Assessment of fNIRS Signal Processing Pipelines: Towards Clinical Applications
Assessment of fNIRS Signal Processing Pipelines: Towards Clinical Applications
复制标题
DOI：
10.3390/app12010316
复制
发表时间：
2022-01-01
影响因子：
2.7
通讯作者：
Baglio, Francesca
中科院分区：
文献类型：
--
作者：
Bonilauri, Augusto;Sangiuliano Intra, Francesca;Baglio, Francesca
作为神经退行性疾病和神经康复的特色应用，随访除了临床认知和运动评分外，还需要工具证据。功能磁共振成像通常不适合，要么是因为患者没有资格接受磁共振成像，要么是因为它容易出现运动伪影。FNIRS技术减弱了这些限制，因为大脑激活可以在更多功能的实验环境中测量，即使仅限于皮质活动。因此，走向完全临床接受fNIRS的路线图旨在为fMRI提供一个额外的、更灵活的解决方案，但它需要标准的信号处理和协议。本研究从上述应用的角度对不同的处理方法进行了比较。功能性近红外光谱技术(FNIRS)可以捕捉皮质区域的激活和抑制，并在临床研究中实现一种可行的神经监测方法。与更先进的方法相比，连续波fNIRS(CW-fNIRS)目前在临床上使用，因为它对整个颅下皮质的标测简单。相反，它往往缺乏硬件减少混杂因素，强调正确的信号处理的重要性。提出的流水线包括运动伪影抑制(MAR)、带通滤波(BPF)和主成分分析(PCA)。对23名青壮年志愿者在运动抓取任务下的8种MAR算法进行了比较。显示的是单个受试者的例子，后面是按单个步骤和累积值计算的能量减少百分比(ERD%)统计数据。将血流动力学响应函数的块平均值与广义线性模型拟合进行比较。给出了显著激活/抑制的图谱。与初始原始信号能量相关的预处理信号的平均ERD%达到4%。一个测试过的多通道MAR变种在4倍的扩展窗口上显示出过度校正。所有的MAR算法都在对侧运动区发现了类似的激活。综上所述，提出了单通道MAR算法，其次是BPF和PCA算法。我们的结果也证实了全皮质标测对于fNIRS整合在临床应用中的重要性。
Featured Application In neurodegenerative diseases and neurorehabilitation, follow-up requires instrumental evidence besides clinical cognitive and motor scores. fMRI is frequently not suitable, either because patients are not eligible for an MRI or because it is prone to motion artifacts. The fNIRS technique attenuates these limitations since brain activations can be measured in a more versatile experimental setting, even if restricted to cortical activity. Therefore, the roadmap towards full clinical acceptance of fNIRS aims to provide an additional and more flexible solution to fMRI when not available or feasible, but it needs standard signal processing and protocols. This study provides comparisons of alternative processing methods in the above applicative perspective. Functional Near-Infrared Spectroscopy (fNIRS) captures activations and inhibitions of cortical areas and implements a viable approach to neuromonitoring in clinical research. Compared to more advanced methods, continuous wave fNIRS (CW-fNIRS) is currently used in clinics for its simplicity in mapping the whole sub-cranial cortex. Conversely, it often lacks hardware reduction of confounding factors, stressing the importance of a correct signal processing. The proposed pipeline includes movement artifact reduction (MAR), bandpass filtering (BPF), and principal component analysis (PCA). Eight MAR algorithms were compared among 23 young adult volunteers under motor-grasping task. Single-subject examples are shown followed by the percentage in energy reduction (ERD%) statistics by single steps and cumulative values. The block average of the hemodynamic response function was compared with generalized linear model fitting. Maps of significant activation/inhibition were illustrated. The mean ERD% of pre-processed signals concerning the initial raw signal energy reached 4%. A tested multichannel MAR variant showed overcorrection on 4-fold more expansive windows. All of the MAR algorithms found similar activations in the contralateral motor area. In conclusion, single channel MAR algorithms are suggested followed by BPF and PCA. The importance of whole cortex mapping for fNIRS integration in clinical applications was also confirmed by our results.