**2.2 Application of NIRS to living tissues**

At least six conditions have to be fulfilled in order for the Beer-Lambert's law to be valid:


Living tissues, especially in humans, are doubtlessly among the most structured and complex in the universe. Their characteristics do not match with the Beer-Lambert's law prerequisites on numerous points. Therefore, the modified Beer-Lambert's law has to be applied in NIRS. As stated in the fifth point of the prerequisites, the incident light must be monochromatic (i.e. only one wavelength λ). In human tissues, lots of chemical species absorb light and account for its loss when travelling. However, there is a range of wavelengths at which light travel is much facilitated. Intuitively, when, in a dark environment, one looks at a flashlight through his finger or his hand, red is invariably the dominant colour. The physical explanation is that the red light travel through the human tissues is easier than for any other wavelengths. Implicitly, in the red portion of the visible

What Does Cerebral Oxygenation Tell Us About Central Motor Output? 301

In eq. 8, OD is measured using Bouguer's idea (eq.1), ε is known from the physicists who are able to measure it (fig.2), ODr and l are unknown but necessary to the computation of c.

Fig. 2. Molar extinction coefficient for haemoglobin in water. O2Hb: plain line; HHb: dashed line; x axis: wavelength (λ) in nm; y axis: extinction coefficient (ε) in cm-1/M. Data compiled

ODr is not expected to change radically in a short lap of time. In other words, it is considered constant between two light impulsions a few tenths of seconds away. Therefore,

<sup>c</sup> <sup>c</sup> <sup>ε</sup>. <sup>l</sup> <sup>ε</sup>. <sup>l</sup>

<sup>c</sup> <sup>c</sup> <sup>ε</sup>. <sup>l</sup>

tissues (Ijichi et al., 2005). Three methods are available to approach the path length:

λ t0 (λ) ( λ t1 λ)

λ t0 λ t1

(9)

(10)

(λ) (λ)

(λ)

Eq. 10 states that the concentration variations depend on the measured OD variations. The advantage of the subtraction in eq. 9 is to get rid of the unknown ODr. However, the absolute concentration of the chromophore becomes unknown as only a concentration

The last unknown parameter missing to compute the concentration variation is l, the path length of light between the transmitter and the receptor (fig. 3). Its measure is almost impossible due to the numerous interactions between the matter and the light in living

(OD OD )

(OD ODr ) (OD ODr )

considering two light impulsions at t0 and t1, it is possible to write:

t0 t1

t0 t1

difference (or variation) can be computed.

by Scott Prahl (Prahl, 2008).

which simplifies into

light, there are a limited number of chemical species which are responsible for the majority of light absorption and diffusion. These species are known to give its colour to the tissue and have been judiciously named chromophores. In human tissues, it is well known that haemoglobin is responsible for the colour given to tissues; in physics, haemoglobin is the chromophore whose concentration can be measured using Bouguer's idea.

#### **2.2.1 The chromophores**

Haemoglobin is a metalloprotein which transports 98% of the oxygen in most vertebrates' blood. When oxygen binds to the iron complex, it causes the iron ion to move back, and changes the optical properties of the molecule. At the human scale the phenomenon is perceptible and results in the long standing view that the red blood is filled-up with oxygen while the blue one has lost the majority of its initial quantity of oxygen. In physics, it can be considered that there are two distinct chromophores: oxygenated haemoglobin (O2Hb) and deoxygenated haemoglobin (HHb). Therefore, according to Bouguer's idea one can compute the concentration of oxy and deoxyhaemoglobin in a tissue by measuring the changes in OD (eq. 4). However, the OD in human tissues is not strictly dependant on haemoglobin. In an imaginary case where there would be no haemoglobin in the explored area, the tissue would still absorb light. Consequently, eq.4 should be rewritten as:

$$\text{OD}\_{\left(\lambda\right)} = \varepsilon\_{\left(\lambda\right)} \cdot \text{c.l} + \text{ODr}\_{\left(\lambda\right)}\tag{5}$$

where ODr is the y-intercept of the linear relation and denotes the OD of the living tissue when there is no haemoglobin. λ denotes the chosen wavelength for the monochromatic light.

#### **2.2.2 Two wavelengths**

The Beer-Lambert's law states that the measured optical density is the sum of the absorbance of the two chromophores. Eq. 5 becomes eq. 6 with the two chromophores appearing:

$$\text{OD}\_{\{\lambda\}} = \varepsilon\_{\text{O2Hb}(\lambda)}.\text{c}\_{\text{O2Hb}}.1 + \varepsilon\_{\text{HHb}(\lambda)}.\text{c}\_{\text{HHb}}.1 + \text{ODr}\_{\{\lambda\}}\tag{6}$$

There are two unknowns in eq. 6 (ie. cO2Hb and cHHb). Thus, two equations are needed to solve the system. The two equations are provided by firing at two different wavelengths λ1 and λ2.

$$\begin{cases} \text{OD}\_{\text{(\lambda1)}} = \text{\text{\textdegree{}}\_{\text{O2Hb}(\lambda1)} \text{ \textdegree{}}\_{\text{O2Hb}} \text{ \textdegree{}}\_{\text{I}} + \text{\text{\textdegree{}}\_{\text{HHb}(\lambda1)} \text{ \textdegree{}}\_{\text{HHb}} \text{ \textdegree{}}\_{\text{I}} + \text{OD}\text{\textdegree{}}\_{\text{\textdegree{}}\text{M}}\\ \text{OD}\_{\text{(\lambda2)}} = \text{\textdegree{}}\_{\text{O2Hb}(\lambda2)} \text{ \textdegree{}}\_{\text{O2Hb}} \text{ \textdegree{}}\_{\text{I}} + \text{\text{\textdegree{}}\_{\text{HHb}} \text{ \textdegree{}}\_{\text{(\lambda 2)}} \text{ \textdegree{}}\_{\text{HHb}} \text{ \textdegree{}}\_{\text{I}} + \text{OD}\text{\textdegree{}}\_{\text{(\lambda 2)}} \end{cases} \tag{7}$$

The main idea is that one needs as many wavelengths as there are chromophores in the investigated area. Only one equation is exposed further down this line for clarity purpose. Note that NIRS systems perform every computation to solve the systems of equations.

#### **2.2.3 Application of the modified Beer-Lambert's law**

Since physiologists use NIRS to compute the haemoglobin concentration, the modified Beer-Lambert's law is then written:

$$\mathbf{c} = \frac{\text{(OD}\_{\{\lambda\}} - \text{ODr}\_{\{\lambda\}})}{\mathbf{c}\_{\{\lambda\}} \cdot 1} \tag{8}$$

In eq. 8, OD is measured using Bouguer's idea (eq.1), ε is known from the physicists who are able to measure it (fig.2), ODr and l are unknown but necessary to the computation of c.

Fig. 2. Molar extinction coefficient for haemoglobin in water. O2Hb: plain line; HHb: dashed line; x axis: wavelength (λ) in nm; y axis: extinction coefficient (ε) in cm-1/M. Data compiled by Scott Prahl (Prahl, 2008).

ODr is not expected to change radically in a short lap of time. In other words, it is considered constant between two light impulsions a few tenths of seconds away. Therefore, considering two light impulsions at t0 and t1, it is possible to write:

$$\mathbf{c}\_{t0} - \mathbf{c}\_{t1} = \frac{(\mathbf{OD}\_{\{\lambda\}\mathbf{t}0} - \mathbf{ODr}\_{\{\lambda\}})}{\varepsilon\_{\{\lambda\}} \cdot 1} - \frac{(\mathbf{OD}\_{\{\lambda\}\mathbf{t}1} - \mathbf{ODr}\_{\{\lambda\}})}{\varepsilon\_{\{\lambda\}} \cdot 1} \tag{9}$$

which simplifies into

300 Neuroimaging – Cognitive and Clinical Neuroscience

light, there are a limited number of chemical species which are responsible for the majority of light absorption and diffusion. These species are known to give its colour to the tissue and have been judiciously named chromophores. In human tissues, it is well known that haemoglobin is responsible for the colour given to tissues; in physics, haemoglobin is the

Haemoglobin is a metalloprotein which transports 98% of the oxygen in most vertebrates' blood. When oxygen binds to the iron complex, it causes the iron ion to move back, and changes the optical properties of the molecule. At the human scale the phenomenon is perceptible and results in the long standing view that the red blood is filled-up with oxygen while the blue one has lost the majority of its initial quantity of oxygen. In physics, it can be considered that there are two distinct chromophores: oxygenated haemoglobin (O2Hb) and deoxygenated haemoglobin (HHb). Therefore, according to Bouguer's idea one can compute the concentration of oxy and deoxyhaemoglobin in a tissue by measuring the changes in OD (eq. 4). However, the OD in human tissues is not strictly dependant on haemoglobin. In an imaginary case where there would be no haemoglobin in the explored area, the tissue would

where ODr is the y-intercept of the linear relation and denotes the OD of the living tissue when there is no haemoglobin. λ denotes the chosen wavelength for the monochromatic light.

The Beer-Lambert's law states that the measured optical density is the sum of the absorbance of the two chromophores. Eq. 5 becomes eq. 6 with the two chromophores

There are two unknowns in eq. 6 (ie. cO2Hb and cHHb). Thus, two equations are needed to solve the system. The two equations are provided by firing at two different wavelengths λ1 and λ2.

> OD ε .c .l ε .c .l ODr OD ε . c . l ε . c . l ODr

The main idea is that one needs as many wavelengths as there are chromophores in the investigated area. Only one equation is exposed further down this line for clarity purpose. Note that NIRS systems perform every computation to solve the systems of equations.

Since physiologists use NIRS to compute the haemoglobin concentration, the modified Beer-

c = (λ) (λ) (λ) (OD ODr ) ε . l

 

(λl) O2Hb(λ1) O2Hb HHb(λ1) HHb (λl) (λ2) O2Hb(λ2) O2Hb HHb(λ2) HHb (λ2)

OD(λ) (λ) (λ) ε .c .l ODr (5)

(7)

(8)

OD(λ) O2Hb(λ) O2Hb HHb(λ) HHb (λ) ε .c .l ε .c .l ODr (6)

chromophore whose concentration can be measured using Bouguer's idea.

still absorb light. Consequently, eq.4 should be rewritten as:

**2.2.3 Application of the modified Beer-Lambert's law** 

**2.2.1 The chromophores** 

**2.2.2 Two wavelengths** 

Lambert's law is then written:

appearing:

$$\mathbf{c}\_{t0} - \mathbf{c}\_{t1} = \frac{(\mathbf{OD}\_{\left(\lambda\right)t0} - \mathbf{OD}\_{\left(\lambda\right)t1})}{\varepsilon\_{\left(\lambda\right)} \cdot 1} \tag{10}$$

Eq. 10 states that the concentration variations depend on the measured OD variations. The advantage of the subtraction in eq. 9 is to get rid of the unknown ODr. However, the absolute concentration of the chromophore becomes unknown as only a concentration difference (or variation) can be computed.

The last unknown parameter missing to compute the concentration variation is l, the path length of light between the transmitter and the receptor (fig. 3). Its measure is almost impossible due to the numerous interactions between the matter and the light in living tissues (Ijichi et al., 2005). Three methods are available to approach the path length:

What Does Cerebral Oxygenation Tell Us About Central Motor Output? 303

of photons between the light transmitter and the receptor (Hiraoka et al., 1993; Simpson et al., 1998; Zhang et al., 2007a, b). This is the most precise method nowadays, usable with regular measurement devices but costly in terms of computation. The Monte Carlo method might be performed after the monitoring session as computers may not be powerful enough to ensure simultaneously proper recording of the data and Monte Carlo analysis (Avrillier et al., 1998a; Avrillier et al., 1998b). Roughly, for all methods the maximum depth of the mean

NIRS data consist of oxy and deoxyhaemoglobin time series (Fig. 4 and Fig. 5), with sampling rate usually ranging from 2 to 20Hz, and occasionally above. Usual measurement sites exclude locations where large arteries or veins would be reachable by the NIRS light as experimenters are rather interested in tissue data. In the tissues, the light crosses three types

NIRS signal is believed to originate in its major part from the venous compartment (approx. 70%); however, vasomotion makes the part of each segment variable (Bourdillon et al., 2009; Peltonen et al., 2009). Briefly, capillaries form an extensive network which connects the arterial and venous sides of the vascular system. The blood flow through a given capillary bed strongly depends on the vascular tone of the parent arteriole and the pre-capillary sphincters. Both adjust the local blood flow to meet the physiological demands. Despite the smooth muscles of the arterioles and sphincters are connected to the sympathetic nervous system, the vascular tone is largely dependent on the local factors (Segal, 2005). Concerning the motor areas of the brain, it is generally assumed that, when activated, the neurons increase their firing rates to generate the motor command and thus increase their metabolic demands (Villringer, 1997). One of the consequences is to increase the local blood flow; this

Fig. 4. Example of a NIRS signal pattern over the motor cortex during a 20 seconds lasting

path length of light is believed to be half of the emitter to receiver distance.

**3. Signal characteristics and interpretations** 



handgrip task (Wolf et al., 2007).

of blood vessels:


Fig. 3. Schematic representation of NIRS applied to human cerebral tissues. The mean path length of light represents the l.DPF in eq.11. Roughly, the maximum depth of the mean path length of light is believed to be half of the emitter to receiver distance.

The DPF method is doubtlessly the easiest to use but also the less precise and less satisfactory. Regrettably it is the most common method nowadays. In this method, l is considered the most direct way between the light transmitter and receptor, DPF is multiplied to l to lengthen the global path length, eq. 10 is then written:

$$
\Delta \mathbf{c} = \frac{\Delta \text{OD}\_{\{\lambda\}}}{\varepsilon\_{\{\lambda\}} \cdot 1. \text{ DPF}} \tag{11}
$$

DPF is arbitrarily set from abacus found in the literature. Only a few studies give the DPF, often as a function of age (Duncan et al., 1995; Essenpreis et al., 1993a; Essenpreis et al., 1993b; Firbank et al., 1993; Ijichi et al., 2005; Kohl et al., 1998a; Kohl et al., 1998b; Nolte et al., 1998; Pringle et al., 1999; Ultman and Piantadosi, 1991; van der Zee et al., 1992; Zhao et al., 2002). Another way to approach the path length of light is to measure the time of flight between the light transmitter and the receptor. The speed of light in the vacuum is used to compute the path length. This method is more precise than the DPF method but costly financially and in terms of load of computation. Billion of photons are detected by the receptor at each light impulsion. One of the advantages is the possibility to select the photons to study; the first detected photons have a priori a shorter path length, which means that they did not go deep into the tissues (Ferrante et al., 2009). The latest photons, which have a longer path length, went a priori deeper into the tissues and carry more information. Finally, the Monte Carlo simulation is a statistical method representing the distribution of energy in the explored volume. It is a way to assume the random path length of photons between the light transmitter and the receptor (Hiraoka et al., 1993; Simpson et al., 1998; Zhang et al., 2007a, b). This is the most precise method nowadays, usable with regular measurement devices but costly in terms of computation. The Monte Carlo method might be performed after the monitoring session as computers may not be powerful enough to ensure simultaneously proper recording of the data and Monte Carlo analysis (Avrillier et al., 1998a; Avrillier et al., 1998b). Roughly, for all methods the maximum depth of the mean path length of light is believed to be half of the emitter to receiver distance.

#### **3. Signal characteristics and interpretations**

NIRS data consist of oxy and deoxyhaemoglobin time series (Fig. 4 and Fig. 5), with sampling rate usually ranging from 2 to 20Hz, and occasionally above. Usual measurement sites exclude locations where large arteries or veins would be reachable by the NIRS light as experimenters are rather interested in tissue data. In the tissues, the light crosses three types of blood vessels:


302 Neuroimaging – Cognitive and Clinical Neuroscience

Fig. 3. Schematic representation of NIRS applied to human cerebral tissues. The mean path length of light represents the l.DPF in eq.11. Roughly, the maximum depth of the mean path

The DPF method is doubtlessly the easiest to use but also the less precise and less satisfactory. Regrettably it is the most common method nowadays. In this method, l is considered the most direct way between the light transmitter and receptor, DPF is

(λ)

DPF is arbitrarily set from abacus found in the literature. Only a few studies give the DPF, often as a function of age (Duncan et al., 1995; Essenpreis et al., 1993a; Essenpreis et al., 1993b; Firbank et al., 1993; Ijichi et al., 2005; Kohl et al., 1998a; Kohl et al., 1998b; Nolte et al., 1998; Pringle et al., 1999; Ultman and Piantadosi, 1991; van der Zee et al., 1992; Zhao et al., 2002). Another way to approach the path length of light is to measure the time of flight between the light transmitter and the receptor. The speed of light in the vacuum is used to compute the path length. This method is more precise than the DPF method but costly financially and in terms of load of computation. Billion of photons are detected by the receptor at each light impulsion. One of the advantages is the possibility to select the photons to study; the first detected photons have a priori a shorter path length, which means that they did not go deep into the tissues (Ferrante et al., 2009). The latest photons, which have a longer path length, went a priori deeper into the tissues and carry more information. Finally, the Monte Carlo simulation is a statistical method representing the distribution of energy in the explored volume. It is a way to assume the random path length

λ

(11)

OD <sup>c</sup> ε . l. DPF 

length of light is believed to be half of the emitter to receiver distance.

multiplied to l to lengthen the global path length, eq. 10 is then written:




NIRS signal is believed to originate in its major part from the venous compartment (approx. 70%); however, vasomotion makes the part of each segment variable (Bourdillon et al., 2009; Peltonen et al., 2009). Briefly, capillaries form an extensive network which connects the arterial and venous sides of the vascular system. The blood flow through a given capillary bed strongly depends on the vascular tone of the parent arteriole and the pre-capillary sphincters. Both adjust the local blood flow to meet the physiological demands. Despite the smooth muscles of the arterioles and sphincters are connected to the sympathetic nervous system, the vascular tone is largely dependent on the local factors (Segal, 2005). Concerning the motor areas of the brain, it is generally assumed that, when activated, the neurons increase their firing rates to generate the motor command and thus increase their metabolic demands (Villringer, 1997). One of the consequences is to increase the local blood flow; this

Fig. 4. Example of a NIRS signal pattern over the motor cortex during a 20 seconds lasting handgrip task (Wolf et al., 2007).

What Does Cerebral Oxygenation Tell Us About Central Motor Output? 305

blood concentration in oxygen or carbon dioxide, baroreflexes and neural feedbacks from metabo and mecano receptors in the skeletal muscle potentially affect the vascular tone and thus the NIRS signal. Consequently, in subjects exercising at high metabolic rates (elevated oxygen consumption), with hyperventilation, hyper/hypocapnia and high cardiac output, the NIRS signal is rather dependant on systemic variables than on motor command (Pereira et al., 2007; Rasmussen et al., 2007). Fig. 5 shows a typical NIRS response from the motor cortex of a subject exercising on a cycloergometer above the ventilatory threshold for 20 minutes (Rooks et al., 2010; Rupp and Perrey, 2008). The amplitudes of the variations are way larger as compared with fig. 4 and the "activation pattern" is altered as there is no apparent decrease in deoxyhaemoglobin. In any case, the interpretation of the NIRS signals has to be modulated following the experimental design and the systemic conditions

As shown in fig. 4 or fig. 5, there is a delay between the stimulus, the neural responses and the hemodynamic modifications as detected by the NIRS systems (Cui et al., 2010b; Yasui et al., 2010). If the NIRS signal depends on the motor command, the delay has been shown to range between 2 and 5 seconds (Fig. 4). In the case of fig. 5 the NIRS signal is rather dependant on systemic parameters (i.e. ventilation) and the delay ranges between 1 and 4 minutes. Such variations in delays are due to the facts that the NIRS records of cerebral hemodynamic parameters depend whether on the motor command or on systemic parameters, following the experimental design. To date, the time course analysis of NIRS signals has yet to be established, notably with regards to the transition periods and the

The only parameters measured by NIRS are the optical densities at two (or more) wavelengths as stated in the part 2.2.2 of this article. O2Hb and HHb are directly computed variables; the computer usually performs calculations during data acquisition. Afterwards, experimenters are using to computing other parameters from O2Hb and HHb to present NIRS data. Among the most often found parameters in the literature there are: the difference between O2Hb and HHb (usually abbreviated O2Hbdiff or Hbdiff); the tissue or capillary saturation (usually abbreviated StO2 or ScO2) and the tissue oxygenation index (TOI). TOI,

2

These indicators are thought to summarize O2Hb and HHb signals and reflect tissue oxygenation. However, physical exercise results in a large heterogeneous increase in cerebral oxygenation (Rooks et al., 2010). It seems that the primary factor influencing this increase is the intensity of exercise, followed by the training status of the subjects, age,

In most studies, NIRS data is pre-processed in order to improve the signal quality (Boas et al., 2004). The first step typically aims at removing noise (Gervain et al., 2011). The noise

O Hb O Hb HHb

(12)

(Gervain et al., 2011; Rooks et al., 2010).

**3.2 Delay** 

experimental designs.

**3.4 Pre-processing** 

**3.3 NIRS computed indicators** 

StO2 and ScO2 are given by the simple formula:

TOI= <sup>2</sup>

health status (i.e., patients vs. healthy subjects) and methodology.

phenomenon can be detected by NIRS. It is obvious that the NIRS measurements are indirect with regards to neuronal activity and rely on the assumption that the latter is coupled to blood supply. Moreover, NIRS measures the concentrations of oxy and deoxyhaemoglobin (the sum of both, Hbtot, giving a proxy of local blood volume), not the blood flow nor the oxygen consumption. Fig. 4 shows a typical NIRS record during a simple motor task (handgrip).

#### **3.1 Patterns**

Empirically, activation pattern in the motor cortex is identified as an increase in oxyhaemoglobin concomitant to a decrease in deoxyhaemoglobin (Fig. 4). The reasons which give the activation pattern such a shape are not fully elucidated (Dai et al., 2001; Harada et al., 2006; Matsuura et al., 2011). However, it is commonly thought that the vasodilation caused by the increase in metabolic demand from the firing neurons overcomes the needs in oxygen; which results in an apparent increase in tissue oxygenation as measured by NIRS (Franceschini and Boas, 2004; Gervain et al., 2011; Leff et al., 2011; Rooks et al., 2010; Shibasaki, 2008; Shibuya and Tachi, 2006). The amplitudes of changes in oxy and deoxyhaemoglobin within the motor cortex areas have been shown to be dependent on the force production: the stronger the push, the higher the oxyhaemoglobin (Shibuya and Tachi, 2006; Smith et al., 2003). However, at low levels of force, there might be no detection by the NIRS systems (at least 10% of maximal voluntary contraction needed); while at high levels (about 50% of maximal voluntary contraction and above) there might be no plateau but only a peak in oxyhaemoglobin (Ekkekakis, 2009). This type of activation pattern is valid only for steady systemic variables (ie. globally non moving body). The NIRS signal, as it comes from the circulatory system, is strongly dependent on the cardio-respiratory parameters. Modifications in cardiac output, autonomic nervous system balance, hormonal response,

Fig. 5. Example of a NIRS signal pattern over the motor cortex during a high intensity whole body cycling exercise at a constant work rate from baseline level (warm up) at 600 s. Personal data.

blood concentration in oxygen or carbon dioxide, baroreflexes and neural feedbacks from metabo and mecano receptors in the skeletal muscle potentially affect the vascular tone and thus the NIRS signal. Consequently, in subjects exercising at high metabolic rates (elevated oxygen consumption), with hyperventilation, hyper/hypocapnia and high cardiac output, the NIRS signal is rather dependant on systemic variables than on motor command (Pereira et al., 2007; Rasmussen et al., 2007). Fig. 5 shows a typical NIRS response from the motor cortex of a subject exercising on a cycloergometer above the ventilatory threshold for 20 minutes (Rooks et al., 2010; Rupp and Perrey, 2008). The amplitudes of the variations are way larger as compared with fig. 4 and the "activation pattern" is altered as there is no apparent decrease in deoxyhaemoglobin. In any case, the interpretation of the NIRS signals has to be modulated following the experimental design and the systemic conditions (Gervain et al., 2011; Rooks et al., 2010).

#### **3.2 Delay**

304 Neuroimaging – Cognitive and Clinical Neuroscience

phenomenon can be detected by NIRS. It is obvious that the NIRS measurements are indirect with regards to neuronal activity and rely on the assumption that the latter is coupled to blood supply. Moreover, NIRS measures the concentrations of oxy and deoxyhaemoglobin (the sum of both, Hbtot, giving a proxy of local blood volume), not the blood flow nor the oxygen consumption. Fig. 4 shows a typical NIRS record during a simple

Empirically, activation pattern in the motor cortex is identified as an increase in oxyhaemoglobin concomitant to a decrease in deoxyhaemoglobin (Fig. 4). The reasons which give the activation pattern such a shape are not fully elucidated (Dai et al., 2001; Harada et al., 2006; Matsuura et al., 2011). However, it is commonly thought that the vasodilation caused by the increase in metabolic demand from the firing neurons overcomes the needs in oxygen; which results in an apparent increase in tissue oxygenation as measured by NIRS (Franceschini and Boas, 2004; Gervain et al., 2011; Leff et al., 2011; Rooks et al., 2010; Shibasaki, 2008; Shibuya and Tachi, 2006). The amplitudes of changes in oxy and deoxyhaemoglobin within the motor cortex areas have been shown to be dependent on the force production: the stronger the push, the higher the oxyhaemoglobin (Shibuya and Tachi, 2006; Smith et al., 2003). However, at low levels of force, there might be no detection by the NIRS systems (at least 10% of maximal voluntary contraction needed); while at high levels (about 50% of maximal voluntary contraction and above) there might be no plateau but only a peak in oxyhaemoglobin (Ekkekakis, 2009). This type of activation pattern is valid only for steady systemic variables (ie. globally non moving body). The NIRS signal, as it comes from the circulatory system, is strongly dependent on the cardio-respiratory parameters. Modifications in cardiac output, autonomic nervous system balance, hormonal response,

Fig. 5. Example of a NIRS signal pattern over the motor cortex during a high intensity whole

body cycling exercise at a constant work rate from baseline level (warm up) at 600 s.

motor task (handgrip).

**3.1 Patterns** 

Personal data.

As shown in fig. 4 or fig. 5, there is a delay between the stimulus, the neural responses and the hemodynamic modifications as detected by the NIRS systems (Cui et al., 2010b; Yasui et al., 2010). If the NIRS signal depends on the motor command, the delay has been shown to range between 2 and 5 seconds (Fig. 4). In the case of fig. 5 the NIRS signal is rather dependant on systemic parameters (i.e. ventilation) and the delay ranges between 1 and 4 minutes. Such variations in delays are due to the facts that the NIRS records of cerebral hemodynamic parameters depend whether on the motor command or on systemic parameters, following the experimental design. To date, the time course analysis of NIRS signals has yet to be established, notably with regards to the transition periods and the experimental designs.

#### **3.3 NIRS computed indicators**

The only parameters measured by NIRS are the optical densities at two (or more) wavelengths as stated in the part 2.2.2 of this article. O2Hb and HHb are directly computed variables; the computer usually performs calculations during data acquisition. Afterwards, experimenters are using to computing other parameters from O2Hb and HHb to present NIRS data. Among the most often found parameters in the literature there are: the difference between O2Hb and HHb (usually abbreviated O2Hbdiff or Hbdiff); the tissue or capillary saturation (usually abbreviated StO2 or ScO2) and the tissue oxygenation index (TOI). TOI, StO2 and ScO2 are given by the simple formula:

$$\text{TOI} = \frac{\text{O}\_2\text{H} \text{lb}}{\text{O}\_2\text{H} \text{b} + \text{H} \text{H} \text{b}} \tag{12}$$

These indicators are thought to summarize O2Hb and HHb signals and reflect tissue oxygenation. However, physical exercise results in a large heterogeneous increase in cerebral oxygenation (Rooks et al., 2010). It seems that the primary factor influencing this increase is the intensity of exercise, followed by the training status of the subjects, age, health status (i.e., patients vs. healthy subjects) and methodology.

#### **3.4 Pre-processing**

In most studies, NIRS data is pre-processed in order to improve the signal quality (Boas et al., 2004). The first step typically aims at removing noise (Gervain et al., 2011). The noise

What Does Cerebral Oxygenation Tell Us About Central Motor Output? 307

beds, corresponding to a greater number of neurons (estimated around 300,000 to 500,000) from various depths in the cortex (Gervain et al., 2011). The pool of capillary beds enlighten by a channel is believed to belong to a given cortex area, which supposedly has a single function. This makes a huge simplification if compared to the brain complexity and its capacity of integration, not to mention the neuro-vascular coupling assumption (see part 4.1.)! Moreover, probe placement is based on the skull anatomy as no direct access to the brain is allowed by NIRS (except in the case of fMRI co-recording) giving a probability to fire over multiple cortex areas or even over a wrong area. Additionally, the proportion of excitatory and inhibitory neurons in the volume aimed by NIRS is unknown yet potentially

At this stage of the article, the most impeding factors have been brought to discussion. However, some factors, not directly linked to the NIRS concepts nor to brain characteristics must be debated. Before entering the tissue of interest, light travels through the skin and the fat layers (as well as the hair and skull layers in case of brain investigations, Fig. 3). The skin colour (and hair colour) has been shown to influence light absorption (Pringle et al., 1999). Intuitively, human eyes perceive various skin colours because skin absorbs and reflects light depending on its properties. The same (or the opposite) happens in the near infrared portion of the spectrum. Light skins are believed to absorb light more than dark skins, while Asian originated skins are the less absorbent. NIRS gain or laser power must then be modulated to fit with the skin properties of a given subject; which can be performed automatically by the

Skin blood flow is one of the main confounding factors as the haemoglobin molecules present in the capillary beds located in the skin are the first (and last) exposed to NIRS light (Tew et al., 2010). In exercising subjects, blood flow is increasing in proportion to the intensity of exercise, for well-known thermoregulation reasons. However, skin is not believed to consume more oxygen at high intensity as compared with low intensity exercises. This means that skin blood flow overcomes by far the local metabolic demands;

The fat and bone layers are probably easier to take into account as they can be integrated in the automatic gain setup which occurs in most modern NIRS devices, before data

Finally, gender has been shown to influence NIRS responses to various stimuli, notably

**4. Measuring the brain activities related to the motor stimulation using NIRS** 

Physiological events associated with brain activity can be subdivided into intracellular events, events occurring at the cell membranes and those that are mediated by neurovascular coupling and occur within the vascular space. Increased brain activity is correlated not only with oxygen consumption but also with glucose consumption. The brain has only negligible stores of glucose and therefore relies both on the circulating glucose and on the active transport system which moves glucose across the blood-brain barrier. Increased activity in brain cells is associated with an increase in glucose consumption and

motor, cognitive tasks and emotions (Marumo et al., 2009; Yang et al., 2009).

**4.1 Physiological processes associated with brain activity** 

affects the results.

acquisition.

**3.7 Confounding factors** 

NIRS hardware before starting the data acquisition.

which necessarily biases the NIRS measurements.

comes from the devices as well as from physiological parameters not a priori linked to the stimulation (eg. Exercise) and are thus undesirable (Nolte et al., 1998). This kind of noise is considered high frequency with regards to the frequencies of interest (Cui et al., 2010a). Low-pass filters are used to remove heart rate, blood pressure variations, breath, swallowing etc. Usually, the cut-off frequency ranges between 0.1 and 1Hz. Detrending is performed using a high-pass filter when NIRS signals slowly drift throughout the experimental session. High-pass filters usually range between 0.01 and 0.05Hz. However experimenters must care as the frequencies of interest could be part of this range. Finally, experimenters have several tools to choose from to remove movement artefacts. If possible set a marker during the experimental session when the subject moved his head is a good start. Retrospectively, the eye of the physiologist is the first tool which can be used. However, its somehow objective behaviour and its inability to treat large amounts of data make its main limits. Abrupt changes in the signals can be detected and corrected by algorithms (Lloyd-Fox et al., 2010; Wilcox et al., 2008). However, the thresholds must be defined carefully in order to preserve the changes that supposedly belong to the awaited hemodynamic response (Gervain et al., 2011).

#### **3.5 Data analysis**

Since NIRS is a relatively new technique for brain investigations, there is no standardised method to analyse data. Up to date, the only invariant is that different experimental designs require different analysis techniques.

In block-designed studies, experimenters are used to analysing time series by averaging multiple trials of the same condition. Mean variations and mean time courses are then obtained for each condition. The critical points of such techniques are the determination of the relevant windows of the time series and the baseline which it is compared to. Once determined, student t-test and analyses of variance are the most often used statistical methods.

More complex, three main freeware packages are downloadable and provide analysis methods derived from the BOLD signal of fMRI: HomER (Huppert et al., 2009), fOSA (Koh et al., 2007) and NIRS-SPM (Ye et al., 2009). The general linear model (GLM) and the statistical parametric mapping (SPM) offer the possibility to create three dimensional pictures of the brain, where activated/inhibited cortex areas are colour encoded (Friston et al., 1999; Plichta et al., 2007; Schroeter et al., 2004; Zarahn et al., 1997). In most studies, the NIRS records are performed off the MRI scan. Then, the input of the three dimensional coordinates of the optodes/channels is crucial for the reconstitution of the pictures. In the case of a co-record of NIRS and fMRI techniques, the coordinates of the NIRS optodes can be precisely assigned; else, skull measurements and probe placement are made either by reference to the 10-20 EEG system or by kinematic acquisition using such devices as optotrack or fastrack.

#### **3.6 Dos and don'ts**

Doubtlessly, the toughest part of the NIRS based studies, is to draw physiological and cognitive conclusions from the data. Multi-channel setups cover wide cortical zones and result in several time series and three dimensional coloured images in which probability to give statistically significant results is high. The question experimenters inevitably face is "What do those results mean?". A typical NIRS channel includes a great number of capillary beds, corresponding to a greater number of neurons (estimated around 300,000 to 500,000) from various depths in the cortex (Gervain et al., 2011). The pool of capillary beds enlighten by a channel is believed to belong to a given cortex area, which supposedly has a single function. This makes a huge simplification if compared to the brain complexity and its capacity of integration, not to mention the neuro-vascular coupling assumption (see part 4.1.)! Moreover, probe placement is based on the skull anatomy as no direct access to the brain is allowed by NIRS (except in the case of fMRI co-recording) giving a probability to fire over multiple cortex areas or even over a wrong area. Additionally, the proportion of excitatory and inhibitory neurons in the volume aimed by NIRS is unknown yet potentially affects the results.
