Department of Imaging and Pathology, Nuclear Medicine and Molecular imaging, Medical Imaging Research Center (MIRC), KU Leuven - University of Leuven, 3000, Leuven, Belgium.
Department of Cardiovascular Sciences, Medical Imaging Research Center (MIRC), KU Leuven - University of Leuven, 3000, Leuven, Belgium.
J Nucl Cardiol. 2019 Dec;26(6):2034-2044. doi: 10.1007/s12350-018-01581-z. Epub 2019 Jan 14.
In a previous study on ex vivo, static cardiac datasets, we investigated the benefits of performing partial volume correction (PVC) in cardiac F-Fluorodeoxyglucose(FDG) PET datasets. In the present study, we extend the analysis to in vivo cardiac datasets, with the aim of defining which reconstruction technique maximizes quantitative accuracy and, ultimately, makes PET a better diagnostic tool for cardiac pathologies.
In vivo sheep datasets were acquired and reconstructed with/without motion correction and using several reconstruction algorithms (with/without resolution modeling, with/without non-anatomical priors). Corresponding ex vivo scans of the excised sheep hearts were performed on a small-animal PET scanner (Siemens Focus 220, microPET) to provide high-resolution reference data unaffected by respiratory and cardiac motion. A comparison between the in vivo cardiac reconstructions and the corresponding ex vivo ground truth was performed.
The use of an edge-preserving prior (Total Variation (TV) prior in this work) in combination with motion correction reduces the bias in absolute quantification when compared to the standard clinical reconstructions (- 0.83 vs - 3.74 SUV units), when the end-systolic gate is considered. At end-diastole, motion correction improves absolute quantification but the PVC with priors does not improve the similarity to the ground truth more than a regular iterative reconstruction with motion correction and without priors. Relative quantification was not influenced much by the chosen reconstruction algorithm.
The relative ranking of the algorithms suggests superiority of the PVC reconstructions with dual gating in terms of overall absolute quantification and noise properties. A well-tuned edge-preserving prior, such as TV, enhances the noise properties of the resulting images of the heart. The end-systolic gate yields the most accurate quantification of cardiac datasets.
在之前对离体、静态心脏数据集的研究中，我们研究了在心脏 F-F 氟脱氧葡萄糖（FDG）PET 数据集中进行部分容积校正（PVC）的益处。在本研究中，我们将分析扩展到体内心脏数据集，旨在确定哪种重建技术可以最大限度地提高定量准确性，最终使 PET 成为心脏病理学的更好诊断工具。
使用小动物 PET 扫描仪（西门子 Focus 220，microPET）对离体羊心进行了体内绵羊数据集的采集和重建，采用了几种重建算法（有/无分辨率建模，有/无非解剖先验）。与体内心脏重建相比，进行了相应的离体心脏扫描。
与标准临床重建相比，使用边缘保持先验（在这项工作中使用总变差（TV）先验）与运动校正相结合，可以在绝对定量方面减少偏倚（在收缩末期门考虑时，为-0.83 与-3.74 SUV 单位）。在舒张末期，运动校正可以改善绝对定量，但与具有运动校正和无先验的常规迭代重建相比，具有先验的 PVC 并不能改善与地面真相的相似性。相对定量受所选重建算法的影响不大。
算法的相对排名表明，在整体绝对定量和噪声特性方面，具有双门控的 PVC 重建具有优势。经过良好调整的边缘保持先验，如 TV，可以增强心脏图像的噪声特性。收缩末期门可实现心脏数据集的最准确定量。
