Dartmouth College, Hanover, NH, 03755, USA.
Department of Psychological and Brain Sciences, Dartmouth College, Hanover, NH, 03755, USA.
F1000Res. 2020 Sep 14;9:1131. doi: 10.12688/f1000research.24544.1. eCollection 2020.
Magnetic resonance imaging (MRI) is an important yet complex data acquisition technology for studying the brain. MRI signals can be affected by many factors and many sources of variance are often simply attributed to "noise". Unexplained variance in MRI data hinders the statistical power of MRI studies and affects their reproducibility. We hypothesized that it would be possible to use phantom data as a proxy of scanner characteristics with a simplistic model of seasonal variation to explain some variance in human MRI data. We used MRI data from human participants collected in several studies, as well as phantom data collected weekly for scanner quality assurance (QA) purposes. From phantom data we identified the variables most likely to explain variance in acquired data and assessed their statistical significance by using them to model signal-to-noise ratio (SNR), a fundamental MRI QA metric. We then included phantom data SNR in the models of morphometric measures obtained from human anatomical MRI data from the same scanner. Phantom SNR and seasonal variation, after multiple comparisons correction, were statistically significant predictors of the volume of gray brain matter. However, a sweep over 16 other brain matter areas and types revealed no statistically significant predictors among phantom SNR or seasonal variables after multiple comparison correction. Seasonal variation and phantom SNR may be important factors to account for in MRI studies. Our results show weak support that seasonal variations are primarily caused by biological human factors instead of scanner performance variation. The phantom QA metric and scanning parameters are useful for more than just QA. Using QA metrics, scanning parameters, and seasonal variation data can help account for some variance in MRI studies, thus making them more powerful and reproducible.
磁共振成像（MRI）是研究大脑的一项重要但复杂的数据采集技术。MRI 信号会受到许多因素的影响，许多来源的方差通常简单地归因于“噪声”。MRI 数据中未解释的方差会降低 MRI 研究的统计效力，并影响其可重复性。我们假设，使用简单的季节性变化模型，将体模数据用作扫描仪特征的代理，可以解释人体 MRI 数据中的一些方差。我们使用了从多个人体参与者研究中收集的 MRI 数据，以及每周为扫描仪质量保证（QA）目的而收集的体模数据。我们从体模数据中确定了最有可能解释获得数据方差的变量，并通过使用这些变量来模拟信噪比（SNR）来评估其统计学意义，SNR 是基本的 MRI QA 指标。然后，我们将体模 SNR 纳入从同一扫描仪获得的人体解剖学 MRI 数据的形态测量指标模型中。在多次比较校正后，体模 SNR 和季节性变化是灰质脑体积的统计学显著预测因子。然而，在对 16 个其他脑区和脑区类型进行扫描后，在经过多次比较校正后，体模 SNR 或季节性变量均未发现统计学显著的预测因子。季节性变化和体模 SNR 可能是 MRI 研究中需要考虑的重要因素。我们的研究结果表明，季节性变化主要是由生物人类因素引起的，而不是扫描仪性能变化。体模 QA 指标和扫描参数不仅对 QA 有用。使用 QA 指标、扫描参数和季节性变化数据可以帮助解释 MRI 研究中的一些方差，从而提高它们的统计效力和可重复性。
