Co-registration of CT-SPECT images for tumor localization and dosimetry
Co-registration of CT-SPECT images for tumor localization and dosimetry
复制标题
用于肿瘤定位和剂量测定的 CT-SPECT 图像联合配准
DOI：
10.1109/cvpr.2013.226
复制
发表时间：
2006
期刊：
影响因子：
--
通讯作者：
P. Cohen
中科院分区：
文献类型：
--
作者：
A. Celler;L. Tang;G. Hamarneh;P. Cohen
目的SPECT和CT图像的配准可以将SPECT图像的功能信息与CT图像的解剖定位信息结合起来，从而大大提高肿瘤的诊断水平。此外，当基于CT的衰减图用于衰减和散射校正时，数据的定量分析可以通过为肿瘤和关键器官提供准确的剂量测定来帮助放射治疗计划。由于CT和SPECT通常在不同的日期进行，并且使用不同的成像系统，因此它们显然需要配准。然而，即使在组合的SPECT/CT相机上完成扫描，由于患者移动，已经报告了大量图像未对准，因此这些图像也可能需要共同配准。然而，共配准提出了一个重大的挑战，特别是在高目标-背景图像中，其中肿瘤很明显，但几乎没有其他身体元素，没有提供用于配准的标志。如果解剖学标记，如在CT中清晰可见的骨结构，可以用于配准，这种情况将得到很大改善。方法为了便于配准，我们的方法提出了骨放射性示踪剂与肿瘤剂同时注射。这种方法不仅有助于配准，而且还可以诊断潜在的转移。我们的自动配准方法使用互信息作为配准标准。为了提高算法的鲁棒性和准确性，使用多分辨率方法，其中以粗到细的方式执行配准。在配准期间，使用梯度下降最小化方案在由平移参数定义的参数空间上优化度量值。该算法进行了测试的临床数据对应的骨盆和胸部CT和SPECT骨研究。结果通过对2D和3D图像的视觉检查以及观察相似性度量和变换参数的收敛性来进行验证。此外，对成对体积的初始未对准的影响和配准的再现性进行了评价。在所有研究中，多分辨率配准收敛接近最终对准。X轴和Y轴的初始未对准高达50 mm，Z轴的初始未对准高达10 mm，导致66例骨盆研究中61例成功配准，66例胸部研究中54例成功配准。平移误差分别为0.712mm±0.313mm和1.312mm±0.623mm。整个配准过程的平均计算时间（在3.0GHz PC上）< 4分钟。结论提出了一种基于骨断层扫描的SPECT-CT图像自动配准方法。
Objectives Co-registration of SPECT and CT images may substantially improve tumor diagnosis by combining the functional information from scintigraphic images with the anatomical localization obtainable from CT images. Additionally, when CT-based attenuation maps are used in attenuation and scatter corrections, quantitative analysis of the data can help radiotherapy planning by providing accurate dosimetry for tumors and critical organs. Since CT and SPECT are usually performed on different days and using separate imaging systems, they clearly require co-registration. However, even if the scans are done on a combined SPECT/CT camera substantial image misalignments have been reported due to patient movement, therefore these images may also need to be co-registered. Co-registration however poses a major challenge, especially in high target-to-background images where tumors are well visible but almost none of the other body elements, providing no landmarks for registration. This situation would be much improved if anatomical markers, such as bone structures that are clearly visible in CT, could be used in co-registration. Methods In order to facilitate co-registration our method proposes an injection of bone radiotracer simultaneously with a tumor agent. This approach not only helps co-registration but may additionally diagnose potential metastases. Our automatic co-registration method uses mutual information as the registration criterion. To improve the robustness and accuracy of the algorithm, a multi-resolution approach is used in which registration is performed in a coarse-to-fine manner. During registration, the metric value is optimized over the parameter space defined by the translation parameters using a gradient-descent minimization scheme. The algorithm was tested on clinical data corresponding to pelvic and thoracic CT and SPECT-bone studies. Results Validation was done by visual inspection of 2D and 3D images and by observing the convergence in the similarity metric and the transformation parameters. Additionally, the effect of initial misalignment of the paired volumes and the reproducibility of the registration were evaluated. In all studies, registration with multi-resolution converged close to a final alignment. Initial misalignments of up to 50mm in the X and Y-axes, and 10mm in the Z-axis resulted in 61 of 66 successful registrations for the pelvic and 54 of 66 for the thoracic studies. The translation errors were 0.712mm±0.313mm and 1.312mm±0.623mm, respectively. The average computation time (on a 3.0GHz PC) was < 4 minutes for the entire registration procedure. Conclusion An automatic SPECT-CT co-registration method based on bone tomoscintigraphy has been developed and validated.