OSA's Digital Library

Biomedical Optics Express

Biomedical Optics Express

  • Editor: Joseph A. Izatt
  • Vol. 3, Iss. 1 — Jan. 1, 2012
  • pp: 64–74
« Show journal navigation

Independent component analysis of broadband near-infrared spectroscopy data acquired on adult human head

Irina Schelkanova and Vladislav Toronov  »View Author Affiliations


Biomedical Optics Express, Vol. 3, Issue 1, pp. 64-74 (2012)
http://dx.doi.org/10.1364/BOE.3.000064


View Full Text Article

Acrobat PDF (962 KB)





Browse Journals / Lookup Meetings

Browse by Journal and Year


   


Lookup Conference Papers

Close Browse Journals / Lookup Meetings

Article Tools

Share
Citations

Abstract

The goal of this study was to investigate the ability of independent component analysis in the time-spectral domain to isolate physiological sources of functional near infrared spectroscopy signals. We apply independent component analysis to the broadband fNIRS data acquired on the human forehead at 650 different wavelengths between 700 nm and 950 nm. To induce cerebral oxygenation changes we use the breath holding paradigm. We found one major independent component during baseline and two major components during exercise. Each independent component corresponds to one oxy-hemoglobin and one deoxy-hemoglobin time courses. The corresponding characteristic spectra of changes in optical absorption suggested that one component represented vasodilation of cerebral arterioles while the delayed component represented the washout of deoxyhemoglobin either in cerebral capillaries and venules or in extra cerebral tissue. We found that both broadband and isolated wavelength data can produce similar independent components.

© 2011 OSA

1. Introduction

Functional near infrared spectroscopy (fNIRS) aims at non-invasively measuring the effects of the brain tissue on near-infrared light propagating through the intact human head [1

1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

,2

2. F. Irani, S. M. Platek, S. Bunce, A. C. Ruocco, and D. Chute, “Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging technology with important applications for the study of brain disorders,” Clin. Neuropsychol. 21(1), 9–37 (2007). [CrossRef] [PubMed]

]. Various fNIRS systems use continuous-wave (CW), frequency-domain and time-resolved sources of light [1

1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

]. Although the time-resolved and frequency-domain systems have capabilities of separating absorption and scattering properties of tissues, they are more expensive and have lower signal-to-noise ratio. Therefore, CW fNIRS is considered to be most practical. CW systems use either broadband [3

3. I. Schelkanova and V. Toronov, “Optimal quantitation of the cerebral hemodynamic response in functional near-infrared spectroscopy,” Opt. Express 18(18), 19386–19395 (2010). [CrossRef] [PubMed]

,4

4. H. Dehghani, F. Leblond, B. W. Pogue, and F. Chauchard, “Application of spectral derivative data in visible and near-infrared spectroscopy,” Phys. Med. Biol. 55(12), 3381–3399 (2010). [CrossRef] [PubMed]

] or selected wavelength light sources [5

5. J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt. 14(5), 054032 (2009). [CrossRef] [PubMed]

]. While both types of sources have their advantages, a number of recent studies demonstrated benefits of broadband systems [3

3. I. Schelkanova and V. Toronov, “Optimal quantitation of the cerebral hemodynamic response in functional near-infrared spectroscopy,” Opt. Express 18(18), 19386–19395 (2010). [CrossRef] [PubMed]

,4

4. H. Dehghani, F. Leblond, B. W. Pogue, and F. Chauchard, “Application of spectral derivative data in visible and near-infrared spectroscopy,” Phys. Med. Biol. 55(12), 3381–3399 (2010). [CrossRef] [PubMed]

]. A drawback of all types of fNIRS systems is that the light integrates signals from all types of small blood vessels: arterioles, capillaries and venules [1

1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

,2

2. F. Irani, S. M. Platek, S. Bunce, A. C. Ruocco, and D. Chute, “Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging technology with important applications for the study of brain disorders,” Clin. Neuropsychol. 21(1), 9–37 (2007). [CrossRef] [PubMed]

]. Also, besides the brain the light travels through all extracerebral head tissues, whose vasomotion biases functional signals. Therefore the separation of functional signals from contamination by systemic and local physiological fluctuations is an important research topic. This problem was approached by using various signal processing methods, including blind signal separation techniques [5

5. J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt. 14(5), 054032 (2009). [CrossRef] [PubMed]

8

8. F. C. Robertson, T. S. Douglas, and E. M. Meintjes, “Motion artifact removal for functional near infrared spectroscopy: a comparison of methods,” IEEE Trans. Biomed. Eng. 57(6), 1377–1387 (2010). [CrossRef] [PubMed]

]. In particular, principal component and independent component analyses (PCA and ICA, respectively) were applied separately either to the oxy- or deoxyhemoglobin time series reconstructed from the bi-wavelength NIRS data acquired on the human head during repetitive hyper- or hypocapnia episodes [5

5. J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt. 14(5), 054032 (2009). [CrossRef] [PubMed]

] with the goal to filter out “surface effects”. The ICA criterion for the “surface effects” was the value of the coefficient of spatial uniformity (CSU) calculated using elements of the spatial mixing matrix [7

7. S. Kohno, I. Miyai, A. Seiyama, I. Oda, A. Ishikawa, S. Tsuneishi, T. Amita, and K. Shimizu, “Removal of the skin blood flow artifact in functional near-infrared spectroscopic imaging data through independent component analysis,” J. Biomed. Opt. 12(6), 062111 (2007). [CrossRef] [PubMed]

]. In [5

5. J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt. 14(5), 054032 (2009). [CrossRef] [PubMed]

] the independent components with higher CSU were excluded and remaining components were recombined into the filtered oxy- and deoxyhemoglobin traces. ICA was also used in NIRS for the motion artifact removal [7

7. S. Kohno, I. Miyai, A. Seiyama, I. Oda, A. Ishikawa, S. Tsuneishi, T. Amita, and K. Shimizu, “Removal of the skin blood flow artifact in functional near-infrared spectroscopic imaging data through independent component analysis,” J. Biomed. Opt. 12(6), 062111 (2007). [CrossRef] [PubMed]

9

9. A. V. Medvedev, J. M. Kainerstorfer, S. V. Borisov, A. H. Gandjbakhche, and J. Vanmeter, ““Seeing” electroencephalogram through the skull: imaging prefrontal cortex with fast optical signal,” J. Biomed. Opt. 15(6), 061702 (2010). [CrossRef] [PubMed]

], and to study the functional cerebral connectivity of the resting state [10

10. H. Zhang, Y. J. Zhang, C. M. Lu, S. Y. Ma, Y. F. Zang, and C. Z. Zhu, “Functional connectivity as revealed by independent component analysis of resting-state fNIRS measurements,” Neuroimage 51(3), 1150–1161 (2010). [CrossRef] [PubMed]

].

Previously fNIRS ICA was performed only in the time-spatial domain on data acquired either at multiple source-detector separations or at different locations [6

6. A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis (Wiley, New York, 2001).

10

10. H. Zhang, Y. J. Zhang, C. M. Lu, S. Y. Ma, Y. F. Zang, and C. Z. Zhu, “Functional connectivity as revealed by independent component analysis of resting-state fNIRS measurements,” Neuroimage 51(3), 1150–1161 (2010). [CrossRef] [PubMed]

]. The goal of our present study was to investigate fNIRS independent components not in the time-spatial but in the time-spectral domain. In particular we wanted to understand relationships between the spectral independent components and the changes in the oxy- and deoxyhemoglobin concentrations. We show that the rows of the time-spectral ICA mixing matrix allow for a physiological interpretation in terms of the oxy- and deoxyhemoglobin concentrations. We apply a well-established Matlab version of ICA FastICA [11

11. C. Hesse and C. James, “The fastica algorithm with spatial constraints,” IEEE Signal Process. Lett. 12(11), 792–795 (2005). [CrossRef]

] to the broadband NIRS data acquired at one location on the human forehead at a single source-detector separation but at 650 different wavelengths between 700 nm and 950 nm. To induce cerebral oxygenation changes the breath holding (BH) paradigm was used because it generated blood carbon dioxide (CO2) [12

12. T. Len, J. Neary, G. Asmundson, D. Goodman, B. Bjornson, and Y. Bhambhani, “Cerebrovascular reactivity impairment after sport-induced concussion,” Med. Sci. Sports Exerc. 43(12), 2241–2248 (2011). [CrossRef]

]. Vasodilation induced by CO2 caused increased cerebral blood flow which resulted in changes in cerebral blood volume and oxygenation [13

13. R. L. Grubb Jr, M. E. Raichle, J. O. Eichling, and M. M. Ter-Pogossian, “The effects of changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean transit time,” Stroke 5(5), 630–639 (1974). [CrossRef] [PubMed]

]. Although the individual wavelength signals were quite diverse and noisy, we found very few independent components whose origin can be interpreted based on their characteristic spectral signatures and time courses.

2. Materials and methods

2.1. Participants and experimental paradigm

2.2. Data acquisition

Near-infrared light was generated by a stabilized fan-cooled AvaLight-HAL Tungsten Halogen Light Source (Avantes Inc., Broomfield, CO) with an adjustable focusing connector to maximize light coupling with the source fiber. The source fiber bundle was made of 30 Thorlabs broadband silica 400 μm core diameter fibers. On the probe the source fibers were

arranged circularly around the location of the detector bundle at a radius of 25 mm (Fig. 1
Fig. 1 The arrangement of source and detector fibers.
). Such geometry allowed for reducing the effects of the scalp’s large blood vessels occasionally located near a light source. Light was collected using a 3-mm diameter fiber optic bundle (Sunoptic Technologies, FL). The detector bundle was connected to a QEB65000 cooled spectrometer (Ocean Optics, Dunedin, FL), which had 1044 wavelength channels in a spectral range between 650 and 1100 nm. The spectrometer output was digitized using the Spectral Suite software (Ocean Optics, Dunedin, FL). The optical probe was positioned on the left side of the forehead near the hairline to avoid sinuses. Spectra were acquired at the sampling rate of one spectrum per second resulting in 360 time points per data set. This sampling rate was selected to ensure that the instrumental noise did not affect physiological data.

2.3. Data processing: independent component analysis and general linear modeling

For 630 signals at different near-infrared wavelengths between 700 nm and 950 nm the absorbance A(λ,t) was calculated as

A(λ,t)=ln(Signal(λ,t)Dark(λ)reference(λ)Dark(λ)),
(1)

where Dark was the dark signal, reference was the signal measured directly through air (without tissue), and λ was the optical wavelength. Each instantaneous spectrum was smoothed using the Matlab smooth function at 15 points. No temporal smoothing was performed. The 630 x 360 matrix of 630 detrended time series of absorbance changes ΔA(λ,t)was fed into the FastICA algorithm [8

8. F. C. Robertson, T. S. Douglas, and E. M. Meintjes, “Motion artifact removal for functional near infrared spectroscopy: a comparison of methods,” IEEE Trans. Biomed. Eng. 57(6), 1377–1387 (2010). [CrossRef] [PubMed]

], which used fpica Matlab function to separate the input into the N independent components (ICs):

ΔA(λ,t)=T(t)×S(λ)+R(λ,t)
(2)

where T(t)was the 360 x N matrix of statistically independent temporal components (ICs) [3

3. I. Schelkanova and V. Toronov, “Optimal quantitation of the cerebral hemodynamic response in functional near-infrared spectroscopy,” Opt. Express 18(18), 19386–19395 (2010). [CrossRef] [PubMed]

], S(λ)was the N x 630 mixing matrix, and R(λ,t)was the matrix of residuals. The columns of T(t) were the time courses of ICs and rows of S(λ)were the time-independent amplitude spectra of the corresponding components. FastICA automatically normalized the time course and the spectrum of each component such that the standard deviation (STD) of the time course Ti(t) was equal to one, so it had arbitrary normalized units, and the corresponding spectrum Si(λ) had units of the optical density. By choosing different numbers of ICs we found that two components always covered more than 90% of non-zero eigenvalues of the covariance matrix, and at the same conditions the norm of R(λ,t)was much smaller than the norm of ΔA(λ,t) at each time moment. If the number of ICs N was set greater than two, the output of FastICA showed one or two non-noisy ICs (such as red traces in Fig. 1 (a,c) whose spectra spanned much greater ranges of values than the spectra of other ICs. ICs whose time courses appeared noisy in Figs. 1(a) and 1(c) also had noisy spectra (blue, green and black curves in Fig. 1(b) and blue and green curves in Fig. 1(d)). Therefore we focused on the analysis of only two components spanning the greatest ranges of Si(λ)values. In this approximation and for each moment of time Eq. (2) reduced to

ΔA(λ,t)=T1(t)S1(λ)+T2(t)S2(λ)
(3)

where T1(t)and T2(t)were the instantaneous values of the temporal independent components, and S1(λ) and S2(λ)were the corresponding time-independent amplitude spectra. One should note that according to Eq. (3) the signs of T1,2(t) and S1,2(λ) were arbitrary up to their simultaneous conjoined inversion. In order to homogenize the cross-subject analysis we fixated the signs of S1(λ)and S2(λ)as they appear in Fig. 1(d). This also fixated the signs of T1(t) and T2(t). We assumed that, apart from the noise, the dynamic of ΔA(λ,t) was only due to changes in the concentrations of oxy-hemoglobin (HbO2) and deoxy-hemoglobin (HHb), and that it follows the modified Lambert-Beers law [1

1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

]:

Δμa(λ,t)=ΔA(λ,t)/(rDPF(λ))
(4)

where DPF(λ)was the differential pathlength factor as a function of wavelength and r was the source detector distance. DPF(λ) was calculated as r1dA(λ)/dμa(λ) [15

15. A. Sassaroli and S. Fantini, “Comment on the modified Beer-Lambert law for scattering media,” Phys. Med. Biol. 49(14), N255–N257 (2004). [CrossRef] [PubMed]

] using the baseline absorbanceA(λ). The baseline absorption coefficient was calculated as

μa(λ)=CHBO2εHBO2(λ)+CHHbεHHb(λ)+CH2OεH2O(λ)
(5)

where CX was the concentration and εX(λ) was the extinction coefficient of the chromophore X. The values of chromophore concentrations were taken from [16

16. W. B. Gratzer, Medical Research Council Laboratories, Holly Hill, London, personal communication.

] and [17

17. N. Kollias, Wellman Laboratories, Harvard Medical School, Boston, MA, personal communication.

]. The concentrations of HbO2, HHb, and water were assumed 55μM(micromolars), 20 μM, and 85%, respectively in accordance with values known from [18

18. H. Q. Woodard and D. R. White, “The composition of body tissues,” Br. J. Radiol. 59(708), 1209–1218 (1986). [CrossRef] [PubMed]

] for water and from [1

1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

,19

19. J. H. Choi, M. Wolf, V. Toronov, U. Wolf, C. Polzonetti, D. Hueber, L. P. Safonova, R. Gupta, A. Michalos, W. Mantulin, and E. Gratton, “Noninvasive determination of the optical properties of adult brain: near-infrared spectroscopy approach,” J. Biomed. Opt. 9(1), 221–229 (2004). [CrossRef] [PubMed]

] for hemoglobin.

In order to analyze the contributions of HbO2 and HHb changes to each IC, and in accordance with Eqs. (3) and (4), we performed the generalized linear model (GLM) fit (glmfit function of MATLAB) of S1(λ)/(rDPF(λ))andS2(λ)/(rDPF(λ)):

S1,2(λ)/(rDPF(λ))=ΨHBO21,2εHBO2(λ)+ΨHHb1,2εHHb(λ)
(6)

where εHBO2(λ)and εHHb(λ) were the regressors of GLM, ΨHBO2and ΨHHb were the parameters of the fit, and the upper indexes indicated particular ICs. It followed from Eqs. (3), (4), and (6) that for each of two ICs the changes in oxy- and deoxyhemoglobin concentrations were

ΔHbO21,2(t)=T1,2(t)ΨHBO21,2 and ΔHHb1,2(t)=T1,2(t)ΨHHb1,2
(7)

respectively. Note that according to Eqs. (7) for the given k both ΔHbO2k(t)and ΔHHb(t)k were synchronous in time with their time course given by Tk(t). Since STDs of T1,2(t)were equal to one, the coefficients ΨHBO2and ΨHHb for each IC provided estimates of the mean magnitudes of changes in HbO2and HHb, respectively. The full HbO2 and HHb changes were obtained by adding partial changes corresponding to both components:

ΔHbO2(t)=ΔHbO21(t)+ΔHbO22(t) and ΔHHb(t)=ΔHHb1(t)+ΔHHb2(t)
(8)

3. Results

Figure 2
Fig. 2 Temporal (left) and spectral (right) components during rest (a and b) and breath hold (c and d) conditions. Respective spectral and temporal components are shown by the same colors. Encircled is the IC2 spectral feature near the peak wavelength of the deoxy-hemoglobin absorption.
showed examples of the time courses (2a and 2c) and the corresponding spectra (2b and 2d) during rest and exercise. The main differences between rest and exercise were the number of physiological ICs (one during rest and two during exercise) and the time course of ICs (chaotic during rest and showing responses to BH during exercise). Since the shapes of the non-noisy spectra shown in Figs. 2(b,d) were qualitatively similar in all subjects we referred to the corresponding independent components as IC1 and as IC2. The only physiological component showing non-periodic baseline hemodynamic fluctuations during rest was IC1(identified based on the shape of its spectrumS1). The shape of S1 (the spectrum of IC1) was topologically similar for all subjects but in some cases quantitatively changed considerably between rest and exercise. IC2 appeared only during exercises. The topological differences between S1 and S2 were shown in Fig. 2(d) (red and black curves, respectively).

Since ΨHBO21was much greater thanΨHHb1, the contribution of the deoxy-hemoglobin changes to IC1 was much smaller than the contribution of the oxy-hemoglobin changes. For IC2 on average both oxy- and deoxy-hemoglobin changes were approximately equal by magnitude and, on average, opposite by sign. Figure 4
Fig. 4 (a) Time courses of IC1 for three different subjects during BH. (b-d) time courses of IC1 (red curve) and IC2 (blue curve) for three sample subjects. The components were extracted from the broadband data sets. Vertical bars showed BH periods. Note that the time courses were normalized to have their STDs equal to one.
shows examples of the time courses of IC1 and IC2 for three subjects during the repetitive BH exercises. In all panels of Fig. 3 the increases in T1(t)corresponded to increases in ΔHbO21(t). The direction of change in ΔHHb1(t)depended on the sign of ΨHHb1in Table 1. The increases in T2(t) in Figs. 3 (b-d) corresponded to the increases in ΔHHb2(t)concentration and to the decreases in ΔHbO22(t). One can see that for different subjects the IC1 responses to BHs (Fig. 4 (a)) were all quite synchronous. The IC2 responses were always delayed compared to the IC1 responses (Figs. 4 (b, c, d). Delays between the peaks of IC1 and IC2 were different for different subjects, varying from 4s to 15s. We have verified that the correlation coefficient between T1(t)and T2(t)was always zero.

Figure 5
Fig. 5 The full ΔHbO2 (t) and ΔHHb(t) traces computed according to Eqs. (8), and the total-hemoglobin changes. T1(t) and T2(t) are also shown by dashed and dotted lines, respectively.
shows the full ΔHbO2(t) and ΔHHb(t)traces computed according to Eqs. (8), and the total-hemoglobin changes ΔtHb(t)=ΔHbO2(t)+ΔHHb(t)for the same data set as Fig. 4b. The scaled T1(t)and T2(t) were also shown. The figure demonstrates that ΔHbO2(t)and ΔtHb(t)were well correlated with T1(t), while ΔHHb(t)was better correlated with T2(t). However, these features did not necessarily mean that IC1 represented mainly ΔHbO2(t) in the brain and IC2 represented the cerebral ΔHHb(t). If the actual sources of IC1 and IC2 were located in different compartments of the vascular bed, the similarities of ΔHbO2(t) to T1(t) and of ΔHHb(t) to T2(t) in Fig. 6
Fig. 6 Comparison of temporal components IC1 (a) and IC2 (b), obtained from broadband data (solid line) and using two narrow wavebands n 760 and 900 nm (dashed line).
could result from the purely numeric fact that the values of ΨHBO21were much greater than all other coefficients and conjointly the values of ΨHHb1were much smaller than all other coefficients in Table 1.

In order to investigate whether T1(t) and T2(t) could be obtained using a small number of wavelengths, we created five signals by averaging ten adjacent wavelength signals around each of 760 nm, 800 nm, 830 nm, 860 nm, and 900 nm wavelengths. Then T1(t) and T2(t)were obtained by applying FastICA to the pairs of signals, one at 760 nm and another either at 800 nm, or at 830 nm, or at 860 nm, or 900 nm. The comparisons of the time courses of ICs obtained using the full spectrum and a pair of wavelengths (760 nm and 900 nm) were presented in Fig. 5. One can see that the time courses obtained using the full and reduced spectra were similar and the main difference was a much noisier T2(t) obtained using two wavelengths. Figure 6 compares ICs obtained using the full spectrum and pairs of wavelengths in terms of STDs of the differences between the corresponding temporal ICs. We note that the values of standard deviations in Fig. 6 should be also compared with the standard deviations of T1(t) and T2(t), which were both equal to one due to the normalization. From Fig. 6 one can conclude that (a) for all pairs of wavelengths the differences in T2(t) from the full spectrum were much larger than for T1(t), (b) for T1(t) the difference was minimal at 860 nm, and (c) for T2(t) the difference was minimal at 830 nm. Also from Fig. 6 one can see that the wavelength of 830 nm allowed for about 15-20% improvement compared to 800 nm and 900 nm. The effect of the wavelength tuning on T1(t) was very modest. Also we found that using combined signals at all five wavelengths (760 nm, 800 nm, 830 nm, 860 nm, and 900 nm) gave almost the same accuracy of T1(t) as the pair of 760 nm and 860 nm, and the same accuracy of T2(t) as the pair of 760 nm and 830 nm (Fig. 7
Fig. 7 Standard deviations of IC time courses computed using the full spectrum relative to those obtained using two wavelengths: one at 760 nm and another at a wavelength shown.
).

4. Discussion

We found that the responses from 630 spectral channels to BHs can be represented by only two ICs. Such a dramatic reduction of the data dimension creates a question of whether instead of the entire spectrum one can use a few isolated wavelengths. Since the signal-to-noise ratio is an important issue of all types of measurements, a complete answer to this question requires an additional study comparing data obtained using the broadband and multi-wavelength devices. However, the results presented in Fig. 6 showed that the time courses of ICs obtained using the full broadband spectrum and two wavelengths at 760 nm and 900 nm were similar. We stress that ICA applied to the two-wavelength data returned two ICs each corresponding to both HbO2 and HHb changes, and a 2 x 2 mixing matrix, whose four elements together with the extinction coefficients εHBO2(λ)and εHHb(λ) can be used to calculate contributions of bothΔHbO21,2(t) and ΔHHb1,2(t)to each component. Therefore, if different ICs indeed have different and typical physiological sources, ICA can be used to separate contributions of those sources to data acquired at one location and single source-detector separation even if the number of wavelengths is small. This approach can be combined with the multi-distance approach (such as in [6

6. A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis (Wiley, New York, 2001).

]) to better separate extracerebral and cerebral signals. The results of our investigation of the reproducibility of the time courses of the broadband ICs using few selected wavelengths showed that IC1 was much less affected by the spectrum reduction than IC2 and that the tuning of isolated wavelengths or increasing of their number caused only modest improvements.

5. Conclusion

By applying independent component analysis to broadband fNIRS data we found one major IC during baseline and two major temporal ICs, one leading and another delayed, during repetitive breath holds. The corresponding characteristic spectra of changes in optical absorption suggested that the leading component represented vasodilation of cerebral arterioles while the delayed component represented the washout of deoxyhemoglobin in either cerebral venules or in extracerebral tissue.

References and links

1.

D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab. 20(3), 469–477 (2000). [CrossRef] [PubMed]

2.

F. Irani, S. M. Platek, S. Bunce, A. C. Ruocco, and D. Chute, “Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging technology with important applications for the study of brain disorders,” Clin. Neuropsychol. 21(1), 9–37 (2007). [CrossRef] [PubMed]

3.

I. Schelkanova and V. Toronov, “Optimal quantitation of the cerebral hemodynamic response in functional near-infrared spectroscopy,” Opt. Express 18(18), 19386–19395 (2010). [CrossRef] [PubMed]

4.

H. Dehghani, F. Leblond, B. W. Pogue, and F. Chauchard, “Application of spectral derivative data in visible and near-infrared spectroscopy,” Phys. Med. Biol. 55(12), 3381–3399 (2010). [CrossRef] [PubMed]

5.

J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt. 14(5), 054032 (2009). [CrossRef] [PubMed]

6.

A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis (Wiley, New York, 2001).

7.

S. Kohno, I. Miyai, A. Seiyama, I. Oda, A. Ishikawa, S. Tsuneishi, T. Amita, and K. Shimizu, “Removal of the skin blood flow artifact in functional near-infrared spectroscopic imaging data through independent component analysis,” J. Biomed. Opt. 12(6), 062111 (2007). [CrossRef] [PubMed]

8.

F. C. Robertson, T. S. Douglas, and E. M. Meintjes, “Motion artifact removal for functional near infrared spectroscopy: a comparison of methods,” IEEE Trans. Biomed. Eng. 57(6), 1377–1387 (2010). [CrossRef] [PubMed]

9.

A. V. Medvedev, J. M. Kainerstorfer, S. V. Borisov, A. H. Gandjbakhche, and J. Vanmeter, ““Seeing” electroencephalogram through the skull: imaging prefrontal cortex with fast optical signal,” J. Biomed. Opt. 15(6), 061702 (2010). [CrossRef] [PubMed]

10.

H. Zhang, Y. J. Zhang, C. M. Lu, S. Y. Ma, Y. F. Zang, and C. Z. Zhu, “Functional connectivity as revealed by independent component analysis of resting-state fNIRS measurements,” Neuroimage 51(3), 1150–1161 (2010). [CrossRef] [PubMed]

11.

C. Hesse and C. James, “The fastica algorithm with spatial constraints,” IEEE Signal Process. Lett. 12(11), 792–795 (2005). [CrossRef]

12.

T. Len, J. Neary, G. Asmundson, D. Goodman, B. Bjornson, and Y. Bhambhani, “Cerebrovascular reactivity impairment after sport-induced concussion,” Med. Sci. Sports Exerc. 43(12), 2241–2248 (2011). [CrossRef]

13.

R. L. Grubb Jr, M. E. Raichle, J. O. Eichling, and M. M. Ter-Pogossian, “The effects of changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean transit time,” Stroke 5(5), 630–639 (1974). [CrossRef] [PubMed]

14.

U. E. Emir, C. Ozturk, and A. Akin, “Multimodal investigation of fMRI and fNIRS derived breath hold BOLD signals with an expanded balloon model,” Physiol. Meas. 29(1), 49–63 (2008). [CrossRef] [PubMed]

15.

A. Sassaroli and S. Fantini, “Comment on the modified Beer-Lambert law for scattering media,” Phys. Med. Biol. 49(14), N255–N257 (2004). [CrossRef] [PubMed]

16.

W. B. Gratzer, Medical Research Council Laboratories, Holly Hill, London, personal communication.

17.

N. Kollias, Wellman Laboratories, Harvard Medical School, Boston, MA, personal communication.

18.

H. Q. Woodard and D. R. White, “The composition of body tissues,” Br. J. Radiol. 59(708), 1209–1218 (1986). [CrossRef] [PubMed]

19.

J. H. Choi, M. Wolf, V. Toronov, U. Wolf, C. Polzonetti, D. Hueber, L. P. Safonova, R. Gupta, A. Michalos, W. Mantulin, and E. Gratton, “Noninvasive determination of the optical properties of adult brain: near-infrared spectroscopy approach,” J. Biomed. Opt. 9(1), 221–229 (2004). [CrossRef] [PubMed]

20.

B. J. MacIntosh, L. M. Klassen, and R. S. Menon, “Transient hemodynamics during a breath hold challenge in a two part functional imaging study with simultaneous near-infrared spectroscopy in adult humans,” Neuroimage 20(2), 1246–1252 (2003). [CrossRef] [PubMed]

21.

B. A. Poser, E. van Mierlo, and D. G. Norris, “Exploring the post-stimulus undershoot with spin-echo fMRI: implications for models of neurovascular response,” Hum. Brain Mapp. 32(1), 141–153 (2011). [CrossRef] [PubMed]

OCIS Codes
(170.0170) Medical optics and biotechnology : Medical optics and biotechnology
(170.2655) Medical optics and biotechnology : Functional monitoring and imaging

ToC Category:
Spectroscopic Diagnostics

History
Original Manuscript: September 29, 2011
Revised Manuscript: December 1, 2011
Manuscript Accepted: December 2, 2011
Published: December 9, 2011

Citation
Irina Schelkanova and Vladislav Toronov, "Independent component analysis of broadband near-infrared spectroscopy data acquired on adult human head," Biomed. Opt. Express 3, 64-74 (2012)
http://www.opticsinfobase.org/boe/abstract.cfm?URI=boe-3-1-64


Sort:  Author  |  Year  |  Journal  |  Reset  

References

  1. D. A. Benaron, S. R. Hintz, A. Villringer, D. Boas, A. Kleinschmidt, J. Frahm, C. Hirth, H. Obrig, J. C. van Houten, E. L. Kermit, W. F. Cheong, and D. K. Stevenson, “Noninvasive functional imaging of human brain using light,” J. Cereb. Blood Flow Metab.20(3), 469–477 (2000). [CrossRef] [PubMed]
  2. F. Irani, S. M. Platek, S. Bunce, A. C. Ruocco, and D. Chute, “Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging technology with important applications for the study of brain disorders,” Clin. Neuropsychol.21(1), 9–37 (2007). [CrossRef] [PubMed]
  3. I. Schelkanova and V. Toronov, “Optimal quantitation of the cerebral hemodynamic response in functional near-infrared spectroscopy,” Opt. Express18(18), 19386–19395 (2010). [CrossRef] [PubMed]
  4. H. Dehghani, F. Leblond, B. W. Pogue, and F. Chauchard, “Application of spectral derivative data in visible and near-infrared spectroscopy,” Phys. Med. Biol.55(12), 3381–3399 (2010). [CrossRef] [PubMed]
  5. J. Virtanen, T. Noponen, and P. Meriläinen, “Comparison of principal and independent component analysis in removing extracerebral interference from near-infrared spectroscopy signals,” J. Biomed. Opt.14(5), 054032 (2009). [CrossRef] [PubMed]
  6. A. Hyvarinen, J. Karhunen, and E. Oja, Independent Component Analysis (Wiley, New York, 2001).
  7. S. Kohno, I. Miyai, A. Seiyama, I. Oda, A. Ishikawa, S. Tsuneishi, T. Amita, and K. Shimizu, “Removal of the skin blood flow artifact in functional near-infrared spectroscopic imaging data through independent component analysis,” J. Biomed. Opt.12(6), 062111 (2007). [CrossRef] [PubMed]
  8. F. C. Robertson, T. S. Douglas, and E. M. Meintjes, “Motion artifact removal for functional near infrared spectroscopy: a comparison of methods,” IEEE Trans. Biomed. Eng.57(6), 1377–1387 (2010). [CrossRef] [PubMed]
  9. A. V. Medvedev, J. M. Kainerstorfer, S. V. Borisov, A. H. Gandjbakhche, and J. Vanmeter, ““Seeing” electroencephalogram through the skull: imaging prefrontal cortex with fast optical signal,” J. Biomed. Opt.15(6), 061702 (2010). [CrossRef] [PubMed]
  10. H. Zhang, Y. J. Zhang, C. M. Lu, S. Y. Ma, Y. F. Zang, and C. Z. Zhu, “Functional connectivity as revealed by independent component analysis of resting-state fNIRS measurements,” Neuroimage51(3), 1150–1161 (2010). [CrossRef] [PubMed]
  11. C. Hesse and C. James, “The fastica algorithm with spatial constraints,” IEEE Signal Process. Lett.12(11), 792–795 (2005). [CrossRef]
  12. T. Len, J. Neary, G. Asmundson, D. Goodman, B. Bjornson, and Y. Bhambhani, “Cerebrovascular reactivity impairment after sport-induced concussion,” Med. Sci. Sports Exerc.43(12), 2241–2248 (2011). [CrossRef]
  13. R. L. Grubb, M. E. Raichle, J. O. Eichling, and M. M. Ter-Pogossian, “The effects of changes in PaCO2 on cerebral blood volume, blood flow, and vascular mean transit time,” Stroke5(5), 630–639 (1974). [CrossRef] [PubMed]
  14. U. E. Emir, C. Ozturk, and A. Akin, “Multimodal investigation of fMRI and fNIRS derived breath hold BOLD signals with an expanded balloon model,” Physiol. Meas.29(1), 49–63 (2008). [CrossRef] [PubMed]
  15. A. Sassaroli and S. Fantini, “Comment on the modified Beer-Lambert law for scattering media,” Phys. Med. Biol.49(14), N255–N257 (2004). [CrossRef] [PubMed]
  16. W. B. Gratzer, Medical Research Council Laboratories, Holly Hill, London, personal communication.
  17. N. Kollias, Wellman Laboratories, Harvard Medical School, Boston, MA, personal communication.
  18. H. Q. Woodard and D. R. White, “The composition of body tissues,” Br. J. Radiol.59(708), 1209–1218 (1986). [CrossRef] [PubMed]
  19. J. H. Choi, M. Wolf, V. Toronov, U. Wolf, C. Polzonetti, D. Hueber, L. P. Safonova, R. Gupta, A. Michalos, W. Mantulin, and E. Gratton, “Noninvasive determination of the optical properties of adult brain: near-infrared spectroscopy approach,” J. Biomed. Opt.9(1), 221–229 (2004). [CrossRef] [PubMed]
  20. B. J. MacIntosh, L. M. Klassen, and R. S. Menon, “Transient hemodynamics during a breath hold challenge in a two part functional imaging study with simultaneous near-infrared spectroscopy in adult humans,” Neuroimage20(2), 1246–1252 (2003). [CrossRef] [PubMed]
  21. B. A. Poser, E. van Mierlo, and D. G. Norris, “Exploring the post-stimulus undershoot with spin-echo fMRI: implications for models of neurovascular response,” Hum. Brain Mapp.32(1), 141–153 (2011). [CrossRef] [PubMed]

Cited By

Alert me when this paper is cited

OSA is able to provide readers links to articles that cite this paper by participating in CrossRef's Cited-By Linking service. CrossRef includes content from more than 3000 publishers and societies. In addition to listing OSA journal articles that cite this paper, citing articles from other participating publishers will also be listed.


« Previous Article  |  Next Article »

OSA is a member of CrossRef.

CrossCheck Deposited