Functional Optical Brain Imaging

本文是从《Biomedical Signals, Imaging, andInformatics》摘录，本章节作者 Meltem Izzetoglu，Drexel University。在此感谢作者及相关人的辛勤付出。

6.1 Introduction

Functional imaging is typically conducted in an effort to understand the activity in a given brain region in terms of its relationship to a particular behavioral state, or its interactions with inputs from another region’s activity. The advances in noninvasive functional brain monitoring technologies provide opportunities to accurately examine the living brains of large groups of subjects over long periods of time, with little impact on their well-being. Neurophysiological and neuroimaging technologies have contributed much to our understanding of normative brain function, as well as to our understanding of the neural underpinnings of various neurological and psychiatric disorders. Commonly employed techniques such as electroencephalography (EEG), event-related brain potentials (ERPs), magnetoencephalography (MEG), positron emission tomography (PET), single-positron emission computed tomography (SPECT), and functional magnetic resonance imaging (fMRI), have dramatically increased our understanding of a broad range of brain disorders. Nevertheless, there is still much unknown about these syndromes. This is due, in large part, to the inherent complexity of the neurobiological substrates of these disorders and of the mind itself. In addition, each of the research methods used to study brain function and its disorders have methodological strengths as well as their own inherent limitations. These limitations place constraints on our ability to fully explicate the neural basis of neurological and psychiatric disorders both inside and outside of the laboratory setting, and to use the information gleaned from laboratory studies for clinical applications in real-world environments. New techniques that allow data to be gathered under more diverse circumstances than is possible with extant neuroimaging systems should facilitate a more thorough understanding of brain function and its pathologies.

Functional near-infrared spectroscopy (fNIR) has been introduced as a new neuroimaging modality with which to conduct functional brain imaging studies [1–24]. fNIR technology uses specific wavelengths of light, irradiated through the scalp, to enable the noninvasive measurement of changes in the relative ratios of deoxygenated hemoglobin (deoxy-Hb) and oxygenated hemoglobin (oxy-Hb) during brain activity. This technology allows the design of portable, safe, affordable, noninvasive, and minimally intrusive monitoring systems. These qualities make fNIR suitable for the study of hemodynamic changes due to cognitive and emotional brain activity under many working and educational conditions, as well as in the field. Using this technique, several types of brain activity have been assessed, including motor activity, visual activation, auditory stimulation, and the performance of cognitive tasks (e.g., [9]). To date, the outcomes of studies utilizing fNIR compare favorably with previous fMRI and complement EEG findings [21–23].

The purpose of the present chapter is to describe an emerging neuroimaging technology, fNIR, which has several attributes that make it possible to conduct neuroimaging studies of the cortex in clinical offices and under more realistic, ecologically valid parameters. In this chapter, we will first describe general working principles of fNIR and different fNIR instrumentations based on these working principles. Then, we will review algorithms developed for fNIR data processing for artifact removal and used in the extraction of hemodynamic signals from the fNIR intensity measurements. We will discuss the results of various fNIR applications in laboratory settings and in field conditions. Finally, the merits of optical imaging in brain research will be summarized.

6.2 Working Principles

6.2.1 Physiological Principles: How Can Brain Activity Be Measured through Hemodynamic Changes?

Neural activity has a direct relation with hemodynamic changes in the brain [25]. Research on brainenergy metabolism has elucidated the close link between hemodynamic and neural activity [26,27]. Understanding the brain energy metabolism and associated neural activity is of importance for realizing principles of fNIR in assessing brain activity. The brain has small energy reserves and a great majority of the energy used by brain cells is for processes that sustain physiological functioning [28]. Ames III reviewed the studies on brain energy metabolism as related to function and reported that the oxygen (O2) consumption of the rabbit vagus nerve increased 3.4-fold when it was stimulated at 10 Hz and O2 consumption in rabbit sympathetic ganglia increased 40% with stimulation at 15 Hz. Furthermore, glucose utilization by various brain regions increased several fold in response to physiological stimulation or in response to pharmacological agents that affect physiological activity [28]. These studies provide clear evidence that large changes occur in brain energy metabolism in response to changes in activity. The levels of compounds involved in energy metabolism and energy metabolites can be outlined as

• Neuronal activity is fueled by glucose metabolism, so increases in neural activity result in increased glucose and oxygen consumption from the local capillary bed. Brain cells consume energy when activated and oxygen is required to metabolize the glucose. The oxygen concentration in the capillaries can support the normal oxygen consumption (about 3.5 μmol g−1 min−1) for 2 s. [27]. For that reason, a reduction in local glucose and oxygen stimulates the brain to increase local arteriolar vasodilation, which increases local cerebral blood flow (CBF) and cerebral blood volume (CBV), a mechanism known as neurovascular coupling.

• Over a period of several seconds, the increased CBF carries both glucose and oxygen to the neural tissue in the area. Oxygen is transported via oxy-Hb in the blood. The oxygen exchange occurs in the capillary beds. As oxy-Hb gives up oxygen, it is transformed into deoxy-Hb.

• The increased oxygen transported to the area via increased CBF typically exceeds the local neuronal rate of oxygen utilization, the cerebral metabolic rate of oxygen (CMRO2), resulting in an overabundance of cerebral blood oxygenation in active areas [29]. Therefore, local blood is more oxygenated and hence less deoxy-Hb is present [30]. (Although the initial increase in neural activity is thought to result in a focal increase in deoxygenated hemoglobin in the capillary bed as oxygen is withdrawn from the hemoglobin for use in the metabolization of glucose, this feature of the vascular response has been much more difficult to measure, and more controversial, than hyperoxygenation. Please see [31,32], for a more detailed discussion of this topic.) These changes in the hemodynamic signals occur within a few seconds after the onset of the stimulation and may take 10–20 s to evolve [33,34]. Based on the brain energy metabolism, methods and imaging modalities, such as fNIR and fMRI [35,36] for measurements of slowly changing hemodynamic signals, deoxy-Hb and/or oxy-Hb are implemented to provide correlates of brain activity through oxygen consumption by neurons.

6.2.2 Physical Principles: How Can Hemodynamic Activity Be Measured by Optics?

It is well-known that the functional state of tissue can influence its optical properties, for instance, cyanosis in hypoxia; pallor in anemia. The human brain undergoes a number of physiological changes as it responds to environmental stimuli; these changes in blood levels and electrochemical activity also affect its optical properties. Functional optical imaging capitalizes on the changing optical properties of these tissues by using light in the near-infrared range (700–900 nm) to measure physiological changes. Because oxy-Hb and deoxy-Hb have characteristic optical properties in the visible and near-infrared light range between 700 and 900 nm, the change in concentration of these molecules during neurovascular coupling can be measured using optical methods [9,37]. Most biological tissues are relatively transparent to light in the near-infrared range, largely because major components of most tissues, such as water absorb very little energy at these wavelengths (see Figure 6.1). As such, this spectral band is often referred to as the “optical window” for the noninvasive assessment of brain activation. However,

FIGURE 6.1 Absorption spectrum in near-infrared (IR) window.

the chromophores oxy-Hb and deoxy-Hb do absorb a fair amount of energy in this range. Fortunately,in the optical window, the absorption spectra of oxy-Hb and deoxy-Hb remain significantly different than each other as shown in Figure 6.1 allowing spectroscopic separation of these compounds to be possible using only a few sample wavelengths.

Typically, an fNIR apparatus is comprised of a light source that is coupled to the participant’s head via either light-emitting diodes (LEDs) or though fiber-optical bundles (the optode), and a light detector that receives the light after it has interacted with the tissue. Photons that enter tissue undergo two different types of interaction, namely absorption and scattering [4,5,19]. When photons get absorbed by the tissue they lose their energy to the medium and hence cannot continue to travel within the tissue. After entering the tissue, photons get diffused and experience multiple scattering events. Hence, a photodetector placed 2–7 cm away from the light source can collect a relatively predictable quantity of photons that are not absorbed and traveled within the tissue along the “banana-shaped path” between the source and detector due to multiple scattering [9,38] as shown in Figure 6.2. When the distance between the source and photodetector is set at 4 cm, the fNIR signal becomes sensitive to hemodynamic changes within the top 2–3 mm of the cortex and extends laterally 1 cm to either side, perpendicular to the axis of source-detector spacing [39]. Studies have shown that at inter-optode distances as short as 2–2.5 cm,grey matter is part of the sample volume [39,40]. If wavelengths of the light sources are chosen to maximize the amount of absorption by oxy-Hb and deoxy-Hb, changes in the chromophore concentrations cause changes in the number of photons that are absorbed and the number of photons that are scattered back to the surface of the scalp. These changes in the attenuation of light measured at the surface of the scalp can be quantified using different techniques, that is, diffusion equation, modified Beer–Lambert law, and so on, and information on changes in oxy-Hb and deoxy-Hb concentrations can be assessed which will be discussed in Section 6.5.

6.3 Instrumentation

A wide variety of both commercial and custom-built fNIR instruments are currently in use [5]. These systems differ with respect to their use and system engineering, with tradeoffs between light sources, detectors, and instrument electronics that result in tradeoffs in the information available for analysis, safety, and cost. Three distinct types of fNIR implementation have been developed: time domain (TD) systems, frequency domain (FD) systems, and continuous wave (CW) spectroscopy systems, each with their own strengths and limitations [3–5]. In TD systems, also referred to as time-resolved spectroscopy (TRS), extremely short (picosecond-order) incident pulses of laser light are applied to the tissue and the temporal distribution of photons that carry the information about tissue scattering and absorption is measured. The emerging intensity is detected as a function of time (the temporal point spread function [TPSF]) with picosecond resolution [7,39,41]. Streak camera systems can provide high-temporal resolution, but they are large, expensive, and have a limited dynamic range. Time-correlated single-photon counting systems can be built with cheaper components and can provide wide dynamic range, however, they have poor temporal resolution. Even though TRS instruments offer absolute measurements of hemodynamic changes since they can be large, expensive, with limited dynamic range or poor temporal resolution, they have originally been developed as laboratory-based devices, and hence are difficult to be implemented in the clinical environment and in the field applications [5]. In FD or phase modulation spectroscopy (PM) systems, the light source is intensity modulated to the frequencies in the order of tens to hundreds of megahertz. The amplitude decay, phase shift, and modulation depth of the detected light intensity with respect to the incident light are measured to characterize the optical properties of tissue [42]. FD methods are low-cost alternatives to time-resolved methods and hence several multichannel FD instruments are now in common use [7,43].

CW systems apply either continuous or a slow-pulsed light to tissue and are limited to measuring the amplitude attenuation of the incident light [45]. These systems utilize less sophisticated detectors than time-resolved and FD systems and they cannot resolve the time-changing component of the light. As such, CW systems provide somewhat less information than time- or FD systems meaning relative measurements as opposed to absolute ones, but this tradeoff results in the capacity to design more compact, portable, easy to engineer, and affordable hardware making it advantageous for various applications [45,46]. They can be laser-based, but LEDs can also be used in CW designs to increase safety (particularly with respect to eye exposure) and comfort, and to again decrease both instrument size and cost, making it possible to deploy these systems in clinical or educational settings [5,7,32]. A detailed list of both commercially available and laboratory prototype fNIR instruments are presented in [43].

6.4 fNIR Measurements

6.4.1 Fast Neuronal Signal

Fast neuronal signal or event-related optical signal (EROS) capitalizes on the changes in the optical properties of the cell membranes themselves that occur as a function of the ionic fluxes during firing [14]. Using invasive techniques, it has been well-established that the optical properties of cell membranes change in the depolarized state relative to the resting state [32,47–49] and that optical methods can be used to detect these changes that occur within the first 200 ms of functional stimulation. The ability to measure the actual depolarization state of neuronal tissue provides obvious advantages in that it is a direct measure of neural activity, with millisecond-level time resolution as can be measured with EEG and MEG but with the superior spatial resolution lacking in EEG/MEG. However, there are also a number of limitations to the noninvasive use of the EROS signal in humans. A primary disadvantage of the fast optical signal is the low signal- to-noise ratio (SNR) resulting from the need to image through skin, skull, and cerebral-spinal fluid. Basic sensory and motor movements such as tactile stimulation and finger tapping require between 500 and 1000 trials to establish a reliable signal [13]. There have also been failures to replicate the results of experiments reporting the fast optical signal in response to a visual stimulus among normal adult humans [32]. The low SNR may play a role in current difficulties with experimental replication; however, more cross-validation work is warranted. The final constraint is that these methods require a more expensive and cumbersome laser-based light source (vs. an LED-based light source), they are not portable, and the potential risk of inadvertent damage to the eyes is increased relative to the systems available for measuring hemodynamic responses. (LED-based nearinfrared sources pose very little, if any, risk upon eye exposure [50]). In spite of these current limitations, the fast optical signal continues to be an important area of investigation because it offers glimpses of the “holy grail” of neuroimaging: the direct measurement of neuronal activity with millisecond time resolution and superior spatial resolution.

6.4.2 Slow Hemodynamic Signal

As explained in Section 6.2.1 through brain-energy metabolism, during brain activity first local oxygen consumption increases to metabolize glucose and provide energy to the activated neurons. Next, blood flow and perfusion increases in the area of activated neurons through neurovascular coupling which in turn changes the hemoglobin concentrations and oxygenation. These slow changes in hemodynamic signal occur within a few seconds after the onset of the stimulation and takes 10–20 min to take its full course to evolve [33,34] and can be measured by fNIR, PET, and fMRI as blood oxygen level dependent (BOLD) signal that is related with the changes in the concentration of deoxy-Hb.

Most fNIR applications are based upon the measurement of the slow hemodynamic responses in terms of deoxy-Hb and oxy-Hb. Typical slow hemodynamic signals together with corresponding fMRI– BOLD signal collected during a finger tapping experiment is shown in Figure 6.3a. Either the full-time course of the hemodynamic signals and/or features that can be extracted from them such as maximum amplitude, time to peak or reaction time, full width half maximum, and so on, as shown in Figure 6.3b are then used for comparison purposes in different task conditions, spatial locations or subject populations which will be discussed in detail in Sections 6.5 and 6.6.

6.5 fNIR Signal Processing

Analysis of fNIR measurements involves two main steps: (i) artifact and/or noise removal, and (ii) conversion of optical measurements into physiologically relevant signals and/or features. fNIR measurements can get corrupted by different sources of noise that should be suppressed in order to reliably extract information-bearing signals that are related to cognitive activity. Depending on the fNIR system used optical intensity measurements are converted to physiologically relevant signals and features for further processing using different approximations, assumptions, and algorithms. In this p, we will review different sources and characteristics of noise and existing algorithms to suppress them and the algorithms used to extract hemodynamic variables from fNIR intensity measurements.

6.5.1 Signal Separation and Noise Removal

The sources of noise in fNIR measurements can be categorized in two main groups: (i) noise of nonphysiological basis, and (ii) artifacts arise from systemic physiological origins. Noise sources that are not driven by physiological origins can happen as a result of instrumental noise, that is, electrical noise or experimental errors, that is, motion artifact, task-related errors. Since fNIR systems measure changes in hemodynamics, they measure such changes not only related with cognitive activity but also the ones that occur due to heart pulsation, respiration, and so on, resulting in systemic physiological artifacts. A detailed explanation of these noise sources and some algorithms for their suppression is presented in [52]. In this p, we will briefly explain these different sources of noise and summarize currently existing algorithms for their suppression.

Note that, noise in the optical measurements can be removed before and after the conversion of raw intensity measurements to physiological hemoglobin concentrations. The noise of nonphysiological basis such as instrument noise or experimental errors that are not related to underlying physiological changes should be removed prior to conversion of signals into hemodynamic measures. This way, errors across wavelengths will not be propagated causing additional crosstalk in the separation of oxy-Hb and deoxy-Hb. In contrast, noise of physiological origins such as blood pressure changes, heart pulsations, and so on, causes oscillations in hemoglobin concentrations together with changes related to brain activity such biological noise is better removed after the conversion of raw intensity measurements to hemodynamic variables [52].

6.5.1.1 Noise of Nonphysiological Basis

One type of noise that is not related with any physiological changes is electrical/electronic noise from the computer or other hardware in the instrument. Such electrical noise can also be generated by the use of other devices or from the surrounding space that can create interference in the measurements. This type of electrical noise is generally assumed to be white and since hemodynamic response generated by brain activity is generally slow (up to 0.1 Hz), most of the time, high frequency instrument noise are suppressed by simply low-pass filtering the data [52,53] or significantly reduced by adjusting the power of light sources within the safety limits and gain amplification at the detectors.

There can be other types of noise within fNIR measurements whose origins are not directly related with physiological changes within the underlying tissue such as artifacts that may arise as the result of subject motion or noncompliance with the experimental paradigm or the ones that are the direct results of experimental paradigm itself. This type of noise is sometimes called experimental errors [52]. Head movement can cause the fNIR detectors to shift and lose contact with the skin, exposing them to either ambient light or to light emitted directly from the fNIR sources or reflected from the skin, rather than being reflected from tissue in regions of interest. These effects cause sudden increases in the fNIR data. Another consequence of head movement is, it can cause the blood to move toward (or away from) the area that is being monitored, increasing (or decreasing) the amount of oxygen, hence result in an increase (or decrease) in the measured data. Since the dynamics of this type of motion artifact are slow, they can easily be confused with the actual hemodynamic response due to brain activation. Even though it is not as pronounced as in fMRI, motion artifacts in fNIR studies are a serious problem for real-life applications where head immobility is undesirable or untenable such as in studies involving pilots, children, and so on. Hence, cleaning the fNIR data from motion artifacts is an important and necessary task in order to deploy fNIR as a brain monitoring technology in its full potential to many real-life application areas.

The best way to deal with motion artifacts is to avoid them as much as possible which mostly depend on subject compliance, experimental design, and experimenter’s expertise in the placement of the fNIR sensor. However, these conditions can be satisfied for experiments involving adults and the ones that can be carried out in the laboratory, in controlled environments or in sitting positions. Different sensor designs and placements are developed to match the requirements of the experiment and reduce dependence on experimenter’s expertise for its secure placement [52]. However, such sensor designs may only eliminate motion artifacts that can arise from the shift or pop of sources and/or detectors where motion artifacts of slow nature due to blood movement may still be present if head movement occurs during data acquisition as a result of experimental design or subject incompliance. There are several different algorithms developed for the detection and elimination of motion artifacts. For motion artifacts arising from the physical displacement or movement of the optical probe, detection algorithms such as the one based on outlier detection methods was proposed in [52]. In order to suppress the motion artifact, several methods were proposed such as the ones based on principal component analysis (PCA) [52], wavelet analysis [54] or recently a combined PCA and independent component analysis (ICA) [55]. The novelty in this last method was the use of dark current measurements for the identification of the existence of motion artifacts and their removal by the use of a combined ICA/PCA-based method. This method was applied on a clinical data collected in a highly noisy environment namely operating room and shown to successfully remove motion artifacts. A comparison of algorithms in motion artifact removal based on wavelets, autoregression, adaptive filtering, and ICA are discussed in [56]. For motion artifacts that are the result of the movement of the blood within the tissue that are more slow in nature and have similar frequency spectrum as with the optical signals coming from physiological origins such as hemodynamic response to stimulus, respiration, heart pulsation, and so on, there exists different algorithms based on adaptive, Wiener and Kalman filtering. A detailed analysis of these algorithms in comparison to each other based on their ability to improve SNR on 11 subject data set where slow, medium, and high-speed head movement was tested are given in [57,58]. A sample result is shown in Figure 6.4 where motion free data, adaptive, Wiener and Kalman filtered data are presented on a medium speed head movement case. A separate study for similar type of motion artifact used the observation that motion noise may cause the measured oxy-Hb and deoxy-Hb signals, which are typically strongly negatively correlated, to become more

positively correlated [59]. A method has been developed to reduce noise-based on the principle that the concentration changes of oxy-Hb and deoxy-Hb should be negatively correlated.

In addition to motion artifacts, experimental errors can also be introduced simply by the design of the functional task or the study itself of it being an event-related or block design or a longitudinal study, and so on. The predominant paradigms used in functional studies are the repeated single trial paradigm and the blocked trial paradigm (Figure 6.5) [60]. Usually in fNIR studies experimental designs are selected similar to the ones used in fMRI or EEG studies. However, additional care must be taken in adjusting the task or selecting data analysis methods while using fNIR. For example, since hemodynamic responses are slow in nature as compared to the fast neuronal signals as measured by EEG, improper design of the timing of an experimental paradigm can introduce errors. In general, this requires that the timing of stimulus presentation in single trial paradigms to be spaced widely apart in time to allow the evoked-hemodynamic signal to evolve completely. If a block design or a rapid event-related design is used, selection of the interstimulus interval is very important for the evoked- hemodynamic responses to still hold the linearity assumption (the assumption that repeated evoked responses will add linearly). For event-related stimulus designs or in block designs where evoked- hemodynamic responses are required to be extracted [61], the total hemodynamic responses measured may be approximated as the linear summation of evoked responses to each single stimulus, provided that the interstimulus interval is longer than around 3–4 s. Another study related error in the measurements can occur during longitudinal studies over a subject group. If the fNIR sensor is not placed on the exact same location at different recording times, the results may not reflect possible changes over time but the optical differences in different tissue area monitored [52].

6.5.1.2 Artifacts That Arise from Systemic Physiological Origins

fNIR technology measures hemodynamic changes which can be as a result of brain activity or simply due to systemic physiological variations such as cardiac pulsations, respiration, and blood pressure. In humans, the cardiac pulsation and respiration typically have a period of 0.7–1.5 s and 3–8 s, respectively. The arterial blood pressure varies on multiple time scales approximately 10 s Mayer waves and slower N50-s variation [51,52,62,63]. Note that, although often times these physiological fluctuations are considered to be artifacts because they may interfere with the evoked-functional responses, if they can be effectively and reliably separated from evoked-hemodynamic responses, it may in fact be a very valuable characteristic of fNIR to be able measure all these physiological signals with just one sensor.

In order to improve the sensitivity and interpretability of fNIR measurements to brain activation, it is necessary to separate systemic physiological signal sources spatially and temporally. There are different techniques developed to suppress or separate the systemic physiological signals from evoked- hemodynamic responses based on either (i) known approximate frequency content of signals, (ii) the spatial covariance of

physiology, or (iii) the ability to remotely measure physiological signals correlated with the ones appearing in fNIR measurements of interest. Since the frequency bands of the physiological signal sources are known approximately, classical filters, that is, low-pass, band-pass, high-pass filters can effectively and efficiently be used to separate some of those signals. A classic example of the power spectral density (PSD) spectrum of the poststimulus signal segment is shown in Figure 6.6 [60]. The hemodynamic response following neural activation is embedded in the frequencies up to approximately 0.1 Hz. The 0.8 Hz and 0.2 Hz bands correspond to heart rate and respiration, respectively. The B-waves (~0.03 Hz) represent very low frequency oscillations (VLFO) generated by the various brain stem nuclei in the vasomotor tone of cerebral arterioles [64]. The Mayer waves (~0.1 Hz) represent low frequency oscillations reflecting oscillations due to blood pressure. As an example, if a band-pass filter with pass band selected to be between 0.08 and 0.12 Hz is designed, it can remove the high-frequency instrument noise, the fast cardiac oscillations, respiration, and VLFO. Note that, in such application involving band-pass filtering care should be taken to avoid removal of frequencies associated with the evoked-functional hemodynamic response. For example, bandpass filtering can generally not be used to remove Mayer waves, since their frequency band overlaps with the content of the hemodynamic response. Alternatives to conventional bandpass filtering, involve the use of curve fitting [60], adaptive, Wiener and Kalman filtering techniques [60,65,66]. In the Wiener filtering approach, the statistical changes in such physiological artifacts had been taken into account.

Other than classical filters, algorithms that use the spatial persistence nature of systemic physiological signals (specific and repeated spatial pattern across the brain) such as PCA-based physiological filtering methods have been proposed [67] similar to its use for motion correction. There also exist algorithms based on separate measurements providing correlated noise sources with the physiological signals. In these algorithms the cardiac, respiration, and blood pressure signals are separately measured by additional sensors such as with dynamic blood pressure cuff, pneumatic belt, pulse plethysmograph, pulse oximetry, or an electrocardiogram which can provide signals that are correlated with the artifacts that exist within fNIR measurements. Then, algorithm such as least squares regression algorithms [68], adaptive filtering [66], ICA [69] or RETROICOR [70] as developed for fMRI can be used to remove these signals from fNIR measurements to extract evoked-hemodynamic responses. In another technique, correlated measurements of physiological artifacts are measured by using multiple source detector separations. Since penetration depth increases with an increase in source detector separation, by measuring from detectors that are placed close enough to the source, one can guarantee penetration to only superficial tissues and not to brain. On the contrary, detectors placed farther away from the source can measure signals coming from the brain together with all the other superficial layers. Such arrangement is used in different applications to remove physiological signals [66] as well as motion artifacts [71].

6.5.2 Hemodynamic Signal Extraction from fNIR Intensity Measurements

Several different algorithms based on different mathematical models and assumptions have been developed to quantify the fNIR measurements and obtain information about the physiological/optical changes related with brain activity. Depending on the fNIR system and measurement type, appropriate algorithms can be implemented. Here, we will discuss two of the most commonly used methods (i) diffusion theory and (ii) modified Beer–Lambert law both of which are derived from linear transport theory to extract absolute or relative measurements of hemodynamic signals. Within each category, we will describe the mathematical model used, assumptions needed to meet for the algorithm to be applicable, and existing analytical and experimental solutions.

6.5.2.1 Diffusion Theory

Propagation of light in highly diffusive media such as tissue is usually modeled by using linear transport theory (radiative transport or Boltzman transport theory) [7,72–74]. However, analytical solutions to radiative transport equation can only be obtained for very simple geometries. Much simpler approximation can be derived from radiative transport equation based on diffusion theory [7,72–77] under certain assumptions such as (i) scattering coefficient is much greater than the absorption coefficient, (ii) scattering is isotropic, (iii) source detector separations are not too small, and (iv) measurement times are apart from the input times [72,77]. TD diffusion equation is given as [7,72–74]

where Φ(r,t) is the diffuse photon fluence rate at position r and time t, S(r,t) is the photon source, c is the speed of light in the tissue, D is the diffusion coefficient, μa is the linear absorption coefficient, μs is the linear scattering coefficient, and g is the mean cosine of the scattering angle.

Analytical solutions to diffusion equation as a forward problem (derivation of the resulting measurement given the distribution of light sources and detectors, tissue parameters, and boundaries) can be obtained using Green’s function which provides the photon distribution when the source is an impulsive function. Green’s function solutions to diffusion approximation have been driven for slab, cylindrical, and spherical geometries. For a semi-infinite half-space geometry one such solution for an impulse input is found as [7,78]

where R(ρ,t) is the reflected light intensity, zo = 1/μ′s and μ′s = (1 − g)μs is the reduced scattering coefficient. Solutions to diffusion equation in TD and FD can also be found in [77].

Experimentally, as an inverse problem (derivation of the distribution of tissue optical parameters given the distribution of light sources and measurements [72]), in TD systems the time course of the measured light intensities at different wavelengths are fitted to the solution equation and hence the absolute absorption and scattering parameters of the underlying tissue of interest are estimated [7]. Note that, since information contained in temporal profile of the measured light will also be present in FD, similar approaches to TD techniques can be applied to FD systems to extract optical parameters as an inverse problem.

Optical properties of the tissue such as absorption and scattering coefficients are wavelength dependent. Furthermore, absorption parameter of the tissue changes with the change in the concentrations of underlying tissue chromophores. Since within the optical window (700–900 nm) the dominant tissue chromophores are the oxy-Hb and deoxy-Hb, their absolute concentrations can be calculated using the absorption coefficient estimated at multiple wavelengths [51] using:

where λ is the photon wavelength, ε is the wavelength and chromophore-dependent extinction coefficients (experimental values are tabulated in [19]) and CHBO2 and CHB are the concentrations of oxy-Hb and deoxy-Hb, respectively.

Solutions to diffusion equation cannot be obtained analytically for more complex inhomogeneous geometries. For such cases, numerical solutions are studied based on numerical integral and differential approximation methods such as finite difference methods (FDM), finite element methods (FEM), or perturbative approaches using truncated series approximations such as Born and Rytov approximations and their higher order extensions [72,73]. These numerical solutions suffer from extensive computational cost. Another numerical method used for the solution of the forward problem to study the light propagation in tissue is the Monte Carlo modeling (MCM) [73]. In MCM, photons are assumed to have certain probability of scattering and absorption where many photons are injected into the medium at every source, treated separately and at the end statistical maps of a combination of photons collected at the detectors are obtained.

6.5.2.2 Modified Beer–Lambert Law

Under various simplifying assumptions [79], radiative transport equation can be reduced to the Beer–Lambert Law which states that the attenuation of an absorbing compound dissolved in a nonabsorbent compound is proportional to the concentration of the compound and the optical path length [2,4]. In a purely absorbing medium, the transmitted light I can be expressed in terms of the input light Io as follows:

where μa is the absorption coefficient of the homogeneous medium and L is the optical path length. Since absorption coefficient is related to concentration of the absorbing compound C and the extinction coefficient ε as μa = εC, the absorbance or attenuation (A) in terms of optical density (OD) units can be written as

If there is more than one absorbing compound within the medium then the attenuation can be given by

In a scattering medium where there is no absorption, the light will still be attenuated since the photons will be deflected from their path [80]. Hence similar to absorbing case, in a medium having only one scattering compound with a scattering coefficient μs the transmitted light intensity I can be written as

Note that when multiple scattering occurs and the light gets diffuse, then isotropic scattering is often assumed and the reduced scattering coefficient μ′s = (1 − g)μs is used instead to represent it.

Within the tissue, light attenuation happens due to absorption and scattering together. Both of these effects should be taken into account in order to correctly quantify changes in hemodynamics. Modified Beer–Lambert law (MBLL) accounts for the effects of both the absorption and multiple scattering where the attenuation is expressed as follows:

where G represents the light attenuation due to scattering which depends on also on measurement geometry. MBLL relies on several assumptions such as (i) homogeneous tissue, (ii) constant scattering G, and (iii) known path length L [4,74,80]. Even though these may not be exactly met by the biological tissues, MBLL is by far the most widely used technique in calculating hemodynamic changes especially in CW systems mostly due to its simplicity and no computational cost [74]. It is discussed in [80] that scattering in brain tissue occurs mostly due to cell and subcellular membranes and that scattering due to red blood cells is low. Hence, in functional optical imaging studies, the attenuation due to scattering can be assumed to be constant since even though the concentration of oxy-Hb and deoxy-Hb changes due to brain activity their contribution to scattering can be negligible.

The mean path length (L) that light travels in a highly scattering medium such as tissue is larger than the direct path or geometrical path length (d) between light source and detector. This increase in path length due to tissue scattering is corrected by using a differential path length factor (DPF). Then the mean path length is found as

The DPF for tissues at different anatomical regions such as head (forehead, somatosensory, and occipital regions), calf, fore arm, and so on, for adult and infant populations have been both experimentally (using TD and FD systems) and numerically (using Monte Carlo simulations) studied and the values are tabulated [81–83]. It is noted that DPF is wavelength dependent and there can be subject to subject differences based on gender, age, and so on [83,84].

Even though the DPF can be measured or taken from published values assuming small across subject differences, the absolute quantification of chromophore concentrations cannot be possible since G is unknown. If G is assumed to be constant during the measurement period, then by measuring the A at two or more wavelengths, the relative change in tissue chromophores ΔC such as oxy-Hb and deoxy-Hb versus time can be obtained from relative changes in attenuation ΔA to an arbitrary time as follows:

By measuring attenuation change at two distinct wavelengths and using Equation 6.10, changes in oxy-Hb (HbO2) and deoxy-Hb (Hb) can be calculated as follows:

The MBLL has been widely used in CW systems for the quantification of hemodynamic changes. However, care should be taken in the selection of the wavelengths of the light sources implemented in order to correctly extract relative changes in tissue chromophores while reducing crosstalk between them [51].

6.6 fNIR Studies

Since the first in vivo application of fNIR on cat brain by Jobsis 1, there has been a growing interest in this exciting noninvasive, safe, affordable and portable technology and hence many different applications involving healthy and diseased groups in adult and children populations for laboratory and field conditions had been studied. Here, we will summarize some clinical applications and further give some specific examples of some of the basic research and clinical application studies that have been mainly carried out by the Optical Brain Imaging group at Drexel University within the Cognitive Neuroengineering and Quantitative Experimental Research (CONQUER) Collaborative.

6.6.1 Summary of Basic Research and Clinical Application Studies

In the past decade, brain activation studies employing fNIR have been conducted on the visual system [8], the somatosensory system [10], the auditory system [17,16], and the language system [6], and during motor tasks [5]. fNIR studies have also been used to examine a number of cognitive tasks [6,21–23,53]. In general, these studies have reported localized increases in oxy-Hb in response to functional challenge, and the results have largely been in agreement with corollary fMRI studies.

The first clinical applications of fNIR have been in the investigation of fetal, neonatal, and infant cerebral oxygenation and functional activation. For instance, fNIR studies have revealed developmental alterations in the cerebral hemodynamic response to auditory and visual stimulation [17,85]. Neurological applications have included an evaluation of the hemodynamic response during deep brain stimulation in Parkinson’s patients [86], brain activations during induced seizures in patients with intractable epilepsy [87], an exploration of the pathophysiology of seizures in childhood epilepsy [88], and an examination of Alzheimer’s patients during verbal fluency and other cognitive tasks [89]. Psychiatric applications have included the comparison of prefrontal brain activations of schizophrenic patients to healthy subjects during a mirror drawing task [90], a self-face recognition test [24], and during a continuous performance task [91]. fNIR was used to predict treatment response in a study of the effects of transcranial magnetic stimulation on depression [92]. A detailed review on clinical application can be found in [93].

6.6.2 Specific Example Studies

6.6.2.1 fNIR Device

In all of the specific example studies that will be discussed in this p a portable CW–fNIR system developed by Drexel University, optical brain imaging team which was originally described by Chance et al. [46] is used. The main components of the system were: (1) the sensor that covers the entire forehead, (2) a control box with integrated power supply for sensor control and data acquisition (current sampling rate is 2 Hz), and (3) a computer for the data analysis software [49,53]. The flexible sensor consisted of four light sources with three built-in LEDs sources having peak wavelengths at 730, 805, 850 nm and 10 photodetectors designed to image cortical areas underlying the forehead (dorsolateral and inferior frontal cortices). With a fixed source-detector separation of 2.5 cm and the implemented data collection scheme, this configuration resulted in a total of 16 signal channels (voxels) as shown in Figure 6.7. Communication between the data analysis computer and the task presentation computer is established via a serial port connection to time-lock fNIR measurement to the task events. The flexible sensor design consists of three parts: (i) a reusable flexible circuit board that carries the necessary infrared sources and detectors, (ii) a replaceable cushioning material, and (iii) a disposable single-use medical-grade adhesive tape that serves to attach the sensor to the participant (see Figure 6.7). The flexible circuit provides a reliable, integrated wiring solution, as well as consistent and reproducible component spacing and alignment. Because the circuit board and cushioning materials are flexible, the components move and adapt to the various contours of the participant’s head, allowing the sensor elements to maintain an orthogonal orientation to the skin surface, dramatically improving light coupling efficiency and signal strength.

6.6.2.2 Basic Research Studies Targeting Different Cognitive Domains

To date, the fNIR studies of cognition and emotion have focused on functions associated with Brodman’s areas BA9, BA10, BA46, BA45, BA47, and BA44. Recent PET and fMRI studies have shown that these areas play a critical role in sustained attention, both the short term storage and the executive process components of working memory, episodic memory, problem solving, response inhibition, and the perception of smell [for a recent review, see 94,95]. In addition, word recognition and the storage of verbal materials activate Broca’s area and left hemisphere supplementary and premotor areas [94,96,97]. Some examples of fNIR applications and their results targeting different cognitive domains will be summarized in this p.

6.6.2.2.1 Attention

The protocol used in this study [49,53] to measure attention is a common visual oddball paradigm modified for use with fMRI by McCarthy et al. [98]. The stimuli were two strings of white letters (XXXXX and OOOOO) presented against the center of a dark background. A total of 516 stimuli were presented, 480 context stimuli (OOOOO) and 36 targets (XXXXX). Stimulus duration was 500 ms, with an interstimulus interval of 1500 ms. Target stimuli were presented randomly with respect to context stimuli with a minimum of 12 context stimuli between successive targets to allow the hemodynamic response an opportunity to return to baseline between target presentations. Fifteen right-handed participants (4 females and 11 males with age 20.8 ± 4.2) were required to press one of two buttons on a response pad after each stimulus, while both fNIR and EEG were recorded simultaneously. One button was pressed in response to targets (X’s), and another button was pressed in response to context stimuli (O’s). The results for the ERPs were consistent with the literature [99]; targets elicited a pronounced P3 component with an average peak at 365 ms for both electrodes Cz and Pz (see Figure 6.8a for Pz results). The peak amplitude response to target stimuli was larger than the response to context stimuli at both Cz (t(14) = 7.58; p < 0.001) and Pz (t(14) = 7.81; p < 0.001). These ERP results confirm that the task parameters and participant responses were comparable to other ERP studies. Repeated-measures ANOVA computed on the fNIR oxygenation data (Oxy = oxy-Hb − deoxy-Hb) revealed that oxygenation values were

greater in response to targets than to controls in voxel 11, located over middle frontal gyrus of the right. Differentiation occurred between 6 and 9 s poststimulus as shown in Figure 6.8b. These results are consistent with the fMRI literature for visual target categorization with respect to increased oxygenation in response to targets, cortical location, and time course [98,100]. This study also demonstrated the utility of the combined EEG–fNIR system for studies of event-related designs that tap into ubiquitous cognitive functions such as attention [49].

6.6.2.2.2 Working Memory

In order to assess the working memory, the n-back task which is a sequential letter task with varied workload conditions that has frequently been used in working memory studies by cognitive psychologists and neuroscientists is used [95,96]. The stimuli are single consonants presented centrally, in pseudorandom sequences, on a computer monitor. Stimulus duration is 500 ms, with a 2500 ms interstimulus interval. Four conditions were used to incrementally vary working memory load from zero to three items. In the 0-back condition, subjects respond to a single prespecified target letter (e.g., “X”) with their dominant hand (pressing a button to identify the stimulus). In the 1-back condition, the target is defined as any letter identical to the one immediately preceding it (i.e., one trial back). In the 2-back and 3-back conditions, the targets were defined as any letter that was identical to the one presented two or three trials back, respectively. Subjects pressed one button for targets (approximately 33% of trials) and another for nontargets. This strategy incrementally increased working memory load from the 0-back to the 3-back condition. Each n-back block contained 20 letters whether target

or nontarget and lasted for 60 s with 15 s of rest periods between n-back blocks. Total test included 7 trials of each of the 4 n-back conditions (hence total of 28 n-back blocks) ordered in such a way that within one trial all 4 of the n-back conditions are presented; however, their order is changed randomly from trial to trial.

In the analysis performed on 9 subjects (with age between 18 and 25), statistically significant differences between the n-back conditions in fNIR measurements are obtained on the fourth voxel located on left dorsolateral prefrontal cortex (DLPFC) [53] which is in agreement with the fMRI literature [97]. Statistical analysis revealed that the 0-back condition differed from 1- and 2-back conditions; 1-back > 0-back, t = 3.21, p = .012; 2-back > 0-back, t = 2.58, p = .032. After outlier elimination, 1- and 2-back differed from each other; 2-back > 1-back, t = 2.77, p = 0.0275. No difference was found between 2- and 3-back conditions. A positive relationship between increasing workload and the oxygenation is observed in DLPFC as shown in Figure 6.9, again in agreement with fMRI studies [97]. The drop in the oxygenation values in the most difficult condition (3-back) can be interpreted using a hypothesis similar to the human performance study discussed in Section 6.6.3.1 where subjects get overwhelmed and lost their concentration.

6.6.2.2.3 Learning and Memory

This study focused on the physiological effects of repetition on learning and working memory using an adaptation of Luria’s Memory Word-Task (LMWT) [101]. The hemodynamic response in DLPFC of 13 healthy subjects (9 female), ranging in age from 23 to 43 while completing LMWT is recorded using fNIR. In LMWT, a list of 10 completely unrelated words is read aloud to the subject for he or she to memorize. This procedure is repeated 10 times. Free word recalls were acquired at the beginning, middle, and end of the task. Behavioral results showed that all subjects could recall the complete word list by the tenth trial, which was considered as successful task accomplishment. In terms of hemoglobin concentrations, it was hypothesized that there will be an increase in oxy-Hb and a decrease in deoxy-Hb in the DLPFC, reflecting activation of this area [102,103], during task learning phase, that is, when the subject is not yet able to recall the complete word list. Conversely, a significant drop in the regional-DLPFC oxy-Hb concentration, along with a significant increase in regional deoxy-Hb, is also expected, reflecting deactivation of the region of interest.

Oxy-Hb and deoxy-Hb results showed significant main effects in periods with (F[1,12] = 6.31, p = 0.027) and (F[1,12] = 20.68, p < 0.001), respectively. Post-hoc analyses showed a higher level of oxy-Hb concentration during the first period of LMWT (M = 0.044 μM) than during the second (M = −0.069 μM). Conversely, a lower level of deoxy-Hb concentration during the first period of the test (M = −0.037 μM; corrected p < 0.01) than during the second (M = 0.037 μM) was found in post-hoc analyses. Significant interactions between channels and period were also detected for oxy-Hb (F[15,180] = 2.01; p = 0.016) and deoxy-Hb (F[15,180] = 1.81; p = 0.036). Post-hoc analyses also showed that channels with significant highest differences between the first and second period of the test were left hemisphere channels 2, 3, 4 and right hemisphere channels 13, 14, 15, and 16 (all p’s < 0.05) corresponding to BA 45, 46, 9, and 10 bilaterally. These channels also showed a lower deoxy-Hb concentration during the first period of the test (all p’s < 0.05).

Correlation analyses showed significant positive correlations between oxy-Hb and memory performance in channels 1–4 and 12–16 for the first period (Spearmann’s ρ’s ranging from 0.54 to 0.73; p’s ranging from 0.03 to 0.005). In these channels, significant negative correlations were found between memory performance and deoxy-Hb (ρ’s ranging from −0.59 to −0.78; p’s ranging from 0.02 to 0.003). On the contrary, for the second period, significant negative correlations were found between oxy-Hb and memory performance in channels 1–3 and 11–15 (ρ’s ranging from −0.51 to −0.66; p’s ranging from 0.04 to 0.013). Positive correlations between deoxy-Hb and memory performance were also detected for this period in channels 1–4 and 12–16 (ρ’s ranging from 0.54 to 0.61; p’s ranging from 0.03 to 0.014).

The comparison of DLPFC hemodynamic activation pattern produced during the first period of word list learning by repetition with that produced during the second period when the list was already learned but the words were still being repeated suggested an attenuation of stimulus-evoked neural activity in prefrontal neurons. These findings indicate that the temporal integration of efficient verbal learning is mediated by a mechanism known as neural repetition suppression (NRS). This mechanism facilitates cortical deactivation in DLPFC once learning is successfully completed. This cortical reorganization is interpreted as a progressive optimization of neural responses to produce a more efficient use of neural circuits. NRS could be considered one of the natural mechanisms involved in the processes of memory learning.

6.6.2.2.4 Problem Solving

The protocol used in this study involved presentation of anagram blocks on a computer screen that contains sequences of three letter (3L), four letter (4L), and five letter (5L) anagrams starting from minimal (3L anagrams) proceeding to the maximal level of difficulty (5L anagrams), and then back down again to the starting point of 3L anagrams [61,104]. Between each anagram block session, there is a rest period of 30 s. Each anagram block is displayed for approximately 1 min containing as many anagrams within depending on the number of processed anagrams by the subject. The decision of the subjects on each anagram processed and its timing is recorded on a text file for further analysis.

Since most subjects solve the anagrams within 2–5 s and since hemodynamic response takes 10–20 s period to fully evolve [33] for each anagram stimulus, hemodynamic responses overlap in time which present challenges for data analysis. In order to evaluate the subject’s evoked-response times or brain activation for single anagram presentation within a block for graded difficulty analysis, a novel single trial hemodynamic response estimation algorithm was developed [61]. In this algorithm, each eventrelated hemodynamic response was estimated on the basis of two postulates that: (1) each single-trial hemodynamic response follows a gamma function, as given in Figure 6.3b, and (2) the total oxygenation data can be modeled by the summation of individual hemodynamic responses evoked by rapidly presented stimuli, . Each single trial was estimated by optimizing the error between the total oxygenation data from fNIR measurements and the linear model:

All calculations are applied to the data gathered from the left hemisphere of the prefrontal cortex on voxel 5. In block anagram study based on 14 participants (age between 18 and 23), the averaged recorded behavioral response times, the extracted rise times or time to peak (min), and the maximum amplitudes (as given in Figure 6.3b) from the estimated evoked-hemodynamic responses with respect

to the 3L, 4L, and 5L anagram sets are presented in Figure 6.10a. The estimated rise time which is the time required for the evoked-hemodynamic response to reach its maximum amplitude follows the same pattern as the behavioral (true) response time of the subjects having a correlation of R = 0.94 as presented in the scatter plot of the rise time versus response time in Figure 6.10b. Similarly, the estimated maximum amplitudes are correlated with the true response times (R = 0.73). The rise times and the maximum amplitude values increase as the difficulty level of the anagram solution increases, meaning that subjects need more time and more oxygen to solve difficult anagrams. Estimation of the event-related signals in a block design allows more precise analysis of the brain’s function during a cognitive/problem solving task.

6.6.2.2.5 Emotion

Two affective dimensions have been studied extensively in neuroimaging research on emotional stimuli: arousal (exciting or calming) and valence (positive or negative). In this work, a new paradigm in the study of emotional processes through functional neuroimaging is introduced where the influence of the valence and arousal of visual stimuli on the neuroimaging of the evoked-hemodynamic changes are demonstrated [105]. Using fNIR, evoked-cerebral blood oxygenation (CBO) changes in DLPFC during direct exposure to different emotion-eliciting stimuli (“on” period), and during the period directly following stimulus cessation (“off” period) is investigated. It is hypothesized that the evoked-CBO, rather than return to baseline after stimulus cessation, would show either overshoot or undershoot. The study includes 30 healthy subjects (15 female) between the ages of 19 and 51 and a total of 9 stimuli, which consist of video-clips of moderate length (~20 s) with different emotional content (ranging from violence to cartoons) that can ensure that the stimuli will provoke strong lasting responses. The total sample of trials studied (270) is classified according to the valence and arousal ratings given by the subjects.

A significant negative correlation was found between valence and arousal dimensions (r = −0.58; p < 0.001). That is to say, the lower the value in valence dimension, the higher the ratings in arousal. The four-way ANOVA for repeated measures revealed significant main effects for exposure condition (F[1,234] = 27.4; p < 0.001) and the grouping factors valence (F[2,234] = 3.93; p < 0.05) and arousal (F[1,234] = 6.94; p < 0.01). A Student t-test for repeated measures between mean oxy-Hb during “on” and “off” periods revealed a significant mean difference of −0.11 μM (t[239] = −2.98; p < 0.01). Comparisons between valence groups revealed a significant difference in oxygenation of 0.33 μM in unpleasant trials compared to neutral trials (t[125] = 3.28; p < 0.01); a significant difference of 0.4 μM was found when compared to pleasant trials (t[167] = 3.84; p < 0.001). No statistically significant differences were found between the oxy-Hb levels in neutral versus pleasant trials (t[182] = 0.94; p = 0.35). A Student t-test for independent measures between arousing and nonarousing trials revealed a significant mean difference of 0.33 μM (t(238) = −4.11; p < 0.001). There was no main effect for the prefrontal cortex (PFC) region variable (F[1,234] = 1.03; p = 0.31) and no significant interactions with this variable. A significant interaction was found between arousal category and exposure condition (F[1,234] = 16.14; p < 0.001): in arousing trials, a significant difference of −0.4 μM existed between “on” and “off” conditions (t[68] = −4.56; p < 0.001), while no significant differences were detected in nonarousing trials (t[170] = 0.23; p = 0.82).

As a result, one of the main findings of this study is that when the subjective degree of arousal is high, the representation of the stimulus remains in the prefrontal cortex, even when the stimulus is no longer present. The persistence of sources of DLPFC activation during the “off” period is closely related to the degree of arousal that the subject assigns to the stimulus. Furthermore, these results show that valence alone does not determine the persistence of activation in PFC beyond the “on” period, except when the valence is unpleasant. The new and principle finding of this study is that there is an overshoot related to level of arousal in the DLPFC that persists even when the arousing stimulus has disappeared. Data also confirms that valence and arousal have different effects on the course of evoked-CBO response in DLPFC. Significant differences between “on” and “off” periods of DLPFC activation based on valence ratings was not found, but significant poststimulus overshoot related to arousal ratings was observed. These findings provide the first fNIR evidence showing that an increment in subjective arousal leads to activation in DLPFC which persists after stimulus cessation and does not occur with nonarousing stimuli. Note that since arousing stimuli produce longer periods of brain activation than nonarousing stimuli, neuroimaging studies must consider the duration and affective dimensions of the stimulus as well as the duration of the scanning. Not accounting for this difference may contribute to misinterpretation of the data.

6.6.3 Studies Involving Complex Tasks, Field, and Clinical Applications

6.6.3.1 Cognitive Performance Assessment

In this study, the deployment of fNIR for the purpose of cognitive state assessment while the user performs a complex task is presented [21]. This work is based on data collected during the DARPA Augmented Cognition-Technical Integration Experiment session participated by a total of 8 healthy subjects, 3 females and 5 males, ranging in age from 18 to 50. The experimental protocol for this session used a complex task resembling a videogame called the Warship Commander Task (WCT). The WCT was designed and developed by Pacific Science & Engineering Group under the direction of Space and Naval Warfare Systems Center to simulate naval air warfare management [106]. A sample screen shot during WCT is as shown in Figure 6.11a.

Task load and task difficulty were manipulated by changing (i) the number of airplanes that had to be managed at a given time (6, 12, 18, and 24 plane waves), (ii) the number of unknown versus known airplane identities (two levels of difficulty, low: 33% of the planes were unknown, and high: 67% of the planes were unknown), and (iii) the presence or absence of a verbal memory task (a secondary task causing divided attention). Each participant completed four sets of WCT. Each set was comprised of 3 repetitions of each of the 4 wave sizes (in the order of 6, 18, 12, and 24 planes) where each wave lasted 75 s. The factors of 4 different wave sizes, 2 different task difficulties (high vs. low percentage of unknown airplanes), and full versus divided attention (secondary verbal memory task On or Off) were crossed to create a 4 Å~ 2 Å~ 2 repeated-measures design.

The fNIR data analysis explored (1) the relationships among cognitive workload, the participant’s performance and changes in blood oxygenation levels of the dorsolateral prefrontal cortex, and (2) the effect of divided attention as elicited by the secondary component of the WCT (the auditory task). The primary hypothesis was that blood oxygenation in the prefrontal cortex, as assessed by fNIR, would rise with increasing task load and would exhibit a positive correlation with performance measures. In support of the primary hypothesis, the results indicated that the rate of change in blood oxygenation was significantly sensitive across both hemispheres (F = 16.24, p < 0.001) to task load (wave size) changes (see Figure 6.11b when secondary verbal was off).

When attention is divided by the secondary verbal task, the primary effect occurred in the 24-plane wave (the most difficult condition) causing the mean oxygenation for this case to drop below that of the 18-plane wave (see Figure 6.11c). In line with the stated hypothesis, a preliminary interpretation of this finding was that a number of participants had reached their maximal level of performance in this most difficult task level and lost their concentration/effort resulting in a drop in blood oxygenation. The hypothesis also predicts that individuals who were able to stay on task and continue to perform in this difficult condition should demonstrate increased oxygenation relative to both: (1) their own oxygenation levels in the 18-plane wave and (2) individuals who became overwhelmed and disengaged. Because sustained concentration and engagement in the task should result in increased performance, a positive correlation between performance and blood oxygenation would provide support for this interpretation. A Pearson’s product-moment correlation indicated a very strong positive relationship between blood oxygenation and performance in the 24-plane condition (Pearson’s r = .89, p = .003).

A median split on the Percentage Game Score provided further evidence of the hypothesized relationship between cognitive effort and the blood oxygenation response. As can be seen in Figure 6.11c, the mean levels of oxygenation were higher for both high and low performers in the 24-plane wave than the 18-plane wave when the secondary verbal task was off. However, when the secondary verbal task was on, the more difficult condition, the individuals who performed well on the 24-plane wave showed a higher mean level of oxygenation for the 24-plane wave than for the 18-plane wave, whereas those who performed poorly showed a decrease in oxygenation relative to the 18-plane wave.

6.6.3.2 Brain–Computer Interface

Brain–computer interface (BCI) is defined as a system that translates neurophysiological signals detected from the brain to supply input to a computer or to control a device. BCI research largely targets to eliminate the need for motor movement and develop mechanisms to relay information directly from the brain to a computer which, in turn, can be used to control or communicate with outside world. Development of alternative communication strategies are a recognized need for clinical applications involving patients with complete paralysis, locked-in syndrome, spinal cord injury, or muscular dystrophy. A technique that bypasses muscles and acquires signals directly from brain would be a notable help. Moreover, this technique should be minimally intrusive, noninvasive, accessible, and safe to be used continuously. In addition to their use in neuroprosthetics, noninvasive BCI systems also have potential applications for healthy individuals especially for enhancing or accelerating the learning process, or in entertainment domains such as in computer games and multimedia applications as a neurofeedback mechanism.

The purpose of this research is to develop a new fNIR-based BCI to allow communication directly from the brain to a computer. A bar-size-control task based on a closed-loop system was designed and tested with 5 healthy subjects across two days [107]. In a single trial, subjects are first asked to rest for 20 s with a blank screen, after which a vertical or horizontal bar will appear (see Figure 6.12a). Initially, the bar is at 50% size and is mapped to the oxygenation data calculated from fNIR data that is updated at a frequency of 2 Hz. The subject is asked to concentrate on the bar for up to 120 s. Finally, the subject is asked to rate their effort on scale from 0 to 10 with 0 lowest and 10 highest effort/difficulty [108]. The subject has 30 s to complete this effort rating activity. Each trial lasts a maximum of 170 s. Comparisons of the average task and rest period oxygenation changes are significantly different (p < 0.01). The average task completion time (reaching + 90%) decreases with practice: day1 (mean 52.3 s) and day2 (mean 39.1 s). These preliminary results suggest that a closed-loop fNIR-based BCI can allow for a human–computer interaction with a mind switch task. Very preliminary studies incorporated fNIR technology into various 3D gaming environments that were built by student teams from Drexel University Digital Media (Figure 6.12b). Use of fNIR in this field is still an ongoing research, however preliminary results suggested that fNIR can allow participants to interact with virtual objects within the 3D environment by their thoughts.

6.6.3.3 Enhancement of Unmanned Aerial Vehicle Operator Training, Evaluation,and Interface Development

As the use of unmanned aerial vehicle (UAV) expand to near-earth applications and force-multiplying scenarios, current methods of operating UAVs and evaluating pilot performance need to expand as well. Research on human factors of UAV flight has identified several reasons underlying mishaps in UAV operations [109]. First, UAV operators have limited situational awareness due to the disembodied nature of UAV flight where operators need to fly UAVs by relying on limited camera angles. Since commands are transmitted over satellite links, UAVs are less responsive to operator input as compared to manned aircraft. In addition to this, typical UAV missions take long durations of time that require transitions

from long periods of dull flight mode to critical moments where operators need to stay alert to engage with a target or to attend to a contingency. Many human factor studies on UAV operations rely on self-reporting surveys to assess the situational awareness and cognitive workload of an operator during a particular task which can make comparisons between operators subjective. In short, physically and cognitively taxing aspects of UAV flight have resulted in a large number of mishaps during military and civilian use. Therefore, devising reliable indices for assessing cognitive load and level of expertise are of critical importance for evaluating training regiments and interface designs, and ultimately for improving the safety and success of UAV operations.

In this project, fNIR is utilized to monitor UAV operator’s cognitive workload and situational awareness during simulated missions [110]. The simulation platform is based on Microsoft’s Flight Simulator X with the Predator UAV add-on by Firstclass Simulations. Using a complete joystick, throttle, and rudder pedal controller set, this simulation environment approximates an actual MQ-1 Predator user interface (Figure 6.13). After completing a demo session, each subject completes a total of 8 twohour long training missions within 3 weeks. During each session subjects fly variants of the same mission, where they are asked to successfully take off, locate a submarine in a specified geographical area, pass over it to allow identification photographs to be taken, navigate back to an airfield with given coordinates, fly within 500 ft of the ground en route to the airfield over mountainous terrain, and successfully land after following a contingency maneuver revealed toward the end of the mission. These aspects, as well as other factors such as crosswinds, are added to the simulation to create realistic cognitive and physical demands, similar to those experienced by a real UAV pilot. This simulation environment allows replay of each session. In addition to the flight video, brain activation data is being collected by fNIR, as well as additional parameters such as pitch, roll, yaw, altitude, longitude, latitude, and air speed from within the simulation to aid in the assessment of performance (Figure 6.13).

In this ongoing study, both quantitative and qualitative methods to monitor the progress of each subject will be employed. Critical aspects of the mission that are likely to increase or decrease cognitive load (e.g., actively searching for a target, navigating toward a set of coordinates, and so on), will be identified through video analysis and the flight data. Once critical moments are identified and sampled, fNIR data collected at those moments can be correlated with operator performance to identify cognitive markers indicating expertise development and cognitive workload.

6.6.3.4 Cognitive Workload Assessment of Air Traffic Operators

The Next Generation Air Transportation System, developed by the Joint Planning and Development Office (JPDO), outlines a series of transformations designed to increase the capacity, safety, and security of air traffic operations in the United States [111]. A critical element in achieving this vision for future air-traffic management involves augmenting the current auditory-based communications between air traffic control (ATC) and the flight deck with text-based messaging, or DataComm systems. DataComm systems are expected to allow ATC to manage more air traffic at a lower level of cognitive load, thereby increasing both the capacity of the national airspace system and the safety of passengers. Although self-report measures of workload suggest that DataComm systems require less cognitive effort than voice-based systems to manage the same amount of traffic [112], to date this has not been tested using measures of neural function. The purpose of this research is to provide objective, brain-based measures of neural activity, and to determine the relative cognitive workload of DataComm versus voice-based communications systems (VoiceComm) during realistic simulations using fNIR.

In this study, fNIR has been incorporated into ongoing studies at the federal aviation administration (FAA) William J. Hughes Technical Center’s (WJHTC) Research, Development, and Human Factors Laboratory, where 24 certified professional controllers (CPC) between the ages of 24 and 55 who had actively controlled traffic in an Air Route Traffic Control Center between 3 and 30 years were monitored with fNIR while they (i) performed a classic working memory (n-back) task as explained before [95,96] and (ii) managed realistic ATC scenarios under typical and emergent conditions [113]. The primary objective of this study was to use neurophysiological measures to assess cognitive workload during completion of controlled complex cognitive tasks: n-back and ATC.

For the ATC part-task, each CPC controlled traffic on workstations with a high-resolution (2048 x 2048), 29″ radarscope, keyboard, trackball, and direct access keypad for 10 min. To display the air traffic, the DESIREE ATC simulator and the TGF systems that were developed by software engineers at the WJHTC were used. Six simulation pilots were used within scenarios by supporting one sector or two sectors and entering data at their workstations to maneuver aircraft, all based on controller clearances. Two types of communications, either voice (VoiceComm) or data (DataComm) communications were used in separate sessions in a pseudo-random order. For each communication type, task difficulty was varied by the number of aircraft in each sector containing 6, 12, or 18 aircraft.

For the n-back fNIR data, repeated measures ANOVA showed that average oxygenation changes occurred only at voxel 2, that is, close to AF7 in the International 10–20 System, located within the left inferior frontal gyrus in the dorsolateral prefrontal cortex, was significant (F3,69 = 4.37, p < 0.05), see Figure 6.14a. Post-hoc analyses confirmed the differences in oxygenation changes as a function of task difficulty with 3-back is larger than the 0- and 1-back tasks (q0.05/2, 69 = 3.72, p < 0.05). For the ATC data, a 2 (Communication: DataComm, VoiceComm) X 2 (Prefrontal Hemisphere: right, left) X 3 (Task Difficulty: 6, 12, 18 aircrafts) ANOVA with repeated measures on all factors was calculated on mean oxygenation. Two subjects (#10, #11) were excluded from the analyses because of high-motion artifact and low signal-to-noise ratios. There were only two significant main effects: (i) task difficulty denoted by number of aircraft [F2,42 = 4.39, p < 0.05] and (ii) communication [F1,21 = 5.09, p < 0.05], which is depicted in Figure 6.14b. Tukey post-hoc tests for task difficulty (q0.05/2, 42 = 3.44, p < 0.05) showed than the 18 aircrafts condition (M + SD; 0.272 + 0.586 μmol) had significantly higher oxygenation change than the 12 aircrafts condition (−0.015 + 0.409 μmol). There were no other significant differences between aircraft conditions.

As a summary, fNIR results were sensitive to task difficulty specifically at left inferior frontal gyrus in the n-back test. These are in line with the earlier results [21] and with the results of fMRI studies that have used the n-back task [114]. The main hypothesis of this project is that VoiceComm would require more cognitive resources than the DataComm condition. Hence, higher activation for VoiceComm would be expected. The fNIR results from the main effect of communication type (p < 0.05) confirms this hypothesis with a small to moderate effect size (d = 0.36). These fNIR results are also in line with subjective assessments of operators as reported in earlier studies. This study indicated that operator’s cognitive effort for different types of tasks can be objectively assessed by comparing fNIR results. One of the major advantages of using fNIR in this study is that it allowed monitoring brain activity of the ground operators in realistic settings.

6.6.3.5 Cognitive Activity Assessment Following Traumatic Brain Injury on Attention Domain

A commonly observed consequence of traumatic brain injury (TBI) is cognitive impairment, whose assessment represents a considerable challenge. Additionally, the choice of a successful treatment among the many existing neurorehabilitation strategies still relies on behavioral observation, and little information is available about the physiological changes produced at the brain level by the specific intervention. The integration of neuroimaging and electrophysiological measures may be more objective and effective in the evaluation of cognitive impairments. This study evaluates the applicability of fNIR for assessment of TBI-induced impairments of attention [115]. Participants included five male TBI patients between 18 and 37 years old. Brain activation measures were collected during a target categorization (oddball) task as explained before. The pattern of the hemodynamic response elicited by the two classes was in agreement with previous studies that investigated the applicability of fNIR to the study of attention-related tasks. The changes in the concentration of oxy-Hb and deoxy-Hb differ between target-locked and nontarget-locked responses: target stimuli elicited in fact a more marked activation in the areas of the prefrontal cortex monitored by fNIR. This pilot study suggests the potential for fNIR to be applied to the monitoring of attention in the TBI population.

6.6.3.6 Cognitive Activity Assessment Following TBI on Working Memory Domain

Behavioral observation and neuropsychological tests have guided the planning of cognitive rehabilitation and the assessment of its effectiveness. However, the information about the actual changes that rehabilitation interventions induce at the brain level is still limited. The availability of this information would, instead, prove useful for a more objective and individualized approach to the evaluation of a treatment efficacy. To this end, the integration of functional neuroimaging in the evaluation of cognitive impairments could offer a more objective monitoring of rehabilitation outcomes.

The aim of this study is to demonstrate the applicability of fNIR, to the assessment of working memory after TBI [116]. Participants included four TBI and ten healthy controls. Neuropsychological evaluation of attention and working memory was based on the WAIS-III Digit Span and WAIS-III Letter-Number Sequencing. Brain activation measurements were collected through fNIR during a visual n-back task as explained before. Physiologically irrelevant data and noise were eliminated from fNIR measurements and the maximum change in oxy-Hb and deoxy-Hb concentration was obtained for the left and right dorsolateral prefrontal areas.

On the basis of the performance in the neuropsychological tests, the TBI group revealed impaired learning performance as compared to the group of healthy controls. Evaluation of fNIR data revealed a pattern in the hemodynamic activity that was significantly different between the healthy controls and the TBI patients in the DLPFC, known to be involved in the working memory processes. The results obtained in this preliminary study show that fNIR is a promising neuroimaging tool that can prove to be clinically useful to investigate and identify the neurological underpinnings of cognitive functioning after TBI. In particular, fNIR offers to neurorehabilitation some unique advantages over other conventional neuroimaging techniques: fNIR is in fact portable, low-cost, noninvasive, and allows a reasonably flexible task design.

6.6.3.7 Depth of Anesthesia Monitoring

Awareness is an unintended mental state during general anesthesia. An accurate, objective measure of return to consciousness would provide an important safeguard for patients and physicians alike. This exploratory investigation on predicting awareness under general anesthesia examines the hypothesis that the transition from deep to light anesthetic stages is associated with reliable changes in oxygenated, deoxygenated, and total hemoglobin in frontal cortex. Hemodynamic changes during deep and light anesthesia were examined in 26 nonbrain surgery patients [55]. fNIR recordings are collected in the operating room before, during, and after subjects were receiving their scheduled anesthetic and surgical procedures. The results suggest that the rate of deoxygenated hemoglobin change can be used as a descriptive neuromarker to differentiate between deep and light anesthesia stages (F = 7.61, p < 0.01). This marker is proposed for further development as an index of the depth of anesthesia for the purpose of monitoring awareness under general anesthesia.

In addition to the neuropsychological findings, this research demonstrates engineering and signal processing solutions in the form of customized algorithms and procedures that allow fNIR to measure usable signals under field conditions. ICA and PCA were combined in a novel procedure that employed dark current (i.e., signal from noncortical sources) as a reference measurement. This method provided improved signal-to-noise ratio for the hemodynamic measurements acquired in the operating room, and can be used to increase the signal quality of fNIR for many other applications and field situations.

6.6.4 Future Directions

Since fNIR technology is still relatively new in comparison with other neuroimaging technologies, the research published to date has been relatively conservative—focusing on establishing fNIR as a valid and reliable neuroimaging technology. Furthermore clinical investigations have also been very limited. As a result, the majority of published fNIR studies have not capitalized on the unique capabilities of the technique because of the need to validate the results with known technologies such as fMRI. We believe that the studies explained in the previous p represent examples of how fNIR can be used in research and clinical paradigms that may not be feasible with other neuroimaging technologies and may facilitate research by acquainting clinicians and researchers with the unique merits of fNIR as a brainimaging technology.

There are several types of clinical applications that could benefit from the unique attributes of fNIR neuroimaging technology in the future:

• Populations that may not be able to readily tolerate the confines of an fMRI magnet, or be able to remain sufficiently still, for example, schizophrenics, autistic children, and neonates

• Populations that require the long-term monitoring of cerebral oxygenation, for example, premature and other high-risk infants

• Studies that require repeated, low-cost neuroimaging, for example, treatment studies that image the cortex for efficacy

• Applications where an fMRI system would be too expensive or cumbersome, for example, for use in a clinical office

• Applications that require ecological validity, for example, working at a computer, or in an educational setting

6.7 Conclusion

fNIR is an emerging technology that uses near-infrared light to measure changes in the concentration of oxygenated and deoxygenated hemoglobin in the cortex. The use of fNIR technology has increased in recent years as a means to measure hemodynamic changes in the cortex in response to cognitive activity. Although fNIR imaging is limited to the outer cortex, it provides neuroimaging that is safe, portable, and very affordable relative to other neuroimaging technologies that can be applied in the laboratory as well as field conditions. Moreover, it is a noninvasive and negligibly intrusive optical imaging modality. It is also relatively robust to movement artifacts, and can readily be integrated with other technologies such as EEG. Current state-of-the-art technologies include portable and wireless fNIR systems, data processing algorithms, and end-user software.

In this chapter, an overview of basis of optical imaging, fNIR instrumentation and signal analysis and cognitive studies carried out in the literature are summarized. In these studies, fNIR has been demonstrated to have adequate validity in the measurement of functional brain activity during a variety of cognitive, emotional, and motor tasks in healthy and diseased subject groups of both adult and children populations, and has the potential to provide a flexible neuroimaging tool for clinicians and researchers alike. The findings are in agreement with the results in current EEG and fMRI literature. The portable and wireless instrumentations, combined with robust analysis algorithms and end-user software, make fNIR a viable option for the study of cognition- and emotion-related hemodynamic changes in both adults and children, under either stationary or ambulant conditions.

Acknowledgments

The author would like to thank Drs. Banu Onaral, Kambiz Pourrezaei, Scott Bunce, Patricia Shewokis, Kurtulus Izzetoglu, Hasan Ayaz, Anna Merzagora, Britton Chance, Shoko Nioka, Jose Leon-Carrion, Murat Cakir, Ajit Devaraj, and Mr. Justin Menda and Mr. Adrian Curtin for their advice and help in the preparation of this chapter and sharing their work and results.

References

\1. Jobsis FF. 1977. Noninvasive infrared monitoring of cerebral and myocardial sufficiency and circulatory parameters. Science, 198:1264–1267.

\2. Cope M, Delpy, D. T. 1988. System for long-term measurement of cerebral blood flow and tissue oxygenation on newborn infants by infra-red transillumination. Med. Biol. Eng. Comput., 26:289–294. Functional Optical Brain Imaging 6-29

\3. Luo Q, Zeng S, Chance B, Nioka S. 2002. Monitoring of brain activation with near infrared spectroscopy. Chapter 8 in Handbook of Optical Biomedical Diagnostics, Editor: Valery V. Tuchin, Press Monograph PM107, Bellingham, WA.

\4. Rolfe P. 2000. In vivo near-infrared spectroscopy. Annu. Rev. Biomed. Eng., 02:715–754.

\5. Strangman G, Boas DA, Sutton JP. 2002. Non-invasive neuroimaging using near-infrared light. Biol. Psychiatry, 52(7):679–693.

\6. Hoshi Y, Tamura M. 1993. Dynamic multichannel near-infrared optical imaging of human brain activity. J. Appl. Phys., 75:1842–1846.

\7. Hoshi Y. 2003. Functional near-infrared optical imaging: Utility and limitations in human brain mapping. Psychophysiology, 40:511–520.

\8. Villringer A, Planck J, Hock C, Schleinkofer L, Dirnagl U. 1993. Near infrared spectroscopy (NIRS): A new tool to study hemodynamic changes during activation of brain function in human adults. Neurosci. Lett., 154:101–104.

\9. Villringer A, Chance B. 1997. Non-invasive optical spectroscopy and imaging of human brain function. Trends Neurosci., 20:435–442.

\10. Suto T, Ito M, Uehara T, Ida I, Fukuda M, Mikuni M. 2002. Temporal characteristics of cerebral blood volume change in motor and somatosensory cortices revealed by multichannel near infrared spectroscopy.Int. Congress Series, 1232:383–388.

\11. Maki A, Yamashita Y, Ito Y, Watanabe E, Mayanagi Y, Koizumi H. 1995. Spatial and temporal analysis of human motor activity by using noninvasive NIR topography. J. Neurosci., 11:1458–1469.

\12. Gratton E, Toronov V, Wolf U, Wolf M, Webb A. 2005. Measurement of brain activity by near infrared light. J. Biol. Optics, 10(1)011008-1-13.

\13. Franceschini MA, Boas DA. 2004. Noninvasive measurement of neuronal activity with near-infrared optical imaging. Neuroimage, 21:372–386.

\14. Gratton G, Corballis PM, Cho E, Fabiani M, Hood DC. 1995. Shades of gray matter: Noninvasive optical images of human brain responses during visual stimulation. Psychophysiology, 32:505–509.

\15. Heekeren HR, Obrig H, Wenzel R, Eberle K, Ruben J, Villringer K, Kurth R, Villringer A. 1997.Cerebral haemoglobin oxygenation during sustained visual stimulation—A near infrared spectroscopy study. Physiol. Trans. Biol. Sci., 352:743–750.

\16. Sato, H, Takeuchi T, Sakai K. 1999. Temporal cortex activation during speech recognition: An optical topography study. Cognition, 40:548–560.

\17. Zaramella P, Freato F, Amigoni A, Salvadori S, Marangoni P, Suppjei A. 2001. Brain auditory activation measured by near-infrared spectroscopy. Ped. Res., 49:213–219.

\18. Son Il-Y, Guhe M, Yazici B. 2005. Human performance assessment using fNIR. Proc. SPIE, 5797:158–169.

\19. Cope M. 1991. The Development of a Near-Infrared Spectroscopy System and Its Application for Noninvasive Monitoring of Cerebral Blood and Tissue Oxygenation in the Newborn Infant. Ph.D. thesis. University College London, London.

\20. Bunce SC, Devaraj A, Izzetoglu M, Onaral B, Pourrezaei K. 2005a. Detecting deception in the brain: A functional near-infrared spectroscopy study of neural correlates of intentional deception. Nondest. Detection Meas. Homeland Security III, Proc. SPIE, 5769:24–32.

\21. Izzetoglu K, Bunce S, Onaral B, Pourrezaei K, Chance B, 2004. Functional optical brain imaging using near-infrared during cognitive tasks. Int. J. Human-Comp. Int., 17(2):211–227.

\22. Izzetoglu K, Yurtsever G, Bozkurt A, Yazici B, Bunce S, Pourrezaei K, Onaral B. 2003a. NIR spectroscopy measurements of cognitive load elicited by GKT and target categorization. Proceedings of 36th Hawaii International Conference on System Sciences, Philadelphia, PA.

\23. Izzetoglu M, Izzetoglu K, Bunce S, Ayaz H, Devaraj A, Onaral B, Pourrezaei K. 2005b. Functional near-infrared neuroimaging. IEEE Trans. Neural Sys. Rehab. Eng., 13(2):153–159.

\24. Platek SM, Fonteyn LCM, Izzetoglu M, Myers TE, Ayaz H, Li C, Chance B. 2005. Functional near infrared spectroscopy reveals differences in self-other processing as a function of schizotypal personality traits. Schizophrenia Res. 73(1):125–127.

\25. Rajapakse JC, Kruggel F, Maisog JM, Von Cramon DY. 1998. Modelin hemodynamic response for analysis of functional MRI time series. Hum. Brain Mapp., 6:283–300.

\26. Magistretti PR, Pellerin L. 1999. Cellular mechanisms of brain energy metabolism and their relevance to functional brain imaging. Phil. Trans. R. Soc. Lond. B, 354(1387):1155–1163.

\27. Magistretti PR. 2000. Cellular bases of functional brain imaging: Insights from neuron-glia metabolic coupling. Brain Res., 886(1–2):108–112.

\28. Ames A III. 2000. CNS energy metabolism as related to function. Brain Res. Brain Res. Rev. 34:42–68.

\29. Fox PT, Raichle ME, Mintun MA, Dence C. 1988. Nonoxidative glucose consumption during focal physiologic neural activity. Science, 241:462–464.

\30. Buxton RB, Uludag K, Dubowitz DJ, Liu TT. 2004. Modeling the hemodynamic response to brain activation. Neuroimage, 23(Suppl. 1):S220–233.

\31. Buxton RB. 2001. The elusive initial dip. Neuroimage, 13: 953–958.

\32. Obrig H, Villringer A. 2003. Beyond the visible—Imaging the human brain with light. J. Cereb. Blood Flow Metab., 23:1–18.

\33. Miezin FM, Maccotta L, Ollinger JM, Petersen SE, Buckner RL. 2000. Characterizing the hemodynamic response: Effects of presentation rate, sampling procedure, and the possibility of ordering brain activity based on relative timing. NeuroImage, 11:735–759.

\34. Haensse D, Szabo P, Brown D, Fauchère JC, Niederer P, Bucher HU, Wolf M. 2005. A new multichannel near infrared spectrophotometry system for functional studies of the brain in adults and neonates. Optics Exp., 13(12):4525–4538.

\35. Ogawa S, Lee TM, Nayak AS, Glynn P. 1990. Oxygenation-sensitive contrast in magnetic resonance image of rodent brain at high magnetic fields. Magn. Reson. Med., 14:68–78.

\36. Kwong KK, Belliveau JW, Chesler DA, Goldberg IE, Weisskoff RM, Poncelet BP, Kennedy DN. et al. 1992. Dynamic magnetic resonance imaging of human brain activity during primary sensory stimulation. PNAS, 89:5951–5955.

\37. Chance B, Cope M, Gratton E, Ramanujam N, Tromberg B. 1998. Phase measurement of light absorption and scatter in human tissue. Rev. Sci. Instrum., 69:3457–3482.

\38. Gratton G, Maier JS, Fabiani M, Mantulinm WW, Gratton E. 1994. Feasibility of intracranial nearinfrared optical scanning. Psychophysiology, 31:211–215.

\39. Chance B, Leig JS, Miyake H, Smith DS, Nioka S, Greenfeld R, Finander M, Kaufmann K, Levy W. et al. 1988. Comparison of time-resolved and un-resolved measurements of deoxyhemoglobin in brain. PNAS, 85:4971–4975.

\40. Firbank M, Okada E, Delpy DT. 1998. A theoretical study of the signal contribution of regions of the adult head to near infrared spectroscopy studies of visual evoked responses. NeuroImage, 8:69–78.

\41. Delpy DT, Cope M, van der Zee P, Arridge S, Wray S, Wyatt J. 1988. Estimation of optical pathlength through tissue from direct time of flight measurement. Phys. Med. Biol, 33:1433–1442.

\42. Lakowicz, J. R, Berndt, K. 1990. Frequency domain measurement of photon migration in tissues. Chem. Phys. Lett., 166:246–252.

\43. Wolf M, Ferrari M, Quaresima V. 2007. Progress of near-infrared spectroscopy and topography for brain and muscle clinical applications. J. Biomed. Optics, 12(6):062104–1–14.

\44. Obrig H, Wenzel R, Kohl M, Horst S, Wobst P, Steinbrink J, Thomas F, Villringer A. 2000. Nearinfrared spectroscopy: Does it function in functional activation studies of the adult brain? Int. J. Psychophysiol., 35:125–142.

\45. Boas DA, Franceschini MA, Dunn AK, Strangman G. 2002. Non-Invasive Imaging of Cerebral Activation with Diffuse Optical Tomography. In-Vivo Optical Imaging of Brain Function. Boca Raton: CRC Press. pp. 193–221.

\46. Chance B, Anday E, Nioka S, Zhou S, Hong L, Worden K, Li C, Murray T, Ovetsky Y, Pidikiti D, Thomas R. 1998. A novel method for fast imaging of brain function, non-invasively, with light. Optics Exp., 2(10):411–423. Functional Optical Brain Imaging 6-31

\47. Rector DM, Poe GR, Kristensen MP, Harper RM. 1997. Light scattering changes follow evoked potentials from hippocampal Schaeffer collateral stimulation. J. Neurophysiol., 78:1707–1713.

\48. Stepnoski RA, LaPorta A, Raccuia-Behling F, Blonder GE, Slusher RE, Kleinfeld D. 1991. Noninvasive detection of changes in membrane potential in cultured neurons by light scattering. PNAS, 88:9382–9386.

\49. Bunce S, Izzetoglu M, Izzetoglu K, Onaral B, Pourrezaei K, 2006. Functional near infrared spectroscopy: An emerging neuroimaging modality. IEEE Engineering in Medicine and Biology Magazine, Special issue on Clinical Neuroengineering, 25(4):54–62.

\50. Bozkurt A, Onaral B. 2004. Safety assessment of near infrared light emitting diodes for diffuse optical measurements. Biomed. Eng. Online, 3(9):1–10.

\51. Boas DA, Dale AM, Francheschini MA. 2004. Diffuse optical imaging of brain activation: Approaches to optimizing image sensitivity, resolution, and accuracy. Neuroimage, 23:S275–S288.

\52. Huppert TJ, Diamond SG, Franceschini MA, Boas DA. 2009. HomER: A review of timeseries analysis methods for near-infrared spectroscopy of the brain. Appl. Opt., 48(10):280–298.

\53. Izzetoglu M, Bunce S, Izzetoglu K, Onaral B, Pourrezaei K. 2007. Functional brain imaging using

near infrared technology for cognitive activity assessment. IEEE Engineering in Medicine and Biology Magazine, Special issue on on the Role of Optical Imaging in Augmented Cognition, 26(4):38–46.

\54. Sato H, Tanaka N, Uchida M, Hirabayashi Y, Kanai M, Ashida T, Konishi I, Maki A. 2006. Wavelet analysis for detecting body-movement artifacts in optical topography signals. Neuroimage, 33:580–587.

\55. Izzetoglu K. 2008. Neural Correlates of Cognitive Workload and Anesthetic Depth: fNIR Spectroscopy Investigation in Humans. Ph.D. thesis, Drexel University, Philadelphia, PA.

\56. Robertson FC, Douglas TS, Meintjes EM. 2010. Motion artefact removal for functional near infrared spectroscopy: A comparison of methods. IEEE Trans. Biomed. Eng., 57(6):1377–1387.

\57. Izzetoglu M, Devaraj A, Bunce S, Onaral B. 2005. Motion artifact cancellation in NIR spectroscopy using Wiener filtering. IEEE Trans. Biomed. Eng., 52(5):934–938.

\58. Izzetoglu M, Chitrapu P, Bunce S, Onaral B. 2010. Motion artifact cancellation in NIR spectroscopy using discrete Kalman filtering. Biomed. Eng. Online, 9(16):1–10.

\59. Cui X, Bray S, Reiss AL. 2010. Functional near infrared spectroscopy (NIRS) signal improvement

based on negative correlation between oxygenated and deoxygenated hemoglobin dynamics. Neuroimage, 49:3039–3046.

\60. Devaraj A. 2005. Signal Processing for Functional Near Infrared Neuroimaging. Master’s thesis, Drexel University, Philadelphia, PA.

\61. Izzetoglu M, Nioka S, Chance B, Onaral B. 2005. Single trial hemodynamic response estimation in a block anagram solution study using fNIR spectroscopy. Proceedings of ICASSP Conference, Vol. 5, pp: 633–636.

\62. Obrig H, Neufang M, Wenzel R, Kohl M, Steinbrink J, Einhaupl K, Villringer A. 2000. Spontaneous low frequency oscillations of cerebral hemodynamics and metabolism in human adults. Neuroimage, 12:623–639.

\63. Mayhew J, Askew S, Zheng Y, Porrill J, Westby GWM, Redgrave P, Rector DM, Harper RM. 1996. Cerebral vasomotion: A 0.1 hz oscillation in refleted light imaging of neural acitvity, Neuroimage, 4:183–193.

\64. Lundberg N. 1960. Continuous recordings and control of ventricular fluid pressure in neurosurgical practice. Acta Psychiatrica et Neurologica Scandinavica, 149:1–193.

\65. Diamond SG, Huppert TJ, Kolehmainen V, Franceschini MA, Kaipio JP, Arridge SR, Boas DA. 2005. Physiological system identification with the Kalman filter in diffuse optical tomography. Lecture Notes in Computer Science, 3750:649–656.

\66. Zhang Q, Strangman GE, Ganis G. 2009. Adaptive filtering to reduce global interference in non-invasive NIRS measures of brain activation: How well and when does it work? Neuroimage, 45:788–794.

\67. Zhang Y, Brooks DH, Francheschini MA, Boas DA. 2005. Eigenvector-based spatial filtering for reduction of physiological interference in diffuse optical imaging. J. Biomed. Opt., 10(1):011014.

\68. Gratton G, Corballis PM. 1995. Removing the heart from the brain: Compensation for the pulse artifact in the photon migration signal. Psychophysiology, 32:292–299.

\69. Morren G, Wolf U, Lemmerling P, Wolf M, Choi JH, Gratton E, De Lathauwer L, Van Huffel S. 2004. Detection of fast neuronal signals in the motor cortex from functional near infrared spectroscopy measurements using independent component analysis. Med. Biol. Eng. Comput., 42(1):92–99.

\70. Glover GH, Li TQ, Ress D. 2000. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn. Reson. Med., 44:162–167.

\71. Yamada T, Umeyama S, Matsuda K. 2009. Multidistance probe arrangement to eliminate artifacts in functional near-infrared spectroscopy. J. Biomed. Opt., 14(6):064034.

\72. Arridge SR, Schweiger M. 1997. Image reconstruction in optical tomography. Phil. Trans. R. Soc. Lond. B, 352:717–726.

\73. Boas DA, Brooks DH, Miller EL, DiMarzio CA, Kilmer M, Gaudette RJ, Zhang Q. 2001. Imaging the body with diffuse optical tomography. IEEE Sign. Proc. Mag., 18:57–75.

\74. Son IY, Yazici B. 2006. Near infrared imaging and spectroscopy for brain activity monitoring. Advances in Sensing with Security Applications, pp: 341–372, NATO Advanced Study Institute, NATO Security through Science Series-A: Chemistry and Biology, Springer, Edited by J. Byrnes.

\75. Ishimaru A. 1997. Wave Propagation and Scattering in Random Media. New York: IEEE Press.

\76. Tromberg BJ, Svaasand LO, Tsay T, Haskell RC. 1993. Properties of photon density waves in multiple- scattering media. Appl. Optics, 32:607–616.

\77. Hillman E. 2002. Experimental and Theoretical Investigations of Near Infrared Tomographic Imaging Methods and Clinical Applications. Ph.D. thesis, University College London, London.

\78. Patterson MS, Chance B, Wilson BC. 1989. Time resolved reflectance and transmittance for the noninvasive measurement of tissue optical properties. Appl. Optics, 28(12):2331–2336.

\79. Sassaroli A, Fantini S. 2004. Comment on the modified Beer–Lambert law for scattering media. Phys. Med. Biol., 49:N255–N257.

\80. Tachtsidis I. 2005. Experimental Measurements of Cerebral Haemodynamics and Oxygenation and Comparisons with a Computational Model: A Near-Infrared Spectroscopy Investigation. Ph.D. thesis, University College London, London.

\81. Duncan A, Meek JH, Clemence M, Elwell CE, Tyszczuk L, Cope M, Delpy DT. 1995. Optical pathlength measurements on adult head, calf and forearm and the head of the newborn infant using phase resolved optical spectroscopy. Phys. Med. Biol., 40:295–304.

\82. Hiraoka M, Firbank M, Essenpreis M, Cope M, Arridge SR, van der Zee P, Delpy DT. 1993. A Monte Carlo investigation of optical pathlength in inhomogeneous tissue and its application to near-infrared spectroscopy. Phys. Med. Biol., 38:1859–1876.

\83. Kohl M, Nolte C, Heekeren HR, Horst S, Scholz U, Obrig H, Villringer A. 1998. Determination of the wavelength dependence of the differential pathlength factor from near-infrared pulse signals. Phys Med Biol., 43(6):1771–1782.

\84. Duncan A, Meek JH, Clemence M, Elwell CE, Fallon P, Tyszczuk L, Cope M, Delpy DT. 1996. Measurement of cranial optical path length as a function of age using phase resolved near infrared spectroscopy. Pediatr. Res., 39(5):889–894.

\85. Meek JH, Firbank M, Elwell CE, Atkinson J, Braddick O, Wyatt JS. 1998. Regional hemodynamic responses to visual stimulation in awake infants. Pediatr. Res., 43:840–843.

\86. Sakatani K, Katayama Y, Yamamoto T, Suzuki S. 1999. Changes in cerebral blood oxygenation of the frontal lobe induced by direct electrical stimulation of thalamus and globus pallidus: A near infrared spectroscopy study. J. Neurol. Neurosurg. Psych., 67:769–773.

\87. Watanabe E, Maki A, Kawaguchi F, Yamashita Y, Koizumi H, Mayanagi Y. 2000. Noninvasive cerebral blood volume measurement during seizures using multichannel near infrared spectroscopic topography. J. Biomed. Opt., 5:287–290.

\88. Haginoya K, Munakata M, Kato R, Yokoyama H, Ishizuka M, Iinuma K. 2002. Ictal cerebral haemodynamics of childhood epilepsy measured with nearinfrared spectrophotometry. Brain, 125(9):1960–1971. Functional Optical Brain Imaging 6-33

\89. Hock C, Villringer K, Muller-Spahn F, Hofmann H, Heekeren H, Schuh-Hofer S. 1996. Near infrared spectroscopy in the diagnosis of Alzheimer’s disease. Ann. N.Y. Acad. Sci., 777:22–29.

\90. Okada F, Tokumitsu Y, Hoshi Y, Tamura M. 1994. Impaired interhemispheric integration in brain oxygenation and hemodynamics in schizophrenia. Eur. Arch. Psych. Clin. Neurosci., 244:17–25.

\91. Fallgatter AJ, Strik WK. 2000. Reduced frontal functional asymmetry in schizophrenia during a cued continuous performance test assessed with nearinfrared spectroscopy. Schizophr. Bull, 26(4):913–919.

\92. Eschweiler GW, Wegerer C, Schlotter W, Spandl C, Stevens A, Bartels M. 2000. Left prefrontal activation predicts therapeutic effects of repetitive transcranial magnetic stimulation (rTMS) in major depression. Psych. Res., 99:161–172.

\93. Irani F, Platek SM, Bunce S, Ruocco AC, Chute D. 2007. Functional near infrared spectroscopy (fNIRS): An emerging neuroimaging technology with important applications for the study of brain

disorders. Clin. Neuropsychol., 21:9–37.

\94. Cabeza R, Nyberg L. 2000. “Imaging Cognition II: An empirical Review of 275 PET and fMRI Studies.” J. Cogn. Neurosci., 12:1–47.

\95. Smith EE, Jonides J. 1997. “Working Memory: A View from Neuroimaging,” Cogn. Psychol. 33:5–42.

\96. Smith EE, Jonides J. 1999. Storage and executive processes in the frontal lobes. Science, 283:1657–1661.

\97. Braver TS, Cohen JD, Nystrom LE, Jonides J. Smith EE, Noll DC. 1997. A parametric study of prefrontal cortex involvement in human working memory. NeuroImage, 5:49–62.

\98. McCarthy G, Luby M, Gore J, Goldman-Rakic P. 1997. Infrequent events transiently activate human prefrontal and parietal cortex as measured by functional MRI. J. Neurophysiol., 77:1630–1634.

\99. Polich J, Kok A. 1995. Cognitive and biological determinants of P300: An integrative review. Biol. Psychol., 41:103–146.

\100. Ardekani BA, Choi SJ, Hossein-Zadeh G, Porjesz B, Tanabe JL, Lim KO, Bilder R, Helpern JA, Begleiter H. 2002. Functional magnetic resonance imaging of brain activity in the visual oddball task. Cogn. Brain Res., 14:347–356.

\101. Leon-Carrion J, Izzetoglu M, Izzetoglu K, Martin-Rodriguez JF, Damas-Lopez J, Martin JM, Dominguez-Morales MR. 2010. Efficient learning produces spontaneous neural repetition suppression in prefrontal cortex. Behav. Brain Res., 208(2):502–508.

\102. Strangman G, Franceschini MA, Boas DA. 2003. Factors affecting the accuracy of near-infrared spectroscopy concentration calculations for focal changes in oxygenation parameters. Neuroimage, 18:865–879.

\103. Huppert TJ, Hoge RD, Diamond SG, Franceschini MA, Boas DA. 2006. A temporal comparison of BOLD, ASL, and NIRS hemodynamic responses to motor stimuli in adult humans. Neuroimage, 29:368–382.

\104. Chance B., Nioka S., Sadi S., Li C. 2003. Oxygenation and blood concentration changes in human subject prefrontal activation by anagram solutions. Adv. Exp. Med. Biol., 510:397–401.

\105. Leon-Carrion J, Martin-Rodriguez JF, Damas-Lopez J, Pourrezai K, Izzetoglu K, Martin JM, Dominguez-Morales MR, 2007. A lasting post-stimulus activation on dorsolateral prefrontal cortex is produced when processing valence and arousal in visual affective stimuli. Neurosci. Lett., 422(3):147–152.

\106. John M, Kobus DA et al. 2002. A multi-tasking environment for manipulating and measuring neural Correlates of cognitive workload”. Proceedings of the IEEE 7th Conference on Human Factors and Power Plants, 7:10–14.

\107. Ayaz H, Shewokis P, Bunce S, Schultheis M, Onaral B. 2009. Assessment of Cognitive Neural Correlates for a Functional Near Infrared-based Brain Computer Interface System. Dylan D. Schmorrow at al. (eds.): Augmented Cognition, HCII 2009, Lecture Notes on Artificial Intelligence, 2009, vol. 5638:699–708, Berlin, Heidelberg: Springer-Verlag (presented at the 13th International Conference on Human-Computer Interaction, July 19–24 2009, San Diego, CA).

\108. Paas FGWC, Van Merriënboer JJG. 1993. The efficiency of instructional conditions: An approach to combine mental effort and performance measures. human factors. J. Hum. Fact. Ergon. Soc., 35:737–743. 6-34 Biosignal Processing

\109. Cooke NJ, Pringle H, Pederson H, Connor O, Salas E. 2006. Human Factors of Remotely Operated Vehicles. The Netherlands: Elsevier.

\110. Cakir MP, Ayaz H, Menda J, Izzetoglu K, Onaral B. 2010. Connecting brain and learning sciences: An optical brain imaging approach to monitoring development of expertise in UAV piloting. Proceedings of ICLS2010 Conference, June 29–July 2 2010, Chicago, IL.

\111. JPDO. 2004. Next Generation Air Transportation System Integrated Plan. Retrieved from http://www.jpdo.gov/library/NGATS_v1_1204r.pdf.

\112. Hah S, Willems B, Phillips R. 2006. The effect of air traffic increase on controller workload. Proceedings of the Human Factors and Ergonomics Society Annual Meeting San Francisco, CA.

\113. Ayaz H, Willems B, Bunce S, Shawokis PA, Izzetoglu K, Hah S, Deshmukh AR, Onaral B. 2010. Cognitive workload assessment of air traffic Controllers Using Optical Brain Imaging Sensors. Proceedings of AHFE2010 Conference, July 17–20 2010, Miami, FL.

\114. Owen AM, McMillan KM, Laird AR, Bullmore E. 2005. N-back working memory paradigm: A metaanalysis of normative functional neuroimaging studies. Hum. Brain Mapp., 25(1):46–59.

\115. Merzagora AC., Izzetoglu M., Onaral B., Schultheis M. 2010. “fNIRS as a useful tool for cognitive evaluation following traumatic brian injury”, 8th World Congress on Brian Injury, March 10–14, 2010, Washington, DC.

\116. Merzagora A.C, Martin-Rodriguez J.F, Longo Perez A, Leon-Dominguez U, Izzetoglu K, Schultheis M, Onaral B, Leon-Carrion J. 2010. “Applicability of fNIRS to the TBI population: Demonstration on an attention task”, 8th World Congress on Brian Injury, March 10–14, 2010, Washington, DC.

谢谢大家观看，如有帮助，来个喜欢或者关注吧！

本文作者：Chen Rui

博客地址 ： Chen Rui Blog
知乎地址 : 知乎专栏
B站地址 : B站主页
书店地址 : 书店主页
网易云音乐地址 : 音乐主页