A frequency multiplexed near infra-red topography
system for imaging functional activation in the brain
N. Everdell, A. Gibson, I. D. C. Tullis, T. Vaithanian, J. Hebden and D. Delpy
Department of Medical Physics & Bioengineering, University College London, 11-20 Capper Street, London WC1E 6JA, UK
Email:everdell@medphys.ucl.ac.uk
Abstract: we have developed a novel near-infrared optical topography system that can acquire
images at 10 frames per second. Sixteen laser diode sources (at 785 and 850 nm) are illuminated
simultaneously, and each of 8 avalanche photodiode detectors records light from multiple sources.
In this system, the contribution from each source is demultiplexed in software using fast Fourier
transforms. This allows for a more flexible, portable and less complex system than using
traditional hardware demodulation techniques. It will eventually incorporate 64 sources and 32
detectors. We will report on the initial clinical investigations that have been undertaken using the
new instrument.
2000 Optical Society of America
OCIS codes: (170.3880) Medical and biological imaging
1. Introduction
Diagnostic methods are being developed which exploit the high transmittance of tissue at near-infrared
wavelengths. In particular, the different specific absorption characteristics of oxyhaemoglobin and
deoxyhaemoglobin enable blood content and oxygenation to be derived. Infra-red topography enables spatially
resolved changes in activation to be recorded from an array of sources and detectors placed against the scalp.
This provides a two dimensional map of functional activity within the tissue.
1
The Hitachi ETG 100 optical topography system has 20 laser diode sources (10 at each of 2
wavelengths of 780 and 830 nm) and 8 avalanche photodiode detectors. Each laser diode is intensity modulated
at a different frequency. The output from each detector is fed into a set of lock-in amplifiers which can
discriminate between the different frequencies in the detected signal. This method, known as frequency
multiplexing, allows all the sources to be illuminated simultaneously, so that an image can be obtained more
quickly. This system has been used in a wide variety of investigations 2, 3, 4, 5. Similar systems have been
6
7
employed to image the cortex of newborn babies. Danen et al. have developed a system with 12 sources and
4 detectors that records both the amplitude and the phase of the modulated signal. Francheschini et al.8 have
built a similar system, based on 16 intensity modulated sources and two photomultiplier tube detectors.
2. The UCL Topography System
At UCL we have developed a new approach to optical topography that involves using software to demultiplex
multiple source signals which are modulated at different frequencies in parallel. This allows great flexibility in
the positioning of sources and detectors. Different optode arrays can be used with only minor changes to the
software. It currently employs 16 laser diode sources, (8 at 785 nm and 8 at 850 nm) and 8 avalanche
photodiode detectors but will eventually have 64 sources and 32 detectors. With this specification hardware
demultiplexing would require over 100 lock-in amplifier circuits, increasing both size and complexity.
A block diagram of the laser sources is shown in figure 1, and the detection system in figure 2. All the
multiplexing frequencies are derived from the same 20 MHz oscillator, which prevents any relative drift in
frequency between different sources. The oscillator signal is fed into a frequency divider which reduces the
frequency to the kilohertz range. A digital potentiometer then enables the amplitude of the resulting square
wave to be adjusted to the required level. The square wave then modulates the laser diode. Currently there are
16 laser diode sources (8 lasers at 785 nm and 8 at 850 nm), driven by frequencies ranging from 2 kHz to 4 kHz
. The frequencies are kept within one octave to prevent any possible interference from harmonics.
There are currently 8 Hamamatsu C5460-01 avalanche photodiode detectors (APDs) with a 3 mm
diameter collecting area and a measurement bandwidth of approximately 100 kHz. The signal from these is ac
coupled into an 8 pole low pass filter that has a 3 dB point at 4 kHz. This allows for 64 dB of roll off before the
Nyquist frequency, so aliasing is prevented. The output from the anti-aliasing filter is fed into an analogue to
digital converter which samples each channel at 20 kHz. The signal can be sampled directly from the APD
(both dc and ac components) or from the output of the low pass filter (just an ac component). The data is then
detrended using a Hanning window on a 2048 point sample. An FFT is then performed to separate the different
source signals. This represents a novel alternative to the more usual technique of using lock-in amplifiers.
Frequency
divider
digital
potentiometer
laser diode
driver
5mW @ 780nm
20 MHz / 10,000 = 2 kHz
20 MHz
oscillator
Frequency
divider
digital
potentiometer
laser diode
driver
To other channels
5mW @ 850nm
20 MHz / 9852 = 2.03 kHz
Figure 1: source
Avalanche
photodiode
detector
Low pass
filter
Under PC
control
DC
AC
Analogue to
digital
converter
PC
Figure 2: Detector
The source signals are demultiplexed in software rather than hardware. From this information a spatial map of
the tissue under interrogation can be built up. In addition to this frequency multiplexed scheme, the system can
also operate in a time multiplexed scheme, which involves illuminating individual laser sources in turn for a
short period. This involves switching individual laser sources on in turn, for a short time. Although this has the
disadvantage of slowing the overall data acquisition rate, the overall signal to noise ratio of the data is increased
as there is less background light contributing shot noise to the measurement. In both schemes of operation, a
subset of sources can be selected so that a smaller area of the cortex can be focussed upon, and the rate of
acquisition of images increased.
Both sources and detectors are coupled to the tissue under investigation via simple multimode PMMA
optical fibres, of 1 mm diameter. These fibres are held in an array by inserting them into a thermoplastic shell
that can be moulded to the shape of the part of the body under investigation. The interior surface of the shell is
lined with soft, light-absorbing foam. The signal to noise ratio across 3 cm of tissue equivalent phantom
material at maximum optical power is approximately 600:1 (56 dB). Currently the data acquisition card encodes
the data to a depth of 14 bits, although this will eventually be increased to 16 bits.
Figure 3
Figure 4
3. Experimental Work
3.1 Liquid Phantom
For a preliminary evaluation of the system an array has been devised that employs 16 sources and 4 detectors,
as shown in figure 3. This array has potentially many distinct source-detector separations. The maximum
separation over which a signal can be detected through tissue has been found to be approximately 40 mm.
Therefore for this array the useful separations range from 15.8 mm to 38.1 mm. For the middle two detectors a
signal can be obtained from the six nearest sources.
The array is shown attached to a liquid phantom in figure 4 above. The phantom consists of a tank of
scattering fluid with one wall made from a 5 mm thick epoxy resin slab. The tank itself is filled with an aqueous
solution of intralipid and absorbing dye, with optical properties of ?a = 0.007 mm-1 and ? s = 0.8 mm-1. A stepper
motor is mounted on the lid, which enables a small target to be rotated within the liquid suspended on a short
length of wire. The target consists of an epoxy resin cylinder, 10 mm in length and 10 mm in diameter, with the
same scattering properties as the surrounding liquid, and ten times the absorption. The motor enables the target
to be rotated in a plane perpendicular to the face of the phantom in a circle of diameter 60 mm, and at a speed of
up to 1 rev/sec. The first experiment with this phantom involved recording two sets of data, first with the target
removed, and second with a target in a fixed position. Data was collected for 30 seconds in each case. The
centre of the cylinder was placed 15 mm below the plane of the optode array, along a normal line that passed
through the centre of the optode array.
Three-dimensional images representing the difference in optical properties between the two states were
reconstructed using the Rytov approximation 9. Measured changes in log(amplitude), y, were assumed to be
related to changes in the optical absorption, x, by the matrix equation y=Ax, where A is the sensitivity matrix.
Here, A was calculated by solving the diffusion equation using the finite element method applied to a finite
10
element mesh of linear tetrahedra which was generated using Netgen (figure 5). A and y were normalised by
dividing by the initial estimated background absorption. Images were generated by Tikhonov regularisation of
the Moore-Penrose generalised inverse x = AT (AAT+λI) y where the regularisation parameter  was set to 1 %
of the largest singular value of AAT. Figure 5 shows three orthogonal slices through the centre of the target. The
position of the image in the (x, y) plane (i.e. in the plane of the optode array) correlates closely with the position
of the target. In the z direction, the centre of the image appears at a depth of 11 mm, but the actual depth of the
centre of the target was 15 mm. The x-y plane image shows absorption increase artefacts coincident with the
optode positions. This is due to increased sensitivity immediately under the optodes. The effect of this will be
11
reduced by normalizing to a standard deviation image calculated from baseline data .
Figure 5
3.2 Motor Cortex Experiment
Initial experiments on human volunteers are currently in progress in order to display functional activation of the
motor cortex. The thermoplastic of the array can be moulded to fit the shape of each individual subject's head.
The array was designed with motor cortex experiments in mind. The inner rectangle in figure 3 represents an
estimate of the size and shape of the adult motor cortex, based on anatomical information.
4. References
1. Yamashita, Y., Maki, A., and Koizumi, H. (1999). Measurement system for noninvasive dynamic optical topography.Journal of
Biomedical Optics, 4(4), 414 - 417.
2. Maki, A., Yamashita, Y., Ito, Y., Watanabe, E., Mayanagi, Y., & Koizumi, H. (1995). Spatial and temporal analysis of human motor
activity using noninvasive NIR topography. Medical Physics, 22, 1997-2005.
3. Yamashita, Y., Maki, A., & Koizumi, H. (2001). Wavelength dependence of the precision of noninvasive optical measurement of
oxy-, deoxy-, and total-hemoglobin concentration. Medical Physics, 28, 1108-1114.
4.
Watanabe, E., Maki, A., Kawaguchi, F., Takashiro, K., Yamashita, Y., Koizumi, H., & Mayanagi, Y. (1998a). Non-invasive
assessment of language dominance with near-infrared spectroscopic mapping. Neuroscience letters, 256, 49-52.
5. Watanabe, E., Maki, A., Kawaguchi, F., Yamashita, Y., Koizumi, H., & Mayanagi, Y. (1998b). Noninvasive cerebral blood volume
measurement during seizures using multichannel near infrared spectroscopic topography. Journal of Epilepsy, 11, 335-340.
6. Hintz, S. R., Benaron, D. A., Siegal, A. M., Zourabian, A., Stevenson, D. K., & Boas, D. A. (2001). Bedside functional imaging of the
premature infant brain during passive motor activation. Journal of Perinatal Medicine, 29, 335-343.
7. Danen, R. M., Wang, Y., Li, X. D., Thayer, W. S., & Yodh, A. G. (1998). Regional imager for low resolution functional imaging of
the brain with diffusing near-infrared light. Photochemistry and Photobiology, 67, 33-40.
8. Franceschini, M. A., Toronov, V., Filiaci, M. E., Gratton, E., & Fantini, S. (2000). On-line optical imaging of the human brain with
160-ms temporal resolution. Optics Express, 6, 49-57.
9. Arridge, SR (1999): Optical tomography in medical imaging. Inverse Problems 15(2), 41-93.
10. Schoberl J 1997 NETGEN-An advancing front 2D/3D-mesh generator based on abstract rules. Comput.Visual. Sci. 1 41-52
11. Culver, J.P., Siegel, A.M., Stott, J.J. and Boas, D.A. "Volumetric diffuse optical tomography of brain activity."Optics Letters 28(21):
2061-2063. 2003.