Current Status and Issues Regarding Pre-processing of fNIRS Neuroimaging Data: An Investigation of Diverse Signal Filtering Methods Within a General Linear Model Framework

Abstract

Functional near-infrared spectroscopy (fNIRS) research articles show a large heterogeneity in the analysis approaches and pre-processing procedures. Additionally, there is often a lack of a complete description of the methods applied, necessary for study replication or for results comparison. The aims of this paper were (i) to review and investigate which information is generally included in published fNIRS papers, and (ii) to define a signal pre-processing procedure to set a common ground for standardization guidelines. To this goal, we have reviewed 110 fNIRS articles published in 2016 in the field of cognitive neuroscience, and performed a simulation analysis with synthetic fNIRS data to optimize the signal filtering step before applying the GLM method for statistical inference. Our results highlight the fact that many papers lack important information, and there is a large variability in the filtering methods used. Our simulations demonstrated that the optimal approach to remove noise and recover the hemodynamic response from fNIRS data in a GLM framework is to use a 1000th order band-pass Finite Impulse Response filter. Based on these results, we give preliminary recommendations as to the first step toward improving the analysis of fNIRS data and dissemination of the results.

Cerebral hemodynamic responses in preterm-born neonates to visual stimulation: classification according to subgroups and analysis of frontotemporal–occipital functional connectivity

“…3. To remove high-frequency noise, improve the signalto-noise ratio (SNR) and to enhance the spectral band with the physiologically relevant information, all NIRS data were band-pass filtered (lower cut-off frequency: 0.0167 Hz ¼ 60 s, upper cut-off frequency: 0.2 Hz ¼ 5 s, phase-corrected finite-impulse response filter of order 500; the high order of the filter was chosen according to a recent investigation pointing out the necessity to use a high-order filter for fNIRS signal analysis 33 ) and afterward low-pass filtered with a Savitzky-Golay filter to further reduce noise (fifth order, window length = 4 s). The selection of the filter type and filter parameter values was done empirically to obtain the optimal SNR.…”

A Mini-Review on Functional Near-Infrared Spectroscopy (fNIRS): Where Do We Stand, and Where Should We Go?

“…The topics include: The basics of NIR photon migration, the state of the art of instrumentations/signal processing/statistical analysis, and the integration of fNIRS with other neuroimaging methods. Factors influencing fNIRS data and recommendations 2010 Orihuela-Espina [34] Caps for long term fNIRS measurements 2015 Kassab [35] Selection of the optimum source-detector distance 2015 Brigadoi [36] Mayer waves interference 2016 Yücel [37] Multiple components of the fNIRS signal 2016 Tachtsidis [19] Signal pre-processing procedures 2019 Pinti [38] Anatomical guidance for fNIRS 2014 Tsuzuki [39] 2015…”

Application of functional near-infrared spectroscopy in the healthcare industry: A review

“…The methods adopted to examine hemodynamic changes via fNIRS are diverse, and it has been eagerly proposed that a standard procedure should be followed. 306 If studies follow a standard data processing pipeline, they can be compared, and a veri¯able knowledge database can be established. The acquired raw fNIRS data are a®ected by various noise sources like physiological (respiratory, cardiac, Mayer waves, etc.…”

“…307 These noises reduce the signal-to-noise ratio of the desired signal, and they can override the neuronal activation for the task performed following an experimental paradigm not carefully designed. 306,308 Therefore, the removal of these noises to obtain a clean fNIRS signal is a pivotal step. Various techniques are employed to remove them as they are identi¯ed by their approximate frequencies like cardiac (1 Hz), respiratory (0.3 Hz), and Mayer waves (0.1 Hz).…”

“…Some researchers prefer to use a¯lter on the light-intensity signals while others use on the hemodynamic signals. 306 In the recent past, the use of short-separation channels to remove the extracerebral e®ects is gaining attention. The short-separation channels are con¯gured by making the source-detector separation lower than 10 mm.…”