【作者】 赵凯；

【导师】 赵宏； 康雁；

【作者基本信息】 东北大学 ， 计算机应用技术， 2012， 博士

【摘要】 针对骨骼医学图像的计算机后处理技术可以改善图像观察效果、实现智能辅助诊断,进而推动相关疾病影像学诊断技术的发展和完善。本文首先综述和分析相关技术发展现状,进而针对骨分割、骨测量、骨移除和骨拼接提出了新的方法或技术,以解决现有技术在精确性、鲁棒性或自动化程度等方面存在的缺陷。第一,在定量CT骨密度测量(尤其是针对腰椎的测量)中,需要借助骨分割和骨测量技术来准确提取测量目标,并进行精确测量。针对现有骨分割技术难以完整、自动化地分割单节腰椎的问题,本文提出了一种高度自动化的单节腰椎分割方法。首先分割整个脊椎,再分别利用不同的图像处理方法断开椎体和椎小关节间的连接。在真实QCT数据和仿真数据上实验,证明此方法在自动化程度和结果准确性方面均优于现有的各种腰椎分割方法。本文提出了一种自动定参的无体模骨密度测量技术,使骨密度测量突破了体模的限制。大数据量实验证实其具有较高的准确性,可重复性也超过同类方法。第二,在使用血管造影CT图像来诊断头颈部和体部血管疾病时,需要利用骨移除技术,以便无干扰地观察和提取血管。本文提出了基于骨轮廓信息融合的头颈部骨移除技术和基于轮廓提取与分类的体部骨移除技术。实验证明这两项技术均在各自应用中得到了较好的去骨效果,大大提高了图像观察质量,进而提高了血管提取准确性。第三,骨拼接技术主要应用于长骨外伤或畸形等疾病的影像学诊断中,将分段采集的图像拼接为便于观察和诊断的全景图像。为提高现有骨拼接技术的鲁棒性,并解决融合图像中存在的“拼缝”伪影和曝光度差异等问题,本文提出了一种旋转无关的图像拼接方法。此方法借助改进的特征配准和相关度匹配等算法实现图像配准。经实验验证,其配准算法在保证结果准确性的同时提高了鲁棒性；其融合算法可有效地平衡曝光度差异,并避免“拼缝”伪影,极大地提高了全景图像质量。另外,设计了三款用于辅助临床诊断的软件系统(或软件功能模块)。这些软件已被投入临床日常诊断中,或已开始临床试用,这也将本研究的理论成果带入了实际应用。更多还原

【Abstract】 The computer post-processing technology for bone medical images can improve the image observation effect, actualize the intelligent aided diagnosis, and thus promote the development of the image diagnosis technology. In this paper, the author first reviewed and analyzed the development status of relevant technologies, and then proposed new methods or technology for bone segmentation, measurement, removal and stitching to fetch up the detect of the current technology in accuracy, robustness or automation.First, in the quantitative analysis of CT bone mineral density (especially for the analysis of the lumbar spine), the bone segmentation and bone measurement technologies should be adopted to extract the accurate measurement target, and then perform accurate measurement.In order to solve the problem that a single section of the lumbar spine couldn’t be segmented completely and automatically, a highly automated single vertebral segmentation method was proposed. It is proved that this method has better automation degree and accuracy than the current methods by the experiments with actual QCT data and synthetic data,.This paper presents a without phantom automatic parameter setting method to measure the bone mineral density. Through large quantity of experimental data, this method is proved to have high accuracy and repeatability.Second, when the vascular diseases in head, neck or body are diagnosed by using angiography CT images, the bone removal technology should be used so that the doctor can observe and extract vessels without interference.This paper presents a head and neck bone removal technology on the basis of bone contour information infusion. And improves a body bone removal technology on the basis of contour extraction and classification also. It is proved that these two technologies both can get better bone removal effect, greatly enhance the image observation quality and improve the vascular extraction accuracy.Third, bone stitching technology is mainly used for the image diagnosis of long bone trauma or deformity, it takes better observation and diagnosis.In order to improve the robustness of the current bone stitching technology and solve the problem of stitching artifacts and exposure differences in the fusion image, this paper presents a rotation-independent image stitching method. The image registration is achieved in virtue of improved feature alignment and correlation matching algorithm. By the experiments, it is proved that the registration algorithm can ensure the accuracy of the result and can improve the robustness; the fusion algorithm can effectively balance the exposure difference and eliminate the stitching artifacts, greatly improving the quality of the panoramic images.In addition, three software systems (or software functional modules) used to assist in clinical diagnosis were designed. They are being widely used or being tested. The theoretical results of this paper have been brought into practical clinical use.更多还原