Functional Near-Infrared Neuroimaging

IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 13, NO. 2, JUNE 2005 153

Functional Near-Infrared NeuroimagingMeltem Izzetoglu, Kurtulus Izzetoglu, Scott Bunce, Hasan Ayaz, Ajit Devaraj, Banu Onaral, Fellow, IEEE, and

Kambiz Pourrezaei

Abstract—Functional near-infrared spectroscopy (fNIR) is aneroimaging modality that enables continuous, noninvasive, andportable monitoring of changes in blood oxygenation and bloodvolume related to human brain function. Over the last decade,studies in the laboratory have established that fNIR spectroscopyprovides a veridical measure of oxygenation and blood flow inthe brain. Our recent findings indicate that fNIR can effectivelymonitor cognitive tasks such as attention, working memory, targetcategorization, and problem solving. These experimental outcomescompare favorably with functional magnetic resonance imaging(fMRI) studies, and in particular, with the blood oxygenation leveldependent signal. Since fNIR can be implemented in the form ofa wearable and minimally intrusive device, it has the capacity tomonitor brain activity under real life conditions and in everydayenvironments. Moreover, the fNIR system is amenable to inte-gration with other established physiological and neurobehavioralmeasures, including electroencephalogram, eye tracking, pupil re-flex, heart rate variability, respiration, and electrodermal activity.

Index Terms—Brain activity, functional near-infrared (fNIR),neuroimaging.

I. INTRODUCTION

WELL-ESTABLISHED neuroimaging techniques suchas functional magnetic resonance imaging (fMRI) and

positron emission tomography (PET) have been widely used toimage brain functions in humans. These techniques have greatlyincreased our knowledge about the neural circuits that underliecognitive and emotional processes [1], [2]. However, each ofthese neuroimaging technologies has both strengths and limita-tions. fMRI is noninvasive and has excellent spatial resolution,but is also expensive, highly sensitive to motion artifact, con-fines the participants to restricted positions inside the magnet,is difficult to integrate with other imaging modalities [suchas electroencephalogram (EEG)], and exposes participants toloud noises. PET also requires a restricted range of motion andconfinement, and requires the injection of radioactive materials.These characteristics make these imaging modalities unsuitablefor many uses, including use with children, the elicitation ofpositive affect, and the monitoring of ongoing cognitive activityunder routine working conditions.

Manuscript received December 24, 2004; revised January 10, 2005; acceptedJanuary 10, 2005. This work was supported in part by the Defense AdvancedResearch Projects Agency (DARPA) under the Augmented Cognition Program,and in part by the Office of Naval Research (ONR) and Homeland Securityunder Agreement Numbers N00014-02-1-0524, N00014-01-1-0986, andN00014-04-1-0119.

M. Izzetoglu, K. Izzetoglu, H. Ayaz, B. Onaral, and K. Pourrezaei are with theSchool of Biomedical Engineering, System and Health Systems, Drexel Univer-sity, Philadelphia, PA 19104 USA (e-mail: [email protected]).

S. Bunce is with the Department of Psychiatry, Drexel University College ofMedicine, Philadelphia, PA 19104 USA.

A. Devaraj is with the Electrical and Computer Engineering Department,Drexel University, Philadelphia, PA 19104 USA.

Digital Object Identifier 10.1109/TNSRE.2005.847377

In the last decade, functional near-infrared (fNIR) spec-troscopy has been introduced as a new neuroimaging modalitywith which to conduct functional brain-imaging studies. fNIRtechnology uses specific wavelengths of light, introduced atthe scalp, to enable the noninvasive measurement of changesin the relative ratios of deoxygenated hemoglobin (deoxy-Hb)and oxygenated hemoglobin (oxy-Hb) in the capillary bedsduring brain activity. This technology allows the design ofportable, safe, affordable, noninvasive, and minimally intrusivemonitoring systems. These qualities make fNIR suitable for thestudy of hemodynamic changes due to cognitive and emotionalbrain activity under many working and educational conditions,as well as in the field.

Photons interact with the tissue in several ways, includingabsorption and scattering [3]. Most biological tissues are rel-atively transparent to light in the near infrared range between700 and 900 nm [3], therefore, relatively little scattering ofphotons occurs when these wavelengths are introduced to tissue.Fortuitously, within this “optical window” the chromophoresoxy-Hb and deoxy-Hb reflect these wavelengths, hence serveas biologically relevant markers with which to monitor neuralactivity [3], [4]. fNIR technology employs specified wavelengthsin the optical window which easily pass through most tissue,but reflect back from oxy- and deoxy-Hb. Since photons scatterin a relatively predictable pattern, they can be measured usingphotodetectors on the surface of the skin. The relative levels ofabsorption and back-scatter from oxy and deoxy-Hb provideinformation about neural activity via a process known as neu-rovascular coupling.

Typically, an fNIR apparatus consists of a light source bywhich tissue is radiated and a light detector that receives lightafter it has interacted with the tissue. According to the modifiedBeer-Lambert Law [3], the light intensity after the photons haveinteracted with the biological tissue is expressed by the equa-tion: where is a factorthat accounts for the measurement geometry and is assumedconstant when concentration changes. is input light intensity,

and are the molar extinction coefficients of deoxy-and oxy-Hb, and are the concentrations of chro-mophores, deoxy- and oxy-Hb, respectively, and is the photonpath which is a function of absorption and scattering coefficients

and . Using the modified Beer-Lambert law and fNIRmeasurements performed at two different wavelengths withinthe near infrared light range and at different times, the relativechanges in the concentrations of deoxy- and oxy-Hb can be ob-tained. Using this technique, several types of brain function havebeen assessed, including motor and visual activation, auditorystimulation and performance of various cognitive tasks [3]–[5].

Functional imaging is typicallyconducted inaneffort tounder-stand the activity in a given brain region in terms of its relation-

1534-4320/$20.00 © 2005 IEEE

154 IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING, VOL. 13, NO. 2, JUNE 2005

ship to a particular behavioral state, or its interactions with inputsfrom another region’s activity. Our program of research in cog-nitive neuroscience has utilized the current generation of fNIRsystem to investigate brain activity, primarily in dorsolateral andinferior frontal cortex. To date, our fNIR studies of cognition andemotion have focused on functions associated with Brodman’sareas BA9, BA10, BA46, BA45, BA47, and BA44. Recent PETand fMRI studies have shown that these areas play a critical rolein sustained attention, both the short term storage and the execu-tive process components of working memory, episodic memory,problem solving, response inhibition and the perception of smell,(for a recent review, see [1] and [6]). In addition, word recogni-tion and the storage of verbal materials activate Broca’s area andleft hemisphere supplementary and premotor areas [1], [7], [8].To date, studies utilizing fNIR have shown results consistentwith fMRI and PET findings for working memory, sustained at-tention [4], and target categorization [9].

Three types of fNIR implementation have been developedby various investigators; time domain, frequency domain, andcontinuous wave (CW) spectroscopy systems. Like other neu-roimaging technologies, each of these systems has its own par-ticular strengths and limitations. Time domain systems, also re-ferred to as time-resolved spectroscopy (TRS), introduces ex-tremely short incident pulses of light to the tissue, and the tem-poral distribution of photons that carry the information abouttissue scattering and absorption is measured. Frequency domainsystems modulate the amplitude of the light source to frequen-cies on the order of tens to hundreds of megahertz. To charac-terize the optical properties of tissue, the amplitude decay andphase shift of the detected signal are measured with respect tothe incident [10]. In CW systems, light is applied to tissue atconstant amplitude. CW systems are limited to measuring theamplitude attenuation of the incident light [10]. At the sametime, CW systems possess a number of advantageous proper-ties that have resulted in wide use among researchers interestedin brain imaging relative to other NIR systems: it is minimallyintrusive and portable, affordable, and easy to engineer rela-tive to frequency and time domain systems [10], [11]. In par-ticular, CW systems hold enormous potential for clinical appli-cations that require quantitative measurements of hemodynamicchanges during brain activation under ambulant conditions. Ourresearch team has been developing a CW fNIR system that lendsitself to both portable and wireless designs for the monitoringof brain function under ecologically valid conditions.

In Section II, both designs are introduced. Common ap-proaches and innovations in fNIR signal analysis and pro-cessing are presented in Section III. Section IV includes asummary of the merits of fNIR neuroimaging and refers topotential deployment areas.

II. INSTRUMENTATION

The portable fNIR system used in our studies was originallydescribed by Chance et al. [11]. The main components of thesystem are: 1) the sensor that covers the entire forehead; 2) acontrol box for data acquisition; and 3) a computer for the dataanalysis software.

The wireless fNIR is a miniaturized system consisting of twoparts: 1) a wearable device to collect and transmit the data and 2)

Fig. 1. Wireless fNIR system.

Fig. 2. (a) Flexible Sensor (b) Participant wearing flexible sensor.

a data analysis and display computer. The graphical representa-tion of the system is given in Fig. 1. This wearable system con-sists of three main parts: 1) a personal digital assistant (PDA);2) a control circuit and battery holder; and 3) the sensor.

The PDA, a Hewlett Packard iPAQ Pocket PC, supports bothintegrated Wi-Fi (IEEE 802.11b standard) and Bluetooth forwireless communication. The customized PDA software is ableto control the circuitry, reading, saving, and sending of the datavia a wireless network. The image acquisition rate of the systemis two images per second.

The current flexible sensor used in the studies described inthis paper consists of four light-emitting diode (LED) lightsources and ten detectors, which is designed to image corticalareas underlying the forehead (dorsolateral and inferior frontalcortices). With a source-detector separation of 2.5 cm, thisconfiguration results in a total of 16 channels. Communicationbetween the data analysis computer and the task presentationcomputer is established via a serial port connection to time-lockfNIR measurement to the task events.

The flexible sensor is a modular design consisting of twoparts: a reusable, flexible circuit board that carries the neces-sary infrared sources and detectors, and a disposable, single-usecushioning material that serves to attach the sensor to the par-ticipant (see Fig. 2). The flexible circuit provides a reliable, in-tegrated wiring solution, as well as consistent and reproduciblecomponent spacing and alignment. Since the circuit board andcushioning materials are flexible, the components move andadapt to the contours of the participant’s head, allowing the

IZZETOGLU et al.: FUNCTIONAL NEAR-INFRARED NEUROIMAGING 155

sensor elements to maintain an orthogonal orientation to the skinsurface, improving light coupling efficiency and signal strength.

Future sensor designs include a modular sensor and a hair-brush sensor. The modular sensor will be scalable to any fore-head shape and size, including adult and infants allowing theadjustment of the sources and detectors according to the Inter-national 10–20 system. This sensor will also permit multiple dis-tance measurements from different layers of human brain. Thehairbrush sensor will have the sources and detectors mountedon a hairbrush which will be used to separate the hair such thatthe sources and detectors make intimate contact with the skinon hairy regions of the head. This design will allow fNIR mea-surements not only from the forehead but also from the hairyregions of the head.

III. FNIR SIGNAL PROCESSING

A. Preprocessing: Artifact Removal

The safe, portable, wearable, minimally intrusive and wire-less qualities of fNIR make it an ideal candidate for monitoringcortical function in the brain while subjects are engaged in var-ious real life or experimental tasks. However, the artifact causedby motion of the participant’s head places an important limita-tion on the use of optical data in these applications, [12]. Headmovement can cause the fNIR detectors to shift and lose con-tact with the skin, exposing them to: 1) ambient light; 2) lightemitted directly from the fNIR sources; or 3) light reflected fromthe skin, rather than being reflected from tissue in the cortex.This type of motion artifact, similar to “electrode pop” in EEG,is readily recognizable because it causes sudden, large spikes inthe fNIR data. A more subtle artifact of head movement is due tothe effects of gravity on the cerebral blood. Rapid head move-ment can cause the blood to move toward (or away from) thearea that is being monitored, rapidly increasing (or decreasing)blood volume with a concomitant skewing of the data. Since thedynamics of this type of motion artifact are slower than LED“pop,” they can be confused with the actual hemodynamic re-sponse due to brain activation. The motion artifacts in fNIRstudies are a serious problem for real life applications wherehead immobility is not feasible. Hence, removing motion arti-fact from fNIR data is an important and necessary step if fNIRis to be deployed as a brain monitoring technology in naturalenvironments.

Adaptive filtering is one approach to dealing with motion ar-tifact [12]. Adaptive filtering has been widely used for noise re-duction in other biomedical applications involving electrocar-diogram (ECG), EEG, and fNIR [13]–[15]. An adaptive filter isusually a finite impulse response (FIR) filter whose coefficientsor tap weights change in response to the changing input signal’scharacteristics. The filter has an adaptation algorithm that moni-tors the environment using additional sensors and hardware andvaries the filter transfer function accordingly [16].

We have implemented a novel approach, i.e., Wiener filtering,to remove motion artifact from fNIR data [17]. The Wiener fil-tering approach is advantageous because it eliminates the needto use additional sensors and extra wiring required for adaptivefiltering. Like adaptive filtering, Wiener filtering is an optimalfiltering technique in that minimization of the mean-square error

TABLE ICOMPARISON OF THE IMPROVEMENT IN SNR DURING SLOW, MEDIUM AND

FAST HEAD MOVEMENTS FOR WEINER VERSUS ADAPTIVE FILTERING

serves as the basis of its function. However, because it uses thestatistics derived from the signals elicited by its current appli-cation to estimate the filter coefficients, it does not require theinput of additional sensor information that adaptive filtering re-quires. If the distorted fNIR measurement signal is mod-eled as where is the true fNIR dataand is the motion artifact signal, assuming that the fNIRsignal and the motion artifact signals are stationary and uncor-related with each other, the Wiener filter in the Fourier domaincan be found as

(1)

where and are the power spectral densities (PSD)of and , respectively. Although there is no need forextra sensor measurements in this approach, andmust be estimated in order to calculate the Wiener filter. Formost practical applications as in our particular application, thePSD’s of the input and the noise are not available apriori and must be estimated from the measurements. For thatpurpose, in ecologically valid applications, prototypes for themotionless fNIR data and motion artifact should be collectedprior to the actual measurement.

In order to test the performances of adaptive and Wiener fil-tering approaches, we carried out a preliminary study with 11participants. Our protocol was composed of three types of 20 sof head movement regions (slow, medium, fast), where subjectis asked to move his/her head up and down continuously and 20s of rest regions in between the head movement regions, wheresubject is asked to stay still by a prompt on the computer screen[17]. This procedure is repeated two times in order to provideprototypes to compute the PSD estimates for Wiener filteringapproach. The extra sensor measurement required by the adap-tive filtering approach is gathered by using an accelerometer si-multaneously with the fNIR system.

We performed an SNR analysis to each of the algorithm re-sults to compare their performances [17]. The estimation SNRis calculated as ) where is the vari-ance of motionless fNIR data, , and is the variance of theestimation error, , where is the filtereddata. The input SNR is calculated as )where is the variance of motion artifact. Then, we obtained

for both the Wiener and adaptive fil-tering results in order to show the SNR improvement on theestimates. This preliminary study showed that in each condi-tion, slow, medium, and fast head movement, Weiner filteringresulted in greater increases in the than did adaptive fil-tering (t-test results are summarized in Table I).


Fig. 3. Typical Gamma-function.

B. Data Analysis and Feature Extraction

As in the case of fMRI, fNIR analysis presents a challengesince hemodynamic responses are slow (10–12 s) and may varyin timing and amplitude across brain regions, across, and withinsubjects, for similar regions and tasks [18], [19]. Several studiesmodeled the hemodynamic response in fMRI including event-related studies that use mathematical model fitting to Gaussian,Poisson, or Gamma-type functions, FIR filters, as well as sta-tistical methods such as Bayesian modeling [19]–[22]. Compa-rable studies modeling the hemodynamic response are rare inthe fNIR literature [23].

It has been shown that the hemodynamic response evolvesover a 10–12 s period, even if the eliciting stimulus itself isrelatively brief (on the order of milliseconds) [19]–[22]. Whenstimuli are presented with sufficient temporal separation, thecorresponding hemodynamic response to each stimulus can beobtained and modeled as a gamma function. However, in mostcognitive studies, stimuli are presented much more rapidlythan every 10–12 s. Although the slow nature of the hemo-dynamic response presents a challenge to event-related dataanalysis, individual trials that overlap within the time con-stant of the hemodynamic response function can be separated.The basis of separation is the finding that sequential hemody-namic events can be modeled by summation of responses in a“roughly” linear fashion [19]. To our knowledge, Izzetoglu etal. [23] presented the first event-related fNIR study in whichindividual event-related hemodynamic responses were mod-eled such that physiologically meaningful features could beextracted for classification purposes. Each event-related hemo-dynamic response was estimated on the basis of two postulates:1) that each single-trial hemodynamic response has the formof a gamma function as given in Fig. 3;and 2) that the total oxygenation data can be modeled by thesummation of individual hemodynamic responses evoked byrapidly presented stimuli . Each single trialwas estimated by optimizing the error between the total oxy-genation data from fNIR measurements and the linear model:

.Using this approach, no further assumptions were necessary

with respect to the amplitude, time parameters, or equivalenceof the hemodynamic responses, which can vary in studies ofcognitive/emotional function. Physiologically relevant features,hypothesized to classify various aspects of cognition such astask load or performance were extracted from the data (e.g.,average amplitude, maximum amplitude, time to peak (or “risetime”), full-width at half-maximum).

The resulting algorithm was tested using a linguistic problemsolving study of graded difficulty (solving anagrams) [24]. Ana-grams were presented on a computer screen in 1-min blocks;each block consisted of either three-, four-, or five-letter ana-grams, with a 30-s rest period between blocks. Each participant

received the blocks in the same order; three-letter,four-letter, five-letter, four-letter and three-letter anagrams. Par-ticipants were instructed to press one response key when theyhad solved the anagram and a second key when they felt the ana-gram was unsolvable or to skip the anagram. The participant’sresponse and response time were recorded for each anagram.

The average of the event-related hemodynamic rise time forleft hemisphere are plotted with the behavioral reaction timesfor three-, four-, and five-letter anagram sets in Fig. 4(a); max-imum amplitudes of the hemodynamic responses are plotted inFig. 4(b). Oxygenation rise times clearly follow the same pat-tern as the behavioral response times. Fig. 4(c) presents a scatterplot of the oxygenation rise time versus reaction time (Pearson’s

). The rise time and maximum amplitude of oxygena-tion both increase as the difficulty level of the anagram solutionsincrease and these brain processes are reflected in longer behav-ioral reaction times [24]. Estimation of the event-related signalsin a block design allows more precise analysis of the brain’sfunction during a cognitive/problem solving task.

C. Data Visualization

For the purpose of data visualization and reporting, we havecreated a topographic mapping tool that is based on the currentsensor geometry (see Fig. 1). The software allows images to begenerated for every time point in the processed data. The activa-tion pattern has been projected onto a brain template (templateused by permission [25]) based on fMRI data indicating thebrain regions underlying the International 10–20 System [26].This registration allows the comparison of fNIR data to otherneuroimaging modalities such as fMRI and PET.

The data imaging software is demonstrated here using the re-sults of a target categorization study [9]. Participantscompleted a standard visual oddball task, i.e., differentiating X’sfrom O’s, modified for use with an event-related hemodynamicresponse study (see [27] for details). Hemodynamic responsesto the target and context stimuli were epoched, scanned for ar-tifacts and outliers, averaged, and baseline corrected. (Note: Inthe current application, we relied on a standard outlier elimina-tion algorithm, removing trials that contained artifacts resultingin average responses 2.5 standard deviation from the mean.We are currently working more sophisticated outlier eliminationalgorithms using clustering algorithms.) The averaged hemody-namic responses to target (X) and context (O) stimuli were an-alyzed for every 1200-ms epoch of the post-stimulus windowusing repeated-measures MANOVAs. Increases in oxygenationwere greater for targets than for context stimuli for voxel 11, lo-cated over the middle frontal gyrus of the right hemisphere from6–9 s post-stimulus 0.015 . These results are in agree-ment with previous fMRI studies relative to both time courseand activation location [27]. Fig. 5 presents these results mappedonto a frontal view of the corresponding brain regions. The vi-sualization toolbox also allows lateral views of left and righthemispheres.


Fig. 4. Subject averages of (a) rise and response times and (b) maximum amplitude. (c) Scatter plot of rise versus response time averages for all anagram sets ofall subjects.

IV. DISCUSSION

The use of fNIR spectroscopy, a wearable, negligibly intru-sive optical imaging modality, has been increasing as a methodwith which to measure hemodynamic changes in the cortex. ThefNIR instrumentation allows for safe, portable, and low-costcortical monitoring that can be applied in ecologically valid en-vironments and is readily combined with other physiologicalmeasures. Current state-of-the-art technologies being developedat Drexel University and the University of Pennsylvania includeboth portable and wireless fNIR systems.

In addition to hardware and sensor development, a numberof important signal-processing procedures have been imple-mented to increase functionality for the end-user. To increase thesignal-to-noise ratio of the fNIR system in the extraction of the

hemodynamic response during cognitive and emotional tasks,signal processing algorithms have been developed to identify,eliminate and/or to compensate for noise. An adaptive filteringtechnique and a Wiener filtering approach have both been usedto successfully remove motion artifacts from the fNIR data. Ahemodynamic response estimation algorithm for event-relateddesigns has also been developed. The algorithm models thehemodynamic response as a gamma function and utilizes theestimated model parameters to identify physiologically relevantfeatures. Imaging software then allows the topographical map-ping of extracted parameters onto the corresponding corticalregions. Wearable instrumentation combined with data analytictools and imaging/presentation algorithms make fNIR suitablefor the study of cognition under ambulant conditions.


Fig. 5. fNIR evidence that targets elicit greater activation of middle frontal gyrus than context stimuli during an event-related target categorization task.

Functional NIR monitoring is gaining acceptance in surgery.The next generation of portable and wireless fNIR systemsequipped with modular hairbrush sensors and efficient, ac-curate, and user friendly signal processing algorithms andvisualization tools for monitoring hemodynamic response re-lated to cognitive activity, are poised for deployment in severalnew application areas including mental health, neurological dis-eases, neurorehabilitation, pediatric, and emergency medicine.Education and training applications are also promising. Mem-bers of our team are currently conducting pilot studies inautism, schizophrenia, and education.

REFERENCES

[1] R. Cabeza and L. Nyberg, “Imaging cognition II: An empirical reviewof 275 PET and fMRI studies,” J. Cogn. Neurosci., vol. 12, pp. 1–47,2000.

[2] R. J. Davidson and S. K. Sutton, “Affective neuroscience: The emer-gence of a discipline,” Curr. Opi. Neurobiol., vol. 5, pp. 217–224,1998.

[3] A. Villringer and B. Chance, “Noninvasive optical spectroscopy andimaging of human brain function,” Trends Neurosci., vol. 20, pp.435–442, 1997.

[4] K. Izzetoglu, S. Bunce, B. Onaral, K. Pourrezaei, and B. Chance, “Func-tional optical brain imaging using near-infrared during cognitive tasks,”Int. J. Human-Comp. Int., to be published.

[5] T. Suto, M. Ito, T. Uehara, I. Ida, M. Fukuda, and M. Mikuni, “Tem-poral characteristics of cerebral blood volume change in motor andsomatosensory cortices revealed by multichannel near infrared spec-troscopy,” Int. Congress Series, vol. 1232, pp. 383–388, 2002.

[6] E. E. Smith and J. Jonides, “Working memory: A view from neu-roimaging,” Cogn. Psychol., vol. 33, pp. 5–42, 1997.

[7] , “Storage and executive processes in the frontal lobes,” Science,vol. 283, pp. 1657–1661, 1999.

[8] T. S. Braver, J. D. Cohen, L. E. Nystrom, J. Jonides, E. E. Smith, and D.C. Noll, “A parametric study of prefrontal cortex involvement in humanworking memory,” NeuroImage, vol. 5, pp. 49–62, 1997.

[9] S. C. Bunce, G. McConnell, K. Izzetoglu, M. Izzetoglu, and B. Onaral,“The Integration of event related fNIR spectroscopy and event-relatedpotentials during a target categorization task,” Drexel Univ. Coll.Medicine, Philadelphia, PA, unpublished manuscript.

[10] D. A. Boas, M. A. Franceschini, A. K. Dunn, and G. Strangman, “Non-invasive imaging of cerebral activation with diffuse optical tomography,”in In-Vivo Optical Imaging of Brain Function. Boca Raton, FL: CRC,2002, pp. 193–221.

[11] B. Chance, E. Anday, S. Nioka, S. Zhou, L. Hong, K. Worden, C. Li, T.Murray, Y. Ovetsky, D. Pidikiti, and R. Thomas, “A novel method forfast imaging of brain function, noninvasively, with light,” Optics Exp.,vol. 2, no. 10, pp. 411–423, 1998.

[12] M. Izzetoglu, K. Izzetoglu, S. Bunce, and B. Onaral, “Motion artifactremoval in fNIR signals using adaptive filtering,” in Proc. IEEE BMES,2003.

[13] B. Widrow et al., “Adaptive noise canceling: Principles and applica-tions,” Proc. IEEE, vol. 63, no. 12, pp. 1692–1716, Dec. 1975.

[14] N. V. Thakor and Y. S. Zhu, “Applications of adaptive filtering to ECGanalysis: Noise cancellation and arrhythmia detection,” IEEE Trans.Biomed. Eng., vol. 38, no. 8, pp. 785–794, Aug. 1991.

[15] N. V. Thakor, “Adaptive filtering of evoked potentials,” IEEE Trans.Biomed. Eng., vol. 34, no. 1, pp. 6–12, Jan. 1987.

[16] S. Haykin, Adaptive Filter Theory. Englewood Cliffs, NJ: Prentice-Hall, 2001.

[17] M. Izzetoglu, A. Devaraj, S. Bunce, and B. Onaral, “Motion artifactcancellation in fNIR spectroscopy using wiener filtering,” IEEE Trans.Biomed. Eng., to be published.

[18] D. R. Gitelman, D. William, W. D. Penny, J. Ashburner, and K. J. Friston,“Modeling regional and psychophysiologic interactions in fMRI: Theimportance of hemodynamic deconvolution,” NeuroImage, vol. 19, pp.200–207, 2003.


[19] F. M. Miezin, L. Maccotta, J. M. Ollinger, S. E. Petersen, and R. L.Buckner, “Characterizing the hemodynamic response: Effects of pre-sentation rate, sampling procedure, and the possibility of ordering brainactivity based on relative timing,” NeuroImage, vol. 11, pp. 735–759,2000.

[20] F. Kruggel and D. Y. Von Cramon, “Modeling the hemodynamic re-sponse in single trial functional MRI experiments,” Magn. ResonanceImag., vol. 42, pp. 787–797, 1999.

[21] C. Gossl, L. Fahrmeir, and D. P. Auer, “Bayesian modeling of the hemo-dynamic response function in BOLD fMRI,” NeuroImage, vol. 14, pp.140–148, 2001.

[22] C. Goutte, F. A. Nielsen, and L. K. Hansen, “Modeling the haemody-namic response in fMRI using smooth FIR filters,” IEEE Trans. Med.Imag., vol. 19, no. 12, pp. 1188–1201, Dec. 2000.

[23] M. Izzetoglu, S. Bunce, and B. Onaral, “Single trial hemodynamic re-sponse estimation in event related fNIR spectroscopy,” in Proc. OpticalSoc. America, Miami, FL, 2004.

[24] M. Izzetoglu, S. Nioka, B. Chance, and B. Onaral, “Single trial hemo-dynamic response estimation in a block anagram solution study usingfNIR spectroscopy,” in Proc. ICASSP, Philadelphia, PA, 2005. acceptedfor publication.

[25] S. W. Bradley, K. Eno, J. Prothero, and J. Brinkley, Digital AnatomistProject. Seattle, WA: Dept. Biological Structure, Univ. Washington,1997. Brain graphic used by permission from the Digital AnatomistProject.

[26] M. Okamoto, H. Dan, K. Sakamoto, K. Takeo, K. Shimizu, S. Kohno, I.Oda, S. Isobe, T. Suzuki, K. Kohyama, and I. Dan, “Three-dimensionalprobabilistic anatomical cranio-cerebral correlation via the international10–20 system oriented for transcranial functional brain mapping,” Neu-roImage, vol. 21, pp. 99–111, 2004.

[27] G. McCarthy, M. Luby, J. Gore, and P. Goldman-Rakic, “Infrequentevents transiently activate human prefrontal and parietal cortex as mea-sured by functional MRI,” J. Neurophysiol., vol. 77, pp. 1630–1634,1997.

Meltem Izzetoglu was born in Samsun, Turkey in1971. She received the B.S. and the M.S. degrees bothin electrical and electronics engineering from MiddleEast Technical University, Ankara, Turkey in 1992and 1995, respectively. She received the Ph.D. degreein electrical and computer engineering from DrexelUniversity, Philadelphia, PA in 2002.

She is currently a Postdoctoral Fellow with theSchool of Biomedical Engineering, Science, andHealth Systems, Drexel University. Her research in-terests include biomedical signal analysis, adaptive,

and optimal signal processing, bio-optics, scale-space processing tools.

Kurtulus Izzetoglu received the M.S.E.E. degreefrom Middle East Technical University, Ankara,Turkey.

He gained his professional software developmentand medical imaging experience as a member ofconsulting companies in the United States and TheNetherlands, respectively. In these positions, heworked as a Senior Analyst as well as a Software andAnalytical Applications Developer. His experiencesinclude the development of professional medicalimaging software package, implementation of quan-

titative analysis and imaging techniques. Subsequent to five years of industrialexperience, he joined the Functional Optical Imaging Research Team at DrexelUniversity, Philadelphia, PA, where he currently serves as the Project Engineer.His technical management responsibilities include development of cognitiveworkload assessment testing and analysis platform, signal processing, andexperimental protocol design and implementation.

Scott Bunce, photograph and biography not available at the time of publication.

Hasan Ayaz, photograph and biography not available at the time of publication.

Ajit Devaraj, photograph and biography not available at the time of publication.

Banu Onaral (S’76–M’78–SM’89–F’93) receivedthe B.S.E.E. and M.S.E.E. degrees from BogaziciUniversity, Istanbul, Turkey, and the Ph.D. degreefrom the University of Pennsylvania, Philadelphia,in 1978.

Her academic focus both in research and teachingis centered on biomedical signals and systemsengineering. She has been a founding member ofthe Biomedical Information Technology Laboratory,Scaling Signals and Systems Laboratory and theBio-Electrode Research Laboratory, Drexel Univer-

sity, Philadelpha, where she is currently the H. H. Sun Professor of Biomedicaland Electrical Engineering. She has lead several curriculum developmentinitiatives including the undergraduate telecommunication and biomedicalengineering programs. Her professional services include Chair and member-ship on advisory boards and strategic planning bodies of several universitiesand funding agencies, including service on the National Science FoundationEngineering Advisory Board (1997–1999), and on the proposal review panelsand study sections. Her editorial responsibilities have included service on theEditorial Board of journals and the CRC Biomedical Engineering Handbook asSection Editor for Biomedical Signal Analysis.

Dr. Onaral has been active in professional society leadership, in particularnational and international technical meeting organization. She served asVice President for Conferences and as President of the IEEE Engineering inMedicine and Biology. She also served on the inaugural Board of the AmericanInstitute for Medical and Biological Engineering. She has developed severalsignals and systems engineering software products and was recognized bythe EDUCOM/NCRIPTAL Best Educational Tool Award. She is the recip-ient of a number of faculty excellence awards including the 1990 LindbackDistinguished Teaching Award of Drexel University. She is a Fellow of theIEEE Engineering in Medicine and Biology Society; Founding Fellow of theAmerican Institute for Medical and Biological Engineering; Fellow of theAmerican Association for the Advancement of Science; Senior Member ofthe Society of Women Engineers; and Member of the American Society forEngineering Education and the Scientific Research Society Sigma Xi. She isListed in the Who’s Who in the East, Who’s Who in Science and Engineering,and Who’s Who in the World.

Kambiz Pourrezaei received the B.S. degree fromTehran University, Tehran, Iran, in 1971, the M.S. de-gree from Tufts University, Medford, MA, in 1976,and the Ph.D. degree from Rensselaer Polytechnic In-stitute, Troy, NY, in 1982.

He is a Professor with the School of Biomed-ical Engineering, Science, and Health Systems,Drexel University, Philadelphia, PA. Currently, heis the Co-Director of the Nanotechnology Institute,Philadelphia. He has active research programs in theareas of bio-nanotechnology and bio-optics.

Documents

Functional Near-Infrared Neuroimaging