基于价值评估的磁共振单序列多层LVP图像压缩算法  

Compression Algorithm of Multi-Layer Low Value Point Image of Magnetic Resonance Single Sequence Based on Value Assessment

作　　者：满盼盼 赵国桦 刘泽华 王梅云[3] 林予松[2,4,5] Man Panpan;Zhao Guohua;Liu Zehua;Wang Meiyun;Lin Yusong(School of Information Engineering,Zhengzhou University,Zhengzhou 450001;Collaborative Innovation Center for Internet Healthcare,Zhengzhou University,Zhengzhou 450052;Department of Radiology,Henan Provincial People Jp Hospital,Zhengzhou 450003;School of Software,Zhengzhou University,Zhengzhou 450002;Hanwei IoTInstitute,Zhengzhou University,Zhengzhou 450002)

机构地区：[1]郑州大学信息工程学院,郑州450001 [2]郑州大学互联网医疗与健康服务河南省协同创新中心,郑州450052 [3]河南省人民医院放射科,郑州450003 [4]郑州大学软件学院,郑州450002 [5]郑州大学汉威物联网研究院,郑州450002

出　　处：《计算机辅助设计与图形学学报》2020年第7期1137-1144,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金面上项目(81772009);河南省科技厅科技攻关项目(182102310162).

摘　　要：针对影像组学过程中医生标注影像数据工作量巨大的问题,提出了一种基于价值评估的磁共振单序列多层低价值点(low value point,LVP)图像压缩算法.对单序列多层LVP原图像使用滤波压缩算法,对已标注的单序列多层LVP分割图像使用池化压缩算法,将单序列多层LVP原图像和分割图像压缩为单层高价值点(highvaluepoint,HVP)原图像和分割图像.将压缩算法应用于磁共振脑胶质瘤的自有数据集和公开数据集中,对比了压缩前原始图像标注时间和压缩成单层HVP图像标注时间,并测试压缩图像进行影像组学分级任务的效果.实验结果表明:压缩后标注一例病例的时间与压缩前相比显著减少,AUC等指标在执行文中算法和未执行下得到的分级结果相差甚微,均高于医生诊断水平.因此,在误差可接受范围内,医生仅需标注一层HVP图像便能进行医学影像分析,并且也能获得较好的分析结果.对于医生来说,可显著缩减医生标注任务的工作量.For the current workload of labeling image data for physicians in the process of radiomics,a compression method of multi-layer LVP image of single sequence based on value assessment is proposed.A filtering compression algorithm is used on the multi-layer LVP original image,the multi-layer LVP segmented image is subjected to a pooling compression algorithm.The compression algorithm is applied to the MRI glioma dataset to compare the labeling time of the original image before compression and the compressed single layer HVP image,and to test the effect of the compressed image on radiomics classification task.The experimental results show that the time to label a case after compression is significantly reduced,compared with that before compression,there is little difference between the result of AUC and that of the algorithm proposed in this paper.Thereby reduce the workload of data annotation significantly for physicians.

关 键 词：医学图像 图像压缩 磁共振图像 滤波压缩 池化压缩 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]