Dias Maria de Fátima Machado, Carvalho Paulo, Castelo-Branco Miguel, Valente Duarte João
CIBIT, ICNAS, University of Coimbra, Portugal.
CISUC, Faculty of Sciences and Technology, University of Coimbra, Portugal.
Neuroimage Rep. 2022 Oct 7;2(4):100137. doi: 10.1016/j.ynirp.2022.100137. eCollection 2022 Dec.
A reproducibility crisis has been reported across many research fields, including neuroimaging, reaching up to 70% of studies. Neuroimaging data, such as magnetic resonance imaging (MRI), requires pre-processing to allow for inter-subject comparison, increase signal contrast and noise reduction. As manual MRI pre-processing is time consuming and requires expertise, multiple automatic pre-processing frameworks have been proposed. However, neuroimaging studies often report divergent results, even for similar populations, thus it is important to determine whether this occurs as a result of different processing tools. Two of the most used tools are FreeSurfer and the Computational Anatomy Toolbox (CAT12). In this study we assessed the reproducibility between these two automatic pre-processing frameworks for structural MRI and test-retest reliability within framework on estimation of cortical thickness. Our results show that the reproducibility between the frameworks is lower at the region-of-interest (ROI) level than at individual level. Furthermore, we found that the reproducibility was lower in paediatric samples than in adults. Finally, an acquisition site effect was also identified. Given the widespread use of these frameworks in basic and clinical neuroscience, the results of multicentric cross-sectional studies must be interpreted with caution, particularly with paediatric samples. The observed reproducibility issue might be one of the sources of discrepancies reported in neuroimaging studies. On a positive note, framework test-retest reliability within subject is high, suggesting that inconsistency of results may be less concerning in longitudinal studies. The code is available at: https://cibit-uc.github.io/fs-cat12-cortical-thickness-reproducibility.
据报道，包括神经影像学在内的许多研究领域都存在可重复性危机，高达70%的研究存在该问题。神经影像学数据，如磁共振成像（MRI），需要进行预处理，以便进行受试者间比较、增加信号对比度并降低噪声。由于手动MRI预处理耗时且需要专业知识，因此已经提出了多个自动预处理框架。然而，神经影像学研究经常报告出不同的结果，即使是针对相似人群，因此确定这种情况是否是由不同的处理工具导致的很重要。两种最常用的工具是FreeSurfer和计算解剖工具箱（CAT12）。在本研究中，我们评估了这两种用于结构MRI的自动预处理框架之间的可重复性，以及在框架内对皮质厚度估计的重测可靠性。我们的结果表明，在感兴趣区域（ROI）水平上，框架之间的可重复性低于个体水平。此外，我们发现儿科样本中的可重复性低于成人。最后，还发现了采集部位的影响。鉴于这些框架在基础和临床神经科学中的广泛应用，多中心横断面研究的结果必须谨慎解释，尤其是对于儿科样本。观察到的可重复性问题可能是神经影像学研究中报告的差异来源之一。值得庆幸的是，框架内受试者的重测可靠性很高，这表明在纵向研究中结果的不一致性可能不太令人担忧。代码可在以下网址获取：https://cibit-uc.github.io/fs-cat12-cortical-thickness-reproducibility 。
