无监督领域适应的结构保护约束 内血管细分 内血管细分
Sizhe Zhao1,2, Qi Sun1,2, Jinzhu Yang3,4
1Key Laboratory of Intelligent Computing in Medical Image, Ministry of Education, Northeastern University, Shenyang, Liaoning, China.
Medical & biological engineering & computing
|October 21, 2024
概括
通过在图像合成过程中保持结构,StruP-Net提高了对内血管细分的无监督域适应. 这种新的方法提高了细分性能,并显示了临床应用的潜力.
科学领域:
- 医疗成像医学成像
- 人工智能的人工智能
- 计算机视觉 计算机视觉
背景情况:
- 无监督域适应 (UDA) 方法减少了数据注释需求,但由于图像合成过程中的结构不匹配,与内血管细分斗争.
- 现有的UDA细分方法显示,细内血管细分任务的性能下降.
研究的目的:
- 提出一种新的UDA细分方法StruP-Net,通过结合结构保存方法来提高图像合成质量和细分性能.
- 为了应对内血管细分的UDA结构不匹配的挑战.
主要方法:
- 在StruP-Net中,用于图像合成的对抗性学习和用于语义一致性的两个域特定细分网络.
- 使用图形卷积网络 (GCNs) 来实现结构相似性的特征级结构保存 (F-SP).
- 实现基于结构相似性约束的感知损失的图像级结构保存 (I-SP).
主要成果:
- 在跨模式实验 (MRA到CTA) 中,StruP-Net与最先进的方法相比,实现了更高的细分性能.
- 该方法显示了高推理效率,表明了临床应用的潜力.
结论:
- StruP-Net有效地提高了UDA中的图像合成质量和细分性能,用于内血管细分.
- 提出的结构保存技术成功地缓解了结构不匹配问题,为医学图像分析提供了有前途的解决方案.
相关概念视频
00:49The Blood-brain Barrier
Overview
01:24Regulation of Angiogenesis and Blood Supply
Rapidly dividing tumors, embryos, and wounded tissues require more oxygen than usual, lowering the oxygen concentration in the blood. At low oxygen or hypoxic conditions, an oxygen-sensitive transcription factor called the hypoxia-inducible factor 1 or HIF1 is activated. HIF1 is a dimeric protein of alpha (ɑ) and beta (β) subunits. Under optimal oxygen conditions, HIF1β is present in the nucleus while HIF1ɑ remains in the cytosol. HIF1ɑ is hydroxylated by prolyl hydroxylase and factor...
01:15Structure of Blood Vessels
Blood is circulated throughout the human body through a network of blood vessels called the circulatory system. This system includes arteries that transport blood from the heart to various body parts. These arterial pathways divide into smaller vessels until they reach the arterioles, which further split into capillaries. It is within these minuscule capillaries that the exchange of nutrients and waste products takes place. After this exchange, the blood is collected by venules, which fuse to...