【作者】 蒋历军；

【导师】 庄天戈；

【作者基本信息】 上海交通大学 ， 生物医学工程， 2008， 博士

【摘要】 随着医院数字化的发展和PACS的普及,医学影像数据急剧膨胀。人们对基于内容的图像检索(Content Based Image Retrieval, CBIR)技术开展了大量研究以解决日益增长的医学影像数据的检索问题。然而,时至今日CBIR技术尚未在临床上广泛应用,这与CBIR技术尚有许多有待完善之处有很大关系。本文提出了应用网格和水印的肝脏CT图像基于内容检索(Content Based Hepatic-CT Image Retrieval using Grid and Watermarking, CBHIRuGW)系统,旨在提升CBIR在查准率、信息安全、检索效率和数据管理等方面的性能,从而提高医学图像信息的利用率,服务于临床、研究和教学。CBHIRuGW系统由存储子系统和检索子系统两部分组成;每个子系统分别由三大模块组成:即图像特征提取/CBIR检索模块、数字水印编码/解码模块、数据网格/计算网格模块。本文围绕这三组模块进行了相关研究:在CBIR模块组,本文提出了基于查询图像分类的图像检索方案以提高系统检索准确率;在水印模块组,利用无损水印技术,提升了系统信息安全和检索效率;在网格模块组,利用网格平台,进一步提升了系统检索效率,并简化了医学数据管理。关于这三组模块的详细叙述,即本文的主要研究内容如下:第一,本文对提升检索系统的查准率进行了研究。在CBIR模块组,本文提出基于查询图像分类的图像检索方案以提高检索准确率。在根据查询图像检索相似图像时,采用某种图像分类方法,将查询图像和数据库中的图像组合在一起进行分类,并将数据库中与查询图像属于同一个类别的图像组成简化数据库;然后,采用CBIR检索方法,从该简化数据库(而不是原始数据库)检索图像。实验表明,通过增加分类处理这一环节,可以在一定程度上提高系统的检索准确率。本文首先采用了支持向量机分类方法用于肝脏CT图像检索,实验表明检索准确率有所提升。随后,根据肝脏CT图像的特点,本文提出了三级阈值分类方法,并将其用于肝脏CT图像检索,实验表明检索准确率获得了进一步提升。第二,本文对检索系统的信息安全和检索效率进行了研究。在水印模块组,将数字水印技术应用于检索系统,以解决图像完整性验证、敏感信息加密等安全性问题以及提升图像相关信息检索效率问题。利用数字水印技术,将各种相关信息(包括患者隐秘信息、肝脏CT图像的Hash值、肝脏和病灶的形态特征,以及病灶区域的位置和纹理信息)以及其它诊断信息等作为水印负载嵌入图像,可以为检索系统带来一系列的好处:首先,水印特殊的编码以及密码保护方式可以为图像信息提供完整性验证以及设置多级读取权限,从而在一定程度上确保了医学数据及其相关信息的安全性,扩展了医学图像的应用范围;其次,由于图像的各种相关信息作为水印嵌入图像,在检索的过程中,一方面可以省去相关信息检索的操作,节约手工操作时间以及额外的系统响应时间,另一方面可以降低网络传输的负载,减少网络传输时间,从而提升图像相关信息的检索效率,这在网络拥塞情况下效果尤其明显;再次,ROI指示性水印(可见水印)可以为检索用户提供一个读片参考;最后,由于图像的各种相关信息作为水印嵌入图像,不再作为单独文件另外存储,在一定程度上简化了图像及其相关信息的管理。针对当前一些水印技术的不足之处,本文分别提出了基于分块排序和基于阈值控制的可扩展差值添加水印方法,从而提高了不同负载情况下的水印性能,降低了含水印图像的失真度。第三,本文对检索系统的检索效率和图像管理进行了研究。在网格模块组,将网格计算技术应用于检索系统,以改善图像数据管理工作和提升系统检索效率。通过整合实验室闲置计算机,构建实验室网格平台对此方案进行了模拟实验:在数据管理方面,通过数据网格的功能模块,系统实现了对PACS系统进行无缝备份和恢复功能;在计算能力方面,利用网格提供的虚拟计算平台,系统对检索过程的最后一个环节(图像刚性配准)进行并行计算,从而大大提升了系统的检索效率。更多还原

【Abstract】 With the development of hospital digitalization and popularity of PACS, medical data expand rapidly. CBIR (Content Based Image Retrieval) was focused for a few years in order to resolve the retrieval issues of medical images. However, CBIR is not applied widely in hospital, for it is still not perfect enough for clinical application.A content-based hepatic CT image retrieval system using grid computing and watermarking technology, CBHIRuGW, is proposed for the retrieval of hepatic CT images, which aims to improve the performance of retrieval precision, data management, retrieval efficiency and patient’s data protection. This system increases the utilization of medical image information in diagnostics, teaching and research. CBHIRuGW is composed of storage and retrieval subsystem, each is composed of three groups separately: image features extracting/CBIR group, digital watermarking encoding/decoding group, data/computing grid group. Within CBIR group, a query image classification based image retrieval method is proposed to improve the retrieval precision; within watermarking group, system data security and retrieval efficiency are advanced with using lossless digital watermarking technology; within grid group, The performance of data management and retrieval efficiency are upgraded with usinggrid computing technology. The details of these groups are discussed as follows: Firstly, a query image classification based image retrieval method is proposed. When retrieving images based on sample, images in the database are classified. Those belonging to the same class as the query sample are grouped as the simplified database and image retrieval is limited within the simplified database. At first, the support vector machine, SVM, was used for classification and the precision of retrival with SVM was better than that without the SVM. Then three level threshold classification method, 3-L was proposed for the hepatic CT images and experiments showed that the precision of retrival with 3-L was better than that with SVM.Secondly, Information security and retrieval efficiency were dissussed to protect the patient’s private information, promote the retrieval efficiency of an image’s revelant information and simplify the data management. Using digital watermarking, all kinds of related information (such as hash value, morphological characteristics, location of the regional focus, texture information and patient’s information) is embedded in the image as watermarking, which has many advantages: Firstly, the special coding and password protection of watermarking enable image integrity verification and multi-levels read permissions, which ensures the security of medical data and expands the application of medical images. Secondly, since all relevant information is embedded within medical image, the operation of retrieving relevant information is not required during the retrieval processing, thus the time for manual operation and additional system costs is saved; morover, the network load is reduced and thus the network transmission time is reduced. As a result of these factors, the retrieval efficiency is improved. Finally, because all relevant information is no longer stored as a separate file but embedded into the medical image, the management of the relevant information is simplified. In order to improve the watermark performance and reduce distortion under low and high load conditions, the block sorting based and differences expansion threshold controlled embedding method are proposed.Thirdly, retrieval efficiency and data management were discussed in the paper. The grid computing technology was applied to CBHIRuGW to improve the performance of data management and retrieval efficiency. The platform was simulated by integrating the idle computers in the laboratory. With the platform, the performance of the system’s data management and computing is improved. In data management, the system can perform seamless storage and backup management for the PACS system with help of the functional module of data grid. In calculating abilities, use of virtual grid computing platforms, the system can process the last step of the image retrieval - the rigid registration of images in parallel, thus greatly enhancing the efficiency of the system crawl.更多还原