695 and the code base at github.com/umutdundar99/progressive-stable-diffusion
关键词
引用
@article{arxiv.2605.01848,
title = {Disentangled Anatomy-Disease Diffusion (DADD) for Controllable Ulcerative Colitis Progression Synthesis},
author = {Umut Dundar and Alptekin Temizel},
journal= {arXiv preprint arXiv:2605.01848},
year = {2026}
}
备注
10 pages, 6 figures. Code and dataset are publicly available. Accepted for presentation at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026, Synthetic Data for Computer Vision Workshop (SynData4CV)