视觉基础模型用于3D磁共振成像细分,分类和注册
Shansong Wang1, Mojtaba Safari1, Qiang Li1
1Department of Radiation Oncology, Emory University School of Medicine, Atlanta, 30322, GA, USA.
Medical image analysis
|February 17, 2026
概括
在3DMRI数据上预训练的视觉基础模型 (VFMs) 改善了细分,分类和注册任务. 新的VFMTriad通过利用最大的3DMRI预训练数据集来提高下游医学成像应用的性能.
科学领域:
- 人工智能的人工智能
- 医疗成像医学成像
- 计算机视觉 计算机视觉
背景情况:
- 视觉基础模型 (VFMs) 为下游任务提供一般表示.
- 现有的VFMs通常在非MRI模式上接受过预先培训,这限制了MRI应用的性能.
- 图像原理和数据分布的差异阻碍了MRI中的VFM多功能性.
研究的目的:
- 介绍Triad,一个专门用于3DMRI细分,分类和注册的VFM.
- 利用最大的3DMRI预训练数据集 (Triad-129K) 进行强大的表示学习.
- 在3DMRI应用中提高VFM性能和多功能性.
主要方法:
- 在129K 3D MRI卷上使用SimMIM框架开发了Triad.
- 使用模态和成像参数的文本描述的受限制的语义分布.
- 在25个下游数据集上对Triad进行了评估,涉及细分,分类和注册任务.
主要成果:
- 在17个数据集上,nnUNet-Triad-SimMIM比nnUNet-Scratch提高了2.13%的细分.
- 在5个数据集上,Swin-B-Triad-SimMIM在分类方面比Swin-B-Scratch有4.38%的改进.
- 在2个数据集上,SwinUNETR-Triad-SimMIM比SwinUNETR-Scratch提升了3.84%的注册.
结论:
- 对3DMRI数据进行大规模的预训练显著提高了下游任务的性能.
- 特里亚德展示了MRI特定的VFMs在医学成像应用中的价值.
- 在预培训和下游任务之间,一致的数据模式和器官特征是提高绩效的关键.
相关概念视频
Magnetic Resonance Imaging
9.9K
Magnetic resonance imaging (MRI) is a noninvasive medical imaging technique based on a phenomenon of nuclear physics discovered in the 1930s, in which matter exposed to magnetic fields and radio waves was found to emit radio signals. In 1970, a physician and researcher named Raymond Damadian noticed that malignant (cancerous) tissue gave off different signals than normal body tissue. He applied for a patent for the first MRI scanning device in clinical use by the early 1980s. The early MRI...
9.9K
Imaging Studies IV: Magnetic Resonance Imaging
299
Introduction:Magnetic Resonance Imaging, or MRI, can include a specialized imaging technique of the urinary system known as Magnetic Resonance Urography (MRU). This radiation-free technique uses strong magnetic fields and radio waves to produce detailed images with the help of a computer. MRU is particularly effective for visualizing fluid-filled structures like the kidneys, ureters, and bladder.Applications of MRI in the Genitourinary SystemKidneys and Ureters: MRI detects tumors, cysts,...
299


