【作者】 陈琴；

【导师】 邰晓英；

【作者基本信息】 宁波大学 ， 计算机应用技术， 2009， 硕士

【摘要】 近年来,随着医学成像技术的迅猛发展和应用,医学图像在临床诊断、治疗以及研究方面起着越来越重要的作用。基于内容的医学图像检索技术是从海量的医学图像中检索出具有相似病理特性图像的技术。由于医学图像具有灰度及空间分辨率高,信息量大,图片比较相似的特点,因此基于内容的医学图像检索往往比一般的图像检索要复杂。本研究针对胸片图像、胸部CT扫描图像和胃窥镜图像的融合进高层语义特征的检索技术进行了较深入的探讨,包括系统的体系结构、医学图像底层特征的提取、高层语义特征的提取、综合检索、相关反馈等。其中医学图像底层特征的提取主要基于颜色、纹理、形状、空间关系四大类,高层语义特征的提取主要基于医生诊断关键词,基于底层特征的潜在语义索引技术以及基于感兴趣区域的目标语义提取技术。对于胸片图像,医生的诊断信息比较全面,因此提出一种针对胸片图像的检索方法。该方法提取医生诊断信息的关键字,以关键字的出现频率作为语义特征向量,再结合底层特征以形成检索特征向量进行检索。对于彩色胃窥镜图像,根据其本身丰富的色彩信息,提出一种针对胃窥镜图像的检索方法。该方法对多种典型的颜色及空间关系提取算法进行了比较分析,选出最适合做进一步语义提取的颜色直方图法和颜色自相关图法,在此基础上利用潜在语义索引技术,提取潜在语义,去除噪声,以期达到提高检索效率的目的。对于胸部CT图像,医生的观察焦点主要在病灶区域,因此提出一种针对胸部CT图像的检索方法。该方法先提取感兴趣目标语义区域,然后在此感兴趣区域的基础上提取视觉特征,最后利用这些目标语义区域上的底层特征来检索图像。在以上思想的基础上,本研究设计了一个原型系统。对于本研究建立的医学图像库,利用不同的特征向量进行检索,并对各种算法的实验结果进行了比较分析。更多还原

【Abstract】 In recent years, medical image is getting more and more important effect in the clinical diagnosis, curing and the aspect of researching with the fact that medical imaging technology swiftness developing and applying. The content-based medical image retrieval is a technology to retrieve the image with similar pathological mechanism property from great quantity of medical images. Since the medical image has the characteristics of high gradation and the space resolution ratio, big information amounts and great similarity, the content-based medical image retrieval sometimes is more complicated than the ordinary image retrieval.Our work has a discussion as to sternum image, gastroscope image and chest CT image’s retrieval technology which fused with high-level semantic features, including system structure, low-level feature extraction, high-level semantic feature extraction, integrated retrieval and relevance feedback. Low-level feature extraction is based on color, texture, shape and spatial information of medical image. High-level semantic feature extraction is based on doctor’s diagnosis keywords, latent semantic indexing technology and object semantic extraction technology based on regions of interest.As to sternum image, the doctor’s diagnosis information is relatively comprehensive. A method for sternum image retrieval is proposed. The method extract the key words of doctor’s diagnosis information, and use the occurrence frequency of key words as semantic feature vector, and then combine low-level features to construct retrieval feature vector to retrieval. As to colorful gastroscope image, according to its own colorful information, a method for gastroscope image retrieval is proposed. We analyze several typical color and spatial information extraction algorithm, and then select the most adapt to semantic extraction algorithm, make use of latent semantic indexing technology to remove noise in order to improve the retrieval performance. As to chest CT image, doctor’s focus is on the focus area of the image. A method for chest CT image retrieval is proposed. We extract the region of interest, and then extract visual feature upon the region of interest, at last use the feature which based on the target semantic region to retrieve image.Based on above algorithms, our research designed a prototype system. As to the image database established by the research, we use different feature vectors to retrieve and several experiments are made to compare the above algorithms.更多还原