基于深度哈希卷积神经网络的医学图像检索  

Medical Image Retrieval Based on Deep Hash Convolution Neural Network

作　　者：崔少国[1] 熊舒羽 刘畅[1] 陈默语 CUI Shaoguo;XIONG Shuyu;LIU Chang;CHEN Moyu(School of Computer Science and Engineering,Chongqing University of Technology,Chongqing 400054,China)

机构地区：[1]重庆理工大学计算机科学与工程学院,重庆400054

出　　处：《重庆理工大学学报（自然科学）》2020年第8期134-142,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国家自然科学基金项目(61502065);重庆市教委人文社科研究重点项目(16SKGH133);重庆市科委基础研究与前沿探索一般项目(cstc2016jcyjA0383,cstc2018jcyjAX0324);重庆市科委重点项目(cstc2015jcyjBX0127)。

摘　　要：医学图像为临床诊断提供了重要的影像学信息,如何从海量图像库中快速准确检索到病人影像信息是一项重要研究课题。提出一种基于深度哈希卷积神经网络的医学图像检索方法,通过建立深度哈希编码神经网络,以成对柯西交叉熵损失函数为优化目标,采用迁移学习方法实现了医学图像的实时精确检索。以ImageCLEF(Med)数据集为实验对象,实验证实新提出的方法是可行、高效的,对于大规模医学图像,基于深度哈希卷积神经网络的医学图像检索方法,明显提高了图像检索性能。当采用48维哈希编码时,平均检索准确率为91.9%,平均检索时间为0.037 s,结果优于当前基于深度学习的图像检索方法。Medical imaging provides important information for clinical diagnosis.How to retrieve patient image information quickly and accurately from image database is an important research topic.A medical image retrieval method based on deep hash CNN is proposed.We establish a deep hash neural network and optimize the model by usingthe paired Cauchy cross entropy loss function.The real-time accurate retrieval model of medical images is achieved by the transfer learning method.Taking the Image CLEF(Med)as the experiment dataset,the results show that the proposed method is feasible and efficient.For large-scale medical images,the proposed retrieval method based on deep hash CNN significantly improves the image retrieval performance.When we use 48-dimensional hash coding,the average retrieval accuracy is 91.9%,and the average retrieval time is 0.037 seconds.The result is better than the current existing image retrieval method.

关 键 词：深度哈希 卷积神经网络 医学图像检索 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]