1.

Changes in Functional Near-Infrared Spectroscopy Signals Represent Neuronal Activity—But Not Always

Functional near-infrared spectroscopy (fNIRS) is an increasingly used method in neuroscience research. It is a hemodynamic-based functional brain imaging technique, like functional magnetic resonance imaging (fMRI), relying on the detection of neurovascular coupling to infer changes in neuronal activity. Neurovascular coupling refers to the fact that changes in local neural activity are correlated with changes in local cerebral blood flow (CBF). The magnitude and spatial location of CBF changes are tightly linked to changes in neuronal activity.^1,2 Neuronal excitation causes changes in oxidative metabolism (neurometabolic coupling) which is linked to neurovascular coupling; blood vessel dilation causes an increase in CBF and cerebral blood volume that overcompensates the metabolic demand. This oversupply of oxygenated blood (functional hyperemia), the cause of which is still a matter of debate,³ causes the fNIRS measurement of oxygenated hemoglobin ([O₂Hb]) to increase, the deoxygenated hemoglobin ([HHb]) to decrease, and the total hemoglobin ([tHb] = [O₂Hb] + [HHb]) to increase. Although this physiological phenomenon is well known and expected, often it is not completely presented in publications (authors tend to focus only on the [O₂Hb] response). In addition, there is a possibility that this pattern of [O₂Hb], [HHb], and [tHb] changes may not uniquely be caused by neurovascular coupling.

The aims of this opinion article are (i) to define and highlight the issue of false positives and negatives in fNIRS, (ii) to describe some of the challenges in recognizing these, and finally (iii) to suggest a way forward in resolving this issue. While recently a substantial number of research and review articles (including ours) have discussed this issue; here, we will attempt to offer a concise and summarized view of the topic that will also represent our opinion. (In this opinion article, we explicitly assume that fNIRS instrumentation can accurately monitor and resolve the concentration changes in [O₂Hb] and [HHb]. The technical specifications and configuration of the fNIRS instrumentation (e.g., type and number of light sources and of photodetectors) can dictate the performance and accuracy of the device in monitoring [O₂Hb] and [HHb]. One aspect, in particular, is the number and choice of wavelengths used for resolving [O₂Hb] and [HHb] as it has been reported that certain wavelength combinations can result in “crosstalk” between the chromophores. By crosstalk we mean to describe the phenomena that a genuine change in one chromophore concentration is also inducing a spurious measured concentration change in another; for a more detailed discussion of this issue see section “Selection of optimum wavelengths” in the recent review article by Scholkmann et al.⁴ and the recent study by Arifler et al.⁵ It is considered good practice for authors to report the wavelengths of their fNIRS instrument used in their study).

2.

What Are False Positives and False Negatives in Functional Near-Infrared Spectroscopy?

We use the terms “false positives” or “false negatives” to state that during a functional experiment a change in the fNIRS signals ([O₂Hb], [HHb], and [tHb]) over a measurement channel that overlays a certain area of the cortex might not be due to the effect of neurovascular coupling, but could also be due to changes in (i) intracerebral hemodynamics caused by task-related systemic activity and/or (ii) extracerebral hemodynamics. These task-related systemic changes may mimic the presence of a neuronally induced hemodynamic response (false positives), or can attenuate (mask) the neuronally induced hemodynamic response (false negatives).

The presence of false positives and false negatives is not a problem exclusively for fNIRS, but is an issue for all hemodynamic-based functional brain imaging techniques including fMRI. However, the sensitivity of fNIRS to hemodynamic and oxygenation changes in the extracerebral (superficial) compartment is a particular confounding factor of this technique.

fNIRS is increasingly being used to investigate complicated cognitive functions, some of which can produce significant changes in systemic variables, which lead to non-neuronal driven changes in hemodynamics/oxygenation happening in the cerebral and extracerebral compartment mainly associated with changes in heart rate (HR), blood pressure, breathing rate, ${\mathrm{CO}}_2$ concentration in the blood, and autonomic nervous system (ANS) activity.

The presence of these systemic effects, and in an extension their consequences for fNIRS measurements, largely depends on the functional protocol (e.g., passive versus active tasks) and the individual subject studied (e.g., two people can react very differently to the same stressor).

3.

What Factors Cause False Positives and False Negatives in Functional Near-Infrared Spectroscopy?

As described recently in detail,⁴ each fNIRS signal comprises different components that can be classified according to their (i) source (cerebral versus extracerebral), (ii) stimulus/task relation (evoked versus nonevoked), and (iii) physiological cause (neuronal versus systemic). The monitoring of the hemodynamic response due to neurovascular coupling that fNIRS aims to detect is only one of these components (i.e., the component neuronal/task-evoked/cerebral), while all the other components are the physiological noise that act as confounders in fNIRS studies (see Fig. 1).

Fig. 1

Visualization of the six components of the fNIRS signals ([O₂Hb], [HHb], [tHb]). (a) An erroneous assumption is that the fNIRS signal represents only changes associated with functional brain activity due to neurovascular coupling. (b) In reality, the fNIRS signals comprise six components so that five components are potentially confounders in every fNIRS study. The contribution of the components to the fNIRS signal is visualized by color-coding (red: 100%, white: 0%). The nonevoked/cerebral/neuronal and nonevoked/cerebral/systemic components can contribute to the fNIRS signal to a significant degree; however, the evoked changes can be generally stronger in an experimental paradigm assessing functional, task-related, brain-activity.

Task-evoked changes in respiration and thus partial pressure of end-tidal carbon dioxide (${\mathrm{P}}_{\mathrm{ET}}{\mathrm{CO}}_2$), a reliable and accurate estimate of the partial pressure of carbon dioxide in the arterial blood (${\mathrm{PaCO}}_2$), was demonstrated in multimodal fNIRS studies (i.e., fNIRS studies including the measurements of systemic physiological changes using additional monitoring devices) investigating different tasks, e.g., speaking (maximal effect: $\mathrm{\Delta }{\mathrm{PaCO}}_2\approx -9\mathrm{mm}\mathrm{Hg}$),⁶ mental arithmetic ($\mathrm{\Delta }{\mathrm{PaCO}}_2\approx 0.5$ to 3 mm Hg),^6,7 and inner speech (i.e., speaking without vocalization, $\mathrm{\Delta }{\mathrm{PaCO}}_2\approx -0.5\mathrm{mm}\mathrm{Hg}$).⁸

Task-evoked change in mean arterial blood pressure ($\mathrm{\Delta }\mathrm{MAP}$) has been demonstrated for a number of different experimental fNIRS protocols, e.g., arm-raising ($\mathrm{\Delta }\mathrm{MAP}\approx 6\mathrm{mm}\mathrm{Hg}$),⁹ visual stimulation ($\mathrm{\Delta }\mathrm{MAP}\approx 2\mathrm{mm}\mathrm{Hg}$),¹⁰ anagram solving ($\mathrm{\Delta }\mathrm{MAP}\approx 3$ to 5 mm Hg,^11,12 6 to 7 mm Hg),¹³ or hypercapnia ($\mathrm{\Delta }\mathrm{MAP}\approx 4\mathrm{mm}\mathrm{Hg}$),¹⁴ hypocapnia ($\mathrm{\Delta }\mathrm{MAP}\approx 4.5\mathrm{mm}\mathrm{Hg}$;¹⁴ $\mathrm{\Delta }\mathrm{MAP}\approx -10\mathrm{mm}\mathrm{Hg}$),¹⁵ hypoxemia ($\mathrm{\Delta }\mathrm{MAP}\approx 2\mathrm{mm}\mathrm{Hg}$),¹⁵ and hyperoxia ($\mathrm{\Delta }\mathrm{MAP}\approx 2\mathrm{mm}\mathrm{Hg}$).¹⁵

Since MAP is a function of cardiac output (CO), systemic vascular resistance (SVR), and central venous pressure (CVP) ($\mathrm{MAP}=\mathrm{CO}\times \mathrm{SVR}+\mathrm{CVP}$), and since CO is the product of HR and stroke volume (SV) ($\mathrm{CO}=\mathrm{HR}\times \mathrm{SV}$), task-evoked changes in HR and SVR will also have an effect on MAP.

A task-evoked change in the activity of the ANS is another factor potentially able to influence fNIRS signals since the ANS has a direct influence on CO, SVR, and in particular on blood flow (BF) in the extracerebral layer. All three of these physiological variables (i.e., respiration leading to ${\mathrm{CO}}_2$ changes, blood pressure, and ANS activity) affect the BF both in the cerebral compartment and in the extracerebral compartment (i.e., the scalp BF, ScBF). Changes in either the CBF or ScBF may have an impact on the recorded fNIRS signals.

The coupling strength between changes in blood pressure, ${\mathrm{CO}}_2$, or ANS activity with BF changes depends at least on two factors: (i) the type of compartment (cerebral versus extracerebral) and (ii) the speed of change (slow versus fast, i.e., minutes range versus seconds range). Table 1 shows the coupling strength values based on the present physiological knowledge. It is interesting to notice that the coupling is generally quite strong in all cases except between slow changes in blood pressure and CBF, where the coupling is limited by dynamic cerebral autoregulation. Note, however, that the efficiency of cerebral autoregulation has also been controversially discussed in the recent literature, with doubt cast on the shape and boundaries of the CBF-transfer function as traditionally described.¹⁶

Table 1

Overview (according to our own opinion based on published literature and own experience) of the coupling strength between BF and systemic variables, i.e., blood pressure, PaCO2, and ANS activity. “*” indicates that at present the coupling strength is not known or controversially discussed. For a critical review of the coupling factors, see Refs. 16 and 17. Coupling between hemodynamic changes in the cerebral layer concern the CBF, whereas changes in the extracerebral compartment are linked to scalp BF.

Extracerebral confounders are an issue because the diffusely reflected light in fNIRS is attenuated by a significant degree in the extracerebral tissue layer of the head and not only in the cerebral compartment.^18–21 The penetration depth ($D$) of light emitted in the head tissue layers is substantially less than half the source–detector separation (SDS) as predicted by the diffusion approximation²² (i.e., $D<0.5$ SDS). Changes in hemodynamics and oxygenation in the extracerebral layer will have a strong impact on the fNIRS signals. This issue has been known in principle for many years (e.g., Refs. 2324.25.26.–27) but the real significance becoming increasingly apparent with a growing number of studies showing that fNIRS is affected by these extracerebral changes and also that the extracerebral hemodynamic changes are spatially heterogeneous.^28–31

4.

How Can We Avoid and/or Remove False Positives and Negatives in Functional Near-Infrared Spectroscopy Studies?

False positives and negatives can be avoided, minimized, and/or dealt with by adopting one or more of the following approaches (listed in no particular order of significance): (i) careful experimental design (e.g., avoiding tasks with strong systemic activation, avoiding additional stress and discomfort for the subject), (ii) depth-resolved fNIRS techniques [e.g., multidistance (MD) measurements], (iii) appropriate statistical and/or signal processing methods [e.g., independent-component analysis (ICA), principal component analysis (PCA), adaptive filtering, analysis with a general linear model (GLM) with systemic regressors], (iv) multimodal monitoring (e.g., measurement of local ScBF or measurements of signals from the systemic physiology).

Although we have focused above on fNIRS monitoring during functional paradigms, there are additional issues to be considered when there is no functional contrast, i.e., in a study assessing the resting-state brain activity. Both neuronal and non-neuronal hemodynamic generating factors are present during the resting-state and should be considered when designing the experimental protocol [e.g., possible environmental noise, change in resting-state activity due to increased drowsiness during the experiment, increased subject stress due to the fNIRS optode(s) attachment which can cause unpleasant sensations and thus induce a change in ANS activity]. When conducting resting-state fNIRS studies one should be aware of the fact that more and more studies report the influence of systemic hemodynamic changes on resting-state fNIRS and fMRI signals. For example, recently the findings of Tong et al.⁴⁵ suggested, based on an fMRI study, that the resting-state “may reflect, to some extent, vascular anatomy associated with systemic fluctuations, rather than neuronal connectivity.” Also, Zhu et al.⁴⁶ recently concluded from their resting-state (rs) fMRI findings that they “raise a critical question of whether a large portion of rs-fMRI signals can be attributed to the vascular effects produced from upstream changes in cerebral hemodynamics.” Since fNIRS signals are additionally affected by changes in ScBF, careful interpretation of resting-state fNIRS signals is warranted. For resting-state fNIRS studies, it thus makes sense to name the measured connectivity accordingly, i.e., “vascular connectivity”⁴⁷ or “sum of intracerebral and extracerebral hemodynamic functional connectivity.”⁴⁸

5.

Way Forward

There are many factors that should be considered in order to improve future fNIRS studies. What needs to be recognized immediately is the imperativeness of considering the time dynamics of both [O₂Hb] and [HHb] signals in fNIRS data analysis, and their interrelationship before formulating any neuroscientific conclusions. Reporting both [O₂Hb] and [HHb] changes, instead of only one of them, allows better physiological interpretation of the functional experimental results (e.g., when both [O₂Hb] and [HHb] show a strong increase and/or when both [O₂Hb] and [HHb] show a strong decrease during the task periods then the signals are probably confounded by systemic changes). However observing the typical pattern of the hemodynamic response during functional activation (i.e., increase in [O₂Hb] and decrease in [HHb]) is not sufficient for concluding that no systemic confounding effect took place. Synchronous measurements of systemic physiology, currently easily achievable with highly portable and relatively inexpensive instrumentation, can then also provide an accessible way to regress out systemic interference and identify individuals with strong task-related systemic changes.

Both [O₂Hb] and [HHb] can be influenced by systemic physiological changes (occurring in the cerebral and/or the extracerebral compartment). However, the degree of the influence often varies and also depends on the origin and type of the systemic change. Kirilina et al.²⁹ demonstrated that [O₂Hb] signals are more strongly affected by global processes in both extracerebral and intracerebral compartments, and local scalp BF regulation, while [HHb] signals are less contaminated by extracerebral processes. The physiological reason for this may be the fact that arterioles are enervated to a larger degree by the fibers from the sympathetic nervous system than the venules;^49,50 during activity of the ANS, the [O₂Hb] signals seem to be more affected than the [HHb] signals—a conclusion also supported by recent findings of Haeussinger et al.⁵¹ In addition, it is well known that cardiac oscillations are more prominent in the [O₂Hb] compared to the [HHb]. Although there is evidence of the [HHb] signal to be less contaminated by systemic changes, this does not imply that the [HHb] signal is not confounded by systemic changes.

In the present state of fNIRS research, most studies are using cw-fNIRS systems that implement only measurements at a fixed 3 cm SDS. In this case, it is recommended that ScBF effects be corrected by either (i) adding additional short channels and using them to regress out the superficial hemodynamics (by using one of the short-channel regression methods developed so far, e.g., Refs. 39, 41, and 5253.54.55.–56), (ii) by calculating the mean signal over all channels and using this channel as the superficial regressor (e.g., Refs. 57 and 58), or (iii) by combining both of these approaches.⁵⁹

More and more fNIRS instrumentation companies currently provide alternative and/or add-on solutions to allow the implementation of MD probe geometries with subsequent short-channel regression frameworks. In the near future, such instrumentation advances will allow fNIRS measurements that better represent the neuronal, cerebral, and evoked responses.