基于动态优化模型的CT金属伪影校正方法研究

Currently, with the growing applications of CT (computed tomography) scan in clinical diagnosis, the metallic artifact problems in CT imaging have attracted more and more concerns, and many clinical practices show that these artifacts severely obscure anatomical structures near the metal objects, making it difficult for radiologists to correctly interpret the images. Therefore, the research on artifacts identification, correction, and assessment is always the hot problems in CT applications. However, the existing methods have the following shortcomings: severely residual artifacts, loss of soft details in background reconstructions, etc. In response to the problems above, our project aims to address residual artifacts suppression and tissue textures preservation via a dynamic regularization model. This study is then to be performed in the following two directions: 1), to improve the quality of prior set, we will focus on the proposed dynamic regularization model for prior restoration, which will lead to better performance in residual artifacts suppression; 2), we can also investigate the dynamic regularization model to exploit the sparseness of reconstruction images, which will lead to superior performance in soft tissue preservation. Through this project, it not only can achieve theoretical achievements in CT imaging, but also can provide credible reference to assist radiologists to accurately describe the tumors and organs surrounding the metallic implants.

当前，随着CT断层扫描在医学诊断中的应用越来越广泛，CT成像中的金属伪影问题已经引起了人们越来越多的关注。大量的临床实践表明金属伪影严重降低了金属植入物周围组织的清晰度，从而影响临床医生对异常组织的诊断分析，所以对它们的识别、校正以及评价一直以来都是医用CT研究领域的热点。但是，现有的校正方法存在以下问题：校正图像中次级伪影严重，金属周围的软组织边缘细节模糊等。本项目旨在通过动态优化模型，探索金属伪影校正中次级伪影抑制和软组织纹理细节保护的方法。研究主要围绕以下两条思路进行：1）研究将动态优化模型应用于先验图像去噪，以提高先验图像质量，进而抑制校正图像中次级伪影形成；2）研究基于动态优化模型增强待重建图像的稀疏性表征，以保护更多的软组织边缘细节。通过本项目研究，不仅可在CT成像领域取得理论成果，还将为放射科医生准确描述金属周围的肿瘤轮廓、组织器官等提供可信的参考。

CT成像中，植入病人体内的金属物体会导致重建图像中产生金属伪影，直接影响临床医生对断层结构的诊断分析以及放疗规划的实施。基于压缩感知理论和动态优化设计，本项目采用动态优化模型增强重建图像的纹理表征，以保护更多的软组织边缘细节。1）针对现有的正则化算法不能同时保证有效校正金属伪影和保证图像清晰度的问题，提出一种基于小波动态阈值的CT图像金属伪影校正方法。实验结果表明，我们提出的算法能有效校正金属伪影并能很好抑制次级伪影的形成，与经典算法相比较，在次级伪影抑制和软组织保护均取得最好效果。2）为了使重建图像能够保留更多的软组织边缘细节信息，提出一种动态变权全变分技术联合统计迭代重建方法抑制CT金属伪影。算法在初始迭代阶段采用较强的正则化因子以更好地抑制图像中的噪声，随着迭代进行，有序降低正则化因子参数以更好地保护软组织细节信息。通过体模数据、及临床数据的实验结果均表明该算法重建结果有效抑制了噪声和伪影，清晰重建出软组织细节，极大提高了重建图像的质量。此外，基于原始CT数据的迭代成像过程中，除了建立CT的统计噪声模型外、还精确建立CT系统的精确光学模型，基于X射线精确物理投射模型，研发了基于动态优化理论的统计迭代成像方法。基于CUDA架构，研究了迭代过程中关键参数的选定及开发高执行效率的并行算法