【作者】 沃飞；

【导师】 庄天戈；

【作者基本信息】 上海交通大学 ， 生物医学工程， 2007， 硕士

【摘要】 X线胸片是临床上使用最多的医学成像诊断方法之一,放射科每天产生大量胸片图像,一旦发生归档错误,将产生严重的后果。本文研究了基于DICOM胸片图像的患者身份识别技术,它能够有效地辅助减少这类误归档错误。研究内容主要包括胸片图像的预处理、分割和身份识别。在胸片的预处理和分割方面,首先,本文通过回顾和研究了以往的经典算法,综合了投影法、聚类法、阈值分割、梯度法、模板匹配、边缘追踪等多种方法实现了胸片ROI的提取、肺野的分割、肋骨锁骨边缘的提取等,取得了良好的效果。其次,本文对图像分割的新方法,如像素点分类法、活动模型法,进行了研究,并标定了12幅胸片的肺野使用AAM模型对另外10幅图像进行分割实验,其中9幅取得了很好的分割效果,只有一幅标定点略有偏移。胸片图像的身份识别技术研究是个新颖的课题,没有过多的经验可以参考。本文参考医生临床上实际辨别方法并借鉴指纹识别和人脸识别的方法,对胸片识别进行了尝试性研究。文章从两个角度展开研究:第一种方法,我们通过胸片图像分割获得了肺野、肋骨和锁骨的信息,从这些信息中提取面积、边缘形态等特征,构建数据库,再从中选择部分重要特征进行识别;第二种方法模仿医生肉眼分辨胸片身份的方法,对比两幅图像中的胸骨框架外形、心脏形状、动脉形状、锁骨和肋骨的弯曲程度长度等等,对两幅图像上的对应部位计算相关系数,并以此为特征,根据一定的规则判别图像是否来自同一个患者,实现了两幅图像之间的身份识别。更多还原

【Abstract】 Chest radiography is one of the most popular medical imaging diagnosis methods in clinical practice. Large amount of chest radiographs are produced in X-ray department every day. Mistakes in archiving these images will lead to serious consequence. In this paper identity recognition based on DICOM chest radiography is researched, which is helpful to reduce mis-archiving. For this purpose, preprocess, segmentation and identification of chest radiographies is discussed.Preprocess and segmentation are discussed in this paper. Firstly, many classical algorithms are reviewed. Combining With these algorithms, such as profiling, clustering, threshold segmentation, gradient edge detection, model matching, edge tracing, A new method is put forward to extract ROI of anatomical structure, segmentation of lung field, edges of ribs and clavicle. This integrated method produces wonderful results. Furthermore, new methods, such as pixel classification and active models, are adopted in this paper. Lung fields in 12 chest CR images are marked to train an AAM model and other 10 images are inputted for testing. We get a good segmentation result that only one mismatchs is identified.Identity recognition of chest radiography is a very new topic and little research experience can be found. In this paper, it is attempted to study patients’identity recognition through imitating how doctors do in clinic and how researchers do in fingerprint recognition and face recognition area. This problem更多还原