Adaptive cyclic physiologic noise modeling and correction in functional MRI
Adaptive cyclic physiologic noise modeling and correction in functional MRI
复制标题
DOI：
10.1016/j.jneumeth.2010.01.013
复制
发表时间：
2010-03-30
影响因子：
3
通讯作者：
Beall, Erik B.
中科院分区：
文献类型：
--
作者：
Beall, Erik B.
众所周知，BOLD加权MRI数据中的生理噪声是方差的重要来源，降低了fMRI和功能连通性分析的统计能力和特异度。我们表明，在fMRI和fcMRI数据中，当前的噪声校正方法都有了显著的改进，避免了过度拟合。传统的噪声模型是一种傅立叶级数展开，叠加在平行测量的呼吸和心脏周期的周期性上。使用该模型的校正将导致去除与生理周期的周期相匹配的方差。使用此框架可以轻松地对噪声进行建模。然而，使用大量的回归变量是以去除与生理噪声无关的方差为代价的，例如，由于功能感兴趣的信号引起的方差(过度拟合数据)。我们的假设是，描述所有显著耦合的生理学噪声的拟合种类很少。如果这是真的，我们可以用较少数量的拟合回归变量来代替模型中使用的大量回归变量，从而以较小的感兴趣的方差减少来解释噪声源。我们描述了这些扩展，并证明了我们可以保留与生理噪声无关的数据的方差，同时等价地去除生理噪声，导致数据具有比现有校正技术更高的有效信噪比。我们的结果表明，功能磁共振成像在灵敏度和功能连接性分析方面有了显着改善(与高阶传统噪声校正相比，功能磁共振成像的激活体积增加了17%)。(C)2010爱思唯尔B.V.保留所有权利。
Physiologic noise in BOLD-weighted MRI data is known to be a significant source of the variance, reducing the statistical power and specificity in fMRI and functional connectivity analyses. We show a dramatic improvement on current noise correction methods in both fMRI and fcMRI data that avoids overfitting. The traditional noise model is a Fourier series expansion superimposed on the periodicity of parallel measured breathing and cardiac cycles. Correction using this model results in removal of variance matching the periodicity of the physiologic cycles. Using this framework allows easy modeling of noise. However, using a large number of regressors comes at the cost of removing variance unrelated to physiologic noise, such as variance due to the signal of functional interest (overfitting the data). It is our hypothesis that there are a small variety of fits that describe all of the significantly coupled physiologic noise. If this is true, we can replace a large number of regressors used in the model with a smaller number of the fitted regressors and thereby account for the noise sources with a smaller reduction in variance of interest. We describe these extensions and demonstrate that we can preserve variance in the data unrelated to physiologic noise while removing physiologic noise equivalently, resulting in data with a higher effective SNR than with current corrections techniques. Our results demonstrate a significant improvement in the sensitivity of fMRI (up to a 17% increase in activation volume for fMRI compared with higher order traditional noise correction) and functional connectivity analyses. (C) 2010 Elsevier B.V. All rights reserved.