基于Flink的海量医学图像检索系统设计与实现

Design and implementation of massive medical image retrieval system based on Flink

摘要:

医学图像检索是有效利用医学资源的基础，而医学图像的海量性和增量性为图像检索带来了新的挑战和要求。为了提高医学图像检索过程的效率，设计并实现一种基于Flink的海量医学图像检索系统。首先，系统通过Web应用作为用户操作入口，在后端搭建数据平台和业务集群；其次，系统通过HBase对医学图像数据进行分布式存储，利用深度卷积神经网络模型提取医学图像特征；然后，将所提取的医学图像特征数据进行乘积量化编码，并通过HBase进行存储；最后，通过基于Flink的内存计算对接Kafka进行实时图像检索，以及对批量导入图像的特征索引编码。系统在4个节点的服务器上部署分布式集群，使用真实医学图像数据集进行测试，通过在MapReduce和Spark两种不同技术模块下的对比实验表明本系统具有更好的检索效率表现。

Abstract:

Medical image retrieval is the basis of effective utilization of medical resources, and the massive and incremental medical image brings new challenges and requirements for image retrieval. In order to improve the efficiency of medical image retrieval process, a Flink based medical image retrieval system is designed and implemented. Firstly, the system uses web application as the users" operation entry, and builds data platform and business cluster at the back end. Secondly, HBase is used to store the medical image data in a distributed way, and extracts the medical image features by using the deep convolution neural network model. Thirdly, the extracted medical image feature data is multiplied and encoded, and stored by HBase. Finally, Kafka is used for real-time image retrieval through memory computing and then consumed on Flink, as well as feature index coding for batch imported images. The system deployed a distributed cluster on four nodes of servers and tested with real medical image data set. The comparison experiment under MapReduce and Spark shows that the system has better retrieval efficiency.