人工智能培训

搜索

人工智能论文:分解操作:结构图像分解的三维医学图像中的可操作对象合成(Decompose to manipulate: Manipulable Object S

[复制链接]
mrlam 发表于 2018-12-6 11:52:09 | 显示全部楼层 |阅读模式
mrlam 2018-12-6 11:52:09 154 0 显示全部楼层
人工智能论文:分解操作:结构图像分解的三维医学图像中的可操作对象合成(Decompose to manipulate: Manipulable Object Synthesis in 3D Medical  Images with Structured Image Decomposition)医学图像分析系统的性能受到高质量图像注释量的限制。这些系统需要经过多年培训的专家对数据进行注释,尤其是在涉及诊断决策时。因此,这样的数据集很难扩展。在这种情况下,监督学习系统很难推广到训练集中罕见的情况,但会出现在现实世界的临床实践中。我们相信,由在真实数据上训练的系统生成的合成图像样本可用于改进医学图像分析应用中的监督学习任务。允许图像合成可被操纵可以帮助合成图像向训练数据提供补充信息,而不是简单地复制真实数据流形。在本文中,我们提出了一个框架,用于在具有可操纵属性的3D医学图像中合成3D对象,例如肺结节。通过将感兴趣对象分解为其分割掩模和包含残差信息的1D向量来实现操纵。通过两个对偶识别器将合成对象细化并混合到图像上下文中。我们评估了所提出的关于肺部结节的三维胸部CT图像的框架,并表明所提出的框架可以产生具有可操纵的形状,纹理和位置等的实际结节。在分类器训练期间从2800个3D CT体积的合成结节和真实结节进行Bysampling,我们展示了合成斑块可以通过平均8.44%的竞争性能指标(CPM)评分来提高整体结节检测性能。
The performance of medical image analysis systems is constrained by thequantity of high-quality image annotations.Such systems require data to beannotated by experts with years of training, especially when diagnosticdecisions are involved.Such datasets are thus hard to scale up.In thiscontext, it is hard for supervised learning systems to generalize to the casesthat are rare in the training set but would be present in real-world clinicalpractices.We believe that the synthetic image samples generated by a systemtrained on the real data can be useful for improving the supervised learningtasks in the medical image analysis applications.Allowing the image synthesisto be manipulable could help synthetic images provide complementary informationto the training data rather than simply duplicating the real-data manifold.Inthis paper, we propose a framework for synthesizing 3D objects, such aspulmonary nodules, in 3D medical images with manipulable properties.Themanipulation is enabled by decomposing of the object of interests into itssegmentation mask and a 1D vector containing the residual information.Thesynthetic object is refined and blended into the image context with twoadversarial discriminators.We evaluate the proposed framework on lung nodulesin 3D chest CT images and show that the proposed framework could generaterealistic nodules with manipulable shapes, textures and locations, etc. Bysampling from both the synthetic nodules and the real nodules from 2800 3D CTvolumes during the classifier training,we show the synthetic patches couldimprove the overall nodule detection performance by average 8.44% competitionperformance metric (CPM) score.人工智能论文:分解操作:结构图像分解的三维医学图像中的可操作对象合成(Decompose to manipulate: Manipulable Object Synthesis in 3D Medical  Images with Structured Image Decomposition) davVrSt5zmGSCkat.jpg
URL地址:https://arxiv.org/abs/1812.01737     ----pdf下载地址:https://arxiv.org/pdf/1812.01737    ----人工智能论文:分解操作:结构图像分解的三维医学图像中的可操作对象合成(Decompose to manipulate: Manipulable Object Synthesis in 3D Medical  Images with Structured Image Decomposition)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

mrlam当前离线
新手上路

查看:154 | 回复:0

快速回复 返回顶部 返回列表