人工智能培训

搜索

人工智能论文:不共享患者数据的多机构深度学习建模:脑肿瘤分割的可行性研究(Multi-Institutional Deep Learning Modeling

[复制链接]
mxs810 发表于 2018-10-11 09:02:45 | 显示全部楼层 |阅读模式
mxs810 2018-10-11 09:02:45 120 0 显示全部楼层
人工智能论文:不共享患者数据的多机构深度学习建模:脑肿瘤分割的可行性研究(Multi-Institutional Deep Learning Modeling Without Sharing Patient Data:  A Feasibility Study on Brain Tumor Segmentation)用于图像语义分割的深度学习模型需要大量数据。在医学成像领域,获取足够的数据是非常重要的挑战。标记医学图像数据需要专业知识。各机构之间的协作可以解决这一挑战,但是向一个集中位置分享医疗数据会面临各种法律,隐私,技术和数据所有权方面的挑战,特别是在国际机构中。在本研究中,我们首次使用联合学习机构多机构协作,在不共享患者数据的情况下实现深度学习建模。我们的定量结果表明联合语义分段模型(Dice = 0.852)在多模式脑机上的表现类似于通过共享数据训练的模型(Dice = 0.862)。我们将联合学习与两种可选的协作学习方法进行比较,发现它们无法与联邦学习的性能相匹配。
Deep learning models for semantic segmentation of images require largeamounts of data.In the medical imaging domain, acquiring sufficient data is asignificant challenge.Labeling medical image data requires expert knowledge.Collaboration between institutions could address this challenge, but sharingmedical data to a centralized location faces various legal, privacy, technical,and data-ownership challenges, especially among international institutions.Inthis study, we introduce the first use of federated learning formulti-institutional collaboration, enabling deep learning modeling withoutsharing patient data.Our quantitative results demonstrate that the performanceof federated semantic segmentation models (Dice=0.852) on multimodal brainscans is similar to that of models trained by sharing data (Dice=0.862).Wecompare federated learning with two alternative collaborative learning methodsand find that they fail to match the performance of federated learning.人工智能论文:不共享患者数据的多机构深度学习建模:脑肿瘤分割的可行性研究(Multi-Institutional Deep Learning Modeling Without Sharing Patient Data:  A Feasibility Study on Brain Tumor Segmentation) OPMs2Dp2pZSMwg5M.jpg
URL地址:https://arxiv.org/abs/1810.04304     ----pdf下载地址:https://arxiv.org/pdf/1810.04304    ----人工智能论文:不共享患者数据的多机构深度学习建模:脑肿瘤分割的可行性研究(Multi-Institutional Deep Learning Modeling Without Sharing Patient Data:  A Feasibility Study on Brain Tumor Segmentation)
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

mxs810当前离线
新手上路

查看:120 | 回复:0

快速回复 返回顶部 返回列表