【作者】 许全盛；

【导师】 叶大田；

【作者基本信息】 清华大学 ， 生物医学工程， 2011， 博士

【摘要】 PET/CT是肿瘤诊断及制定放疗计划的重要依据。然而PET扫描时间很长，患者的呼吸会导致胸腹部PET图像出现运动伪影，造成图像分辨率和对比度下降，肿瘤边界模糊，引起对肿瘤目标体积过高估计和放射性浓度过低估计，从而严重影响肿瘤诊断的准确率和放疗计划的精确性。目前关于PET/CT呼吸运动伪影校正的研究大都是针对呼吸门控采集的图像配准和图像重建运动校正方法，这些方法的实现依赖于对成像系统硬件性能的精确了解，同时复杂的重建和配准算法需要占用大量的计算资源，还可能因为大剂量的CT照射对患者造成额外伤害。针对目前呼吸门控设备价格昂贵，以及相应的配准和重建运动校正方案仍不能在临床中得到广泛应用的现状，本文提出了一种基于图像恢复的常规PET/CT图像呼吸运动伪影校正方案，该方案不依赖于任何硬件设备，肿瘤的呼吸运动参数估计和运动伪影校正都在重建后的图像域实现。在运动估计方面，本文通过伪影图像的方向微分最大最小灰度分析估计运动方向，通过平均自相关分析估计运动幅度，并采用加权平均和局部插值来降低运动估计的随机误差；在运动校正方面，本文提出了基于多孔小波残差去噪的改进的Richardson-Lucy迭代算法，根据估计的运动参数对运动伪影进行反卷积校正。该方法能显著降低迭代时的噪声放大效应，克服环状伪影。此外，本文还提出了图像的边缘保持滤波预处理方法，该方法有助于降低反卷积时噪声的影响。本文通过仿真数据、运动体模数据和临床数据对上述方法进行了验证和评价。本文对多例肺部肿瘤临床PET图像的呼吸伪影校正结果进行了统计研究。通过多元回归的方法证明了呼吸运动伪影与呼吸运动幅度、肿瘤大小之间的内在联系，解释了肿瘤部位、患者身高和性别等因素对呼吸幅度的影响，并检验了回归方程用于估计和评价呼吸运动伪影的能力。针对传统图像融合在提高视觉效果和保留原有信息之间的矛盾，本文提出了基于多孔小波变换的PET/CT融合方法，该方法能在不损失PET图像肿瘤量化信息的前提下，大大增加图像的解剖学信息，增强肿瘤目标边缘，呼吸伪影校正后的融合图像不仅有利于肿瘤诊断，还能直接用于肿瘤靶区勾画和放射性定量计算。更多还原

【Abstract】 PET/CT imaging plays a key role in tumor diagnosis and radiationtreatment planning. However, respiratory motion results in severe motionartifact in thoracic and abdomen PET images due to the long scanning time.Respiratory motion artifact significantly degrades PET images by reducingspatial resolution, tumor-to-background contrast；it also affects quantitationaccuracy, reduces the measured activity level of standard uptake value，causesthe tumor volume to be overestimated and thus increases the planned targetvolume. Existing approaches to reduce the blurring artifact involve acquiringimage in gated mode and using complicated registration-based orreconstruction-based motion correction algorithms. These methods requiremulti-frame acquisitions, detailed understanding of the geometry of scanner andthe response characteristics of each detector, higher CT dose and specializedreconstruction algorithms, thus is expensive in terms of time consuming,memory and health care.In view of the fact that respiratory gating device and its matched motioncorrection algorithms is too expensive to be widely used in domestic clinicalpractice, we proposed a post-reconstruction motion artifact correctionframework for un-gated PET/CT imaging using image restoration techniques.The advantage of proposed method is that it is independent of any particularimaging device and is easy to implement with less computing burden. We justuse the reconstructed image to estimate the original non blurred image throughmotion estimation and motion correction. In motion estimation, we proposed aminmax intensity directional derivative analysis and average auto-correlationanalysis to estimate or identify two PSF parameters: motion direction andmotion extent, respectively, weighted average and local interpolation were alsoused to reduce the stochastic error and improve estimation precision; in motioncorrection, we proposed a modified Richardson-Lucy deconvolution algorithmin conjunction with à trous wavelet residual de-noising to restore the motion blurred image according to estimated motion parameters. The wavelet denoisingmodification is aimed to suppress the deconvolution induced noise amplificationeffect and ringing artifact, moreover, an edge-preservation filteringpreprocessing was also employed to further improve the quality of deconvolutedimage. Simulation and mobile phantom data was used to test and evaluate theproposed methods before applied to clinical lung tumor PET images.We proposed a statistical study based on respiratory motion artifactcorrection results of39cases lung tumor PET data. We statistically showed therelationship between respiratory motion artifact and influences such asbreathing magnitude, tumor size, tumor location, patient height and gender etcthrough multi-regression analysis. We also tested the ability of resultedregression equations to estimate or predict the extent of motion artifact.To further improve the quality of motion corrected PET image, weproposed a new PET/CT image fusion strategy based on à trous wavelettransform. Combined with respiratory motion artifact correction, the method isable to produce a PET/CT fused image with a significantly enhancededge-sharpness and anatomical structure meanwhile preserve the quantitativeinformation of tumor target. Resulted image is therefore able to improve thelung tumor diagnosis but also can be used in tumor target delineation andactivity quantitative analysis.更多还原