Bio Imaging & Signal Processing Lab., Dept. of Bio & Brain Engineering, Korea Advanced Institute of Science and Technology (KAIST), 373-1 Guseong-dong, Yuseong-gu, Daejon 305-701, Republic of Korea; Rotman Research Institute at Baycrest Centre, University of Toronto, Toronto, Ontario M6A 2E1, Canada.
Neuroimage. 2014 Jan 15;85 Pt 1:72-91. doi: 10.1016/j.neuroimage.2013.06.016. Epub 2013 Jun 15.
Functional near-infrared spectroscopy (fNIRS) is a non-invasive method to measure brain activities using the changes of optical absorption in the brain through the intact skull. fNIRS has many advantages over other neuroimaging modalities such as positron emission tomography (PET), functional magnetic resonance imaging (fMRI), or magnetoencephalography (MEG), since it can directly measure blood oxygenation level changes related to neural activation with high temporal resolution. However, fNIRS signals are highly corrupted by measurement noises and physiology-based systemic interference. Careful statistical analyses are therefore required to extract neuronal activity-related signals from fNIRS data. In this paper, we provide an extensive review of historical developments of statistical analyses of fNIRS signal, which include motion artifact correction, short source-detector separation correction, principal component analysis (PCA)/independent component analysis (ICA), false discovery rate (FDR), serially-correlated errors, as well as inference techniques such as the standard t-test, F-test, analysis of variance (ANOVA), and statistical parameter mapping (SPM) framework. In addition, to provide a unified view of various existing inference techniques, we explain a linear mixed effect model with restricted maximum likelihood (ReML) variance estimation, and show that most of the existing inference methods for fNIRS analysis can be derived as special cases. Some of the open issues in statistical analysis are also described.
功能近红外光谱（fNIRS）是一种通过无损颅骨测量大脑活动的非侵入性方法，通过大脑中光学吸收的变化来实现。与正电子发射断层扫描（PET）、功能磁共振成像（fMRI）或脑磁图（MEG）等其他神经影像学模式相比，fNIRS 具有许多优势，因为它可以直接测量与神经激活相关的血氧水平变化，具有高时间分辨率。然而，fNIRS 信号受到测量噪声和基于生理的系统干扰的严重影响。因此，需要仔细的统计分析从 fNIRS 数据中提取与神经元活动相关的信号。在本文中，我们对 fNIRS 信号的统计分析的历史发展进行了广泛的回顾，其中包括运动伪影校正、短源-探测器分离校正、主成分分析（PCA）/独立成分分析（ICA）、错误发现率（FDR）、序列相关误差，以及标准 t 检验、F 检验、方差分析（ANOVA）和统计参数映射（SPM）框架等推断技术。此外，为了提供各种现有推断技术的统一视图，我们解释了具有限制最大似然（ReML）方差估计的线性混合效应模型，并表明 fNIRS 分析的大多数现有推断方法都可以作为特例得出。还描述了统计分析中的一些开放性问题。
