Reproducibility of findings in modern PET neuroimaging: insight from the NRM2018 grand challenge.
Reproducibility of findings in modern PET neuroimaging: insight from the NRM2018 grand challenge.
复制标题
DOI：
10.1177/0271678x211015101
复制
发表时间：
2021-10
期刊：
影响因子：
--
通讯作者：
Grand Challenge Participants#
中科院分区：
文献类型：
--
作者：
Veronese M;Rizzo G;Belzunce M;Schubert J;Searle G;Whittington A;Mansur A;Dunn J;Reader A;Gunn RN;Grand Challenge Participants#
研究结果的可重复性是神经影像学界目前面临的一个引人注目的方法学问题。即使在分析相同的数据时，图像处理、量化和统计的标准化流程的缺乏也在结果的可变性和解释中发挥着重要作用。这个问题在 MRI 研究中是众所周知的，该方法无可争议的价值因大量产生不一致结果的研究而变得复杂化。然而，任何具有复杂数据和灵活分析程序的研究领域都可能遇到类似的重复性缺乏的情况。在本文中，我们研究了脑 PET 成像的这个问题。在 2018 年神经受体图谱会议期间，大脑 PET 社区面临着一项涉及模拟神经递质释放实验的计算竞赛的挑战。十四个国际团队分析了相同的成像数据集，其基本事实是已知的。尽管有多种方法，但参与者的解决方案是一致的，尽管并不完全相同。这些结果应该让人们认识到，仅增加 PET 数据的共享只是增强神经影像结果信心的一个组成部分，并且重要的是用用于量化数据的分析流程和程序的完整细节来补充这一点。
The reproducibility of findings is a compelling methodological problem that the neuroimaging community is facing these days. The lack of standardized pipelines for image processing, quantification and statistics plays a major role in the variability and interpretation of results, even when the same data are analysed. This problem is well-known in MRI studies, where the indisputable value of the method has been complicated by a number of studies that produce discrepant results. However, any research domain with complex data and flexible analytical procedures can experience a similar lack of reproducibility. In this paper we investigate this issue for brain PET imaging. During the 2018 NeuroReceptor Mapping conference, the brain PET community was challenged with a computational contest involving a simulated neurotransmitter release experiment. Fourteen international teams analysed the same imaging dataset, for which the ground-truth was known. Despite a plurality of methods, the solutions were consistent across participants, although not identical. These results should create awareness that the increased sharing of PET data alone will only be one component of enhancing confidence in neuroimaging results and that it will be important to complement this with full details of the analysis pipelines and procedures that have been used to quantify data.