摘要
在胃肠道检查过程中,胶囊内窥镜生成的数十万帧中的大量冗余图像对临床医生在找出病灶图像时产生了极大干扰。为提高医生诊疗效率并缩短阅片时间,提出了一种图像块信息相似性的评估算法对冗余图像进行评估并去除。算法将图像分为几个图像块,并对每个区域求出亮度、二维熵值和结构信息;合并所有图像块信息后作为该图像的唯一识别码并存入json文件中;通过调用json文件中两帧的相应识别码计算出相似度度量值,筛除冗余图像。对2018年从长海医院采集的800帧小肠内窥镜图像进行扩增后进行了冗余图像去除实验。实验结果表明,该算法具有0.03 s/帧的快速数据筛查速度,实验召回率达到99.14%,准确率达到98.42%,F-measure达到98.01%。经验证可知,在有效保留了图像的原始信息同时算法有效提高了筛查速度,对复杂多变的消化道图像具有良好的去冗余性能。
During gastrointestinal examination, capsule endoscopy generates hundreds of thousands of data. When clinicians examine these data, a large number of redundant images will cause great interference. In order to improve the diagnosis efficiency of doctors, proposes an evaluation algorithm of image block information similarity, which can be used to evaluate and remove redundant images. The image is divided into several image blocks, and each image block is calculated with brightness, two-dimensional entropy and structure information. The information of all image blocks will be merged as the unique identification code of the image and stored in the json file. After calculating the similarity measure of the corresponding identification codes of the two frames, the redundant images are filtered out according to the preset conditions. 800 frames of small intestine endoscopic images collected from Changhai Hospital in 2018 were amplified, and then used for redundant image removal experiment. The experimental results show that the data screening speed of the algorithm is 0.03 seconds/frame, the recall rate of the experiment is 99.14%, the accuracy rate is 98.42%, and the F-measure is 98.01%. The algorithm effectively retains the original image information and improves the screening speed, and has good performance for complex and changeable gastrointestinal images.
作者
张林琪
郭旭东
张璐璐
刘张
韦彩诗
薛亦鑫
Zhang Linqi;Guo Xudong;Zhang Lulu;Liu Zhang;Wei Caishi;Xue Yixin(School of Medical Instrument and Food Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)
出处
《电子测量技术》
2020年第22期93-97,共5页
Electronic Measurement Technology
基金
上海市科学自然基金(20ZR1437700)项目资助。
关键词
胶囊内镜
图像块信息相似性评估
冗余去除
capsule endoscopy
block information similarity
redundancy removal
