Nuklearmedinische Klinik der TU München, Ismaningerstrasse 22, 81675 München, Germany.
Med Phys. 2010 Jun;37(6):2414-24. doi: 10.1118/1.3395554.
The combination of sequentially acquired cardiac PET and SPECT data integrating metabolic and perfusion information allows the assessment of myocardial viability, a relevant clinical parameter for the management of patients who have suffered myocardial infarction and are now candidates for complex and cost intensive therapies such as bypass surgery. However, registration of cardiac functional datasets acquired on different imaging systems is limited by the difficulty to define anatomical landmarks and by the relatively poor inherent spatial resolution. In this article, the authors sought to evaluate whether it is possible to automatically register FDG-PET and sestamibi-SPECT cardiac data.
Automatic rigid registration was implemented with the ITK framework using Mattes mutual information as the similarity measure and a quaternion to represent the rotational component. The goodness of the alignment was evaluated by computing the mean target registration error (mTRE) at the myocardial wall. The registration parameters were optimized for robustness and speed using the data from 11 cardiac patients undergoing both PET and SPECT examinations (training datasets). The optimized algorithm was applied on the PET and SPECT data from 11 further patients (evaluation datasets). Quantitative (mTRE calculation) and visual (scoring method) comparisons were performed between automatic and manual registrations. Moreover, the automatic registration was also compared to the registration implicitly defined in the standard clinical analysis.
The registration parameters were successfully optimized and resulted in a mean mTRE of 1.13 mm and 1.2 s average runtime on standard computer hardware for the training datasets. Automatic registration in the 11 validation datasets resulted in an average mTRE of 2.3 mm, with 7.5 mm mTRE in the worst case and an average runtime of 1.6 s. Automatic registration outperformed manual registrations both for the mTRE and for the visual assessment. Automatic registration also resulted in higher accuracy and better visual assessment as compared to the registration implicitly performed in the standard clinical analysis.
The results demonstrate the possibility to successfully perform mutual information based registration of PET and SPECT cardiac data, allowing an improved workflow for the sequentially acquired cardiac datasets, in general, and specifically for the assessment of myocardial viability.
连续获取的心脏 PET 和 SPECT 数据结合代谢和灌注信息,可评估心肌活力,这是评估心肌梗死后接受复杂且昂贵治疗(如搭桥手术)患者的重要临床参数。然而,在不同成像系统上获取的心脏功能数据集的配准受到难以定义解剖学标志和相对较差固有空间分辨率的限制。在本文中,作者旨在评估是否可以自动配准 FDG-PET 和 sestamibi-SPECT 心脏数据。
使用 ITK 框架实现自动刚性配准,以 Mattes 互信息作为相似性度量,并使用四元数表示旋转分量。通过计算心肌壁的平均目标配准误差(mTRE)来评估配准的准确性。使用来自 11 名同时接受 PET 和 SPECT 检查的心脏患者的数据(训练数据集)来优化配准参数的稳健性和速度。将优化的算法应用于另外 11 名患者的 PET 和 SPECT 数据(评估数据集)。对自动和手动配准进行了定量(mTRE 计算)和视觉(评分方法)比较。此外,还将自动配准与标准临床分析中隐含的配准进行了比较。
成功优化了配准参数,对于训练数据集,平均 mTRE 为 1.13mm,平均运行时间为 1.2s。在 11 个验证数据集中的自动配准结果导致平均 mTRE 为 2.3mm,最差情况下为 7.5mm mTRE,平均运行时间为 1.6s。在 mTRE 和视觉评估方面,自动配准均优于手动配准。与标准临床分析中隐含的配准相比,自动配准的精度更高,视觉评估更好。
这些结果表明,成功地进行了基于互信息的 PET 和 SPECT 心脏数据配准,为顺序获取的心脏数据集的工作流程提供了改进,特别是为心肌活力的评估提供了改进。