用SAM驱动的交叉提示与适应性采样一致性,用于半监督的医疗图像细分
Juzheng Miao1, Cheng Chen2, Yuchen Yuan1
1Department of Computer Science and Engineering, The Chinese University of Hong Kong, Hong Kong, China.
Medical image analysis
|February 20, 2026
概括
本研究介绍了CPAC-SAM,这是一种新的半监督学习方法,用于医疗图像细分,利用细分任何模型 (SAM). 它通过有效使用有限的标记数据和丰富的未标记数据,显著提高了细分精度.
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:
- 半监督学习 (SSL) 对于医疗图像细分至关重要,因为标记数据有限.
- 像分段任何模型 (SAM) 这样的视觉基础模型提供了更好的样本效率.
- 在医学成像中利用SSL的基础模型仍然是一个挑战.
研究的目的:
- 为半监督医疗图像细分提出一个新的SAM驱动框架 (CPAC-SAM).
- 通过交叉提示和自适应抽样,增强标记和未标记数据的学习.
- 提高快速可靠性和一致性,以实现强大的细分.
主要方法:
- 开发了一个SAM驱动的交叉提示框架,具有双分支结构.
- 实施了用于自适应提示生成的原型引导网格采样策略.
- 引入了快速一致性规范化,以降低SAM的灵敏度.
主要成果:
- 在五个医疗图像分割任务 (2D和3D) 中,CPAC-SAM在最先进的SSL方法中表现出优越的性能.
- 实现了显著的Dice改善,包括乳腺癌超过4.1%,左前庭细分3.8%.
- 在各种标记数据比率和模式中验证了有效性.
结论:
- CPAC-SAM有效地将基础模型集成到半监督的医疗图像细分中.
- 拟议的交叉提示,自适应抽样和一致性规范化提高了学习效率和准确性.
- 这一框架为推进医疗图像分析提供了一个有前途的方法,使用有限的标记数据.
相关概念视频
Positron Emission Tomography
Positron emission tomography (PET) is a medical imaging technique involving radiopharmaceuticals — substances that emit short-lived radiation. Although the first PET scanner was introduced in 1961, it took 15 more years before radiopharmaceuticals were combined with the technique and revolutionized its potential.
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body being...
One of the main requirements of a PET scan is a positron-emitting radioisotope, which is produced in a cyclotron and then attached to a substance used by the part of the body being...
Imaging Studies II: Positron Emission Tomography and Scintigraphy
Positron Emission Tomography (PET) is a medical imaging technique that provides crucial insights into the body's physiological functions at a molecular level. It is an indispensable resource for diagnosing, staging, and monitoring various illnesses, notably cancer, neurological disorders, and cardiovascular conditions.
Fundamental Principles of PET
Fundamental Principles of PET


