通过基于隐性扩散的数据增强来解决阶级不平衡,以改善儿童胸部X射线中的疾病分类.
Sivaramakrishnan Rajaraman1, Zhaohui Liang1, Zhiyun Xue1
1Division of Intramural Research, National Library of Medicine, National Institutes of Health, Bethesda, MD, USA.
概括
隐性扩散模型 (LDM) 产生合成胸部X射线,以解决深度学习中的数据不平衡,用于医学图像分类. 这种增强显著改善了诊断性能,增强了模型的概括性.
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:
- 深度学习 (DL) 模型与医学成像中常见的不平衡数据集作斗争.
- 合成数据增强可以提高DL模型的性能和概括性.
- 潜在扩散模型 (LDM) 显示了高质量的医学图像合成的潜力.
研究的目的:
- 评估以文本为导向的图像对图像LDMs的有效性,用于合成病情阳性胸部X射线 (CXRs).
- 用合成图像来增强儿科CXR数据集,以提高分类性能.
- 为了减轻阶级不平衡,提高医学图像分类任务的概括性.
主要方法:
- 在不平衡的CXR数据 (正常与肺炎/支气管肺炎) 上,Inception-V3模型的确定的基线性能.
- 精心调整的文本引导的LDMs产生合成CXR,描绘肺炎和支气管肺炎.
- 使用包括LDM合成图像在内的增强数据集重新训练Inception-V3模型.
主要成果:
- 通过LDM合成的图像增强显著改善了Youden指数 (p<0.05).
- 增量显著增强了其他分类指标,包括平衡的准确性,灵敏度,特异性,F-score,MCC和Kappa.
- 该战略有效地解决了阶级不平衡问题,并改善了模型通用化.
结论:
- 基于文本指导的LDM数据增强是一种可行的策略,可以改善对不平衡的医学成像数据集的深度学习分类.
- 与LDM合成疾病阳性CXR可以提高模型的稳定性和诊断准确性.
- 这种方法为克服医疗人工智能研究中的数据限制提供了一个有希望的解决方案.
相关概念视频
Imaging Studies for Cardiovascular System III: X-Ray
118
The most common cardiovascular diagnostic test is an X-ray. It produces images of the heart, blood vessels, and adjacent structures.
Definition and Purpose
An X-ray, or radiograph, is a non-invasive method that uses ionizing radiation to take images of internal structures. It is mainly used in cardiac imaging to examine the heart, lungs, and major blood vessels, aiming to identify abnormalities in the heart's size, shape, and position, such as heart failure, congenital defects, and vascular...
Definition and Purpose
An X-ray, or radiograph, is a non-invasive method that uses ionizing radiation to take images of internal structures. It is mainly used in cardiac imaging to examine the heart, lungs, and major blood vessels, aiming to identify abnormalities in the heart's size, shape, and position, such as heart failure, congenital defects, and vascular...
118
Lung Capacity
48.7K
The air in the lungs is measured in volumes and capacities. Lung volume measures reflect the amount of air taken in, released, or left over after a lung function, like a single inhalation. Lung capacity measures are sums of two or more lung volume measures.
48.7K


