海量医学影像自动语义标注研究 - 中国数字医学.pdf

信息化论坛 Information Forum 海量医学影像自动语义标注研究 万艳丽① 郭珉江① 马豪① 郭琳① 胡红濮 ＊ 摘 要 随着医学影像数量的几何级增长，如何有效标注海量医学影像资源，并使其在检索及辅助诊断中得到充分利用是 大数据时代的迫切需求。针对医学影像语义标注现状，提出了一种医学影像语义描述模型和基于深度学习的自动语义标注 框架，一方面将人从费时费力的手工标注中解脱出来，另一方面通过建立低层视觉特征与高层语义的映射，克服视觉特征 与语义表达之间的鸿沟，使基于语义的海量医学影像数据的检索应用得以实现。 关键词　PACS 视觉特征 语义特征 语义标注 图像检索 Doi:10.3969/j.issn.1673-7571.2015.05.027 [中图分类号] R319;TP391.4 [文献标识码] A Research on Automatic Semantic Annotation of the Massive Medical Images / WAN Yan-li, GUO Min-jiang, MA Hao, et al// China Digital Medicine.-2015 10(5): 83 to 85 Abstract With the number of medical image increasing in a geometric progression, how to carry out efficient annotation and make it play a maximum role in clinical decision support is the urgent demand of the age of big data. Based on the existing problems of the semantic annotation of medical images, this paper proposes a semantic description model and automatic semantic annotation framework based on the deep learning. On one hand, it frees man from the time-consuming manual annotation. On the other hand, through the building mapping between low-level visual features and high-level semantic, it overcomes the gap between the visual feature and semantic expression, which promotes the application of retrieval based on semantic of the massive medical images. Keywords PACS, visual feature, semantic feature, semantic annotation, image retrieval Fund project Social Science Fund Project-Research on Medical Knowledge Organization System Mapping Model and Application Service in Health Insurance (No. 13CTQ032) Corresponding author Institute of Information on Medicine, China Academy of Chinese Medical Sciences, Beijing 100020, P.R.C. 随着多媒体技术与互联网应用的不断发展，医院影像存档与通讯系统（PACS）的成熟和普及，医学影像的数量呈几 何级增长。因此对这些海量医学影像资源的有效组织和标注，并使其在影像检