 |
Description  |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to diagnostic medical instrumentation, or more generally to spectroscopic methods of analytical chemistry. In particular, the present invention defines a class of instruments which may be used to conveniently
monitor various metabolites, or to quantitatively characterize the constituents of a chemical mixture. A primary objective of the present invention is to provide a versatile and relatively cheap spectroscopic device that can be used to determine the
spatial distribution of such metabolites in a tissue, or components in a mixture.
2. Description of Background Art
The present invention concerns a novel design for a class of integrated optical devices. In brief the photons produced by a set of semiconductor light-emitting diodes (LEDs) or injection laser diodes (ILDs) are gathered into a small lightbeam or
lightbeams. The wavelengths of light are chosen from the visible and infrared spectrum so as to permit the quantification of various components of a mixture through characteristic absorption behavior. Moreover the spatial distribution of these
components may be determined in two dimensions so as to produce an image, by the manner in which the light is directed onto the sample and detected after it emerges from the sample.
The invention is described as "discrete" Fourier transform spectroscopy because discrete wavelengths are used rather than a continuum of light. The absorption spectrum is obtained by means of a Fourier transform.
The background an involves several distinct disciplines. Chemometry uses chemical measurements along with the statistical tools of multivariate calibration and analysis to quantify the components of a mixture. The chemical measurements can
include near-infrared spectroscopy, which may be of a sort referred to as Fourier transform spectroscopy. Process analysis and control determine how chemical sensors can best be used to monitor industrial and other processes, while clinical chemistry
identifies the manner in which chemometry may be usefully applied in medicine. Physical optics describes the various tools by which light can be analyzed and controlled, and the means to overcome certain experimental problems. Light-scattering in
turbid samples is one such problem that has been addressed through various analytical models, experimental protocols, and image-processing techniques. Optoelectronics can produce monochromatic light of a desired wavelength at high efficiency, with
semiconductor structures that can be mass-produced at low cost.
An imaging near infrared spectroscope could be used to analyze tissues and organs noninvasively, through the skin. Many chronic illnesses require invasive tests which withdraw blood for analysis, such as for glucose in the case of diabetes.
Such blood tests are inconvenient and sometimes problematic. Simple applications of spectral analyses to such problems have appeared in the published literature, along with a variety of specific multivariate calibrations. However, any single scalar
quantity such as a blood glucose level can probably be more economically and accurately determined by an implantable sensor designed for that specific purpose. Problems of biocompatibility will be overcome, so that implantable sensors seem to offer a
more cost-effective means of monitoring such physiological parameters, while avoiding the painful and repetitive inconvenience of invasive blood tests. The most useful application of spectroscopy as a portable sensor technology seems to be in imaging
samples: scanning the body (transillumination or diaphanography), or monitoring the spatial distributions of the components of a complex chemical mixture undergoing flow or other perturbations.
Chemometry applies mathematics and statistics as well as models of chemical structure to the quantitative analysis of chemical mixtures and solutions. Qualitative chemometric pattern-recognition techniques such as classification and clustering
methods are applied to exploratory studies, while multivariate calibration is used for quantitative analyses of systems that are already relatively well-understood. The approach to calibration is empirical and condenses pertinent information into the
most useful form, while structural models are used for both prediction and interpretation.
Multivariate calibration and analysis use statistical simplifications which have proved expedient to model and analyze complex chemical mixtures and solutions. Typically, the chemist starts by constructing a data matrix D from the spectra for a
set of samples, and a control matrix C from a set of known concentrations. Then a suitable mathematical approach is chosen for reproducing the matrix D from the matrix C. Three common calibration methods include multiple linear regression (MLR),
principal component regression (PCR) and partial least squares regression (PLSR).
MLR is adequate in ideal situations (linear and independent response of absorbing species, high dilution or no interactions of species, and low noise), but implicitly incorporates much irrelevant information into the model. MLR uses a linear
combination of variables to solve the equation C=AD+B, where A is a matrix of coefficients estimated by linear regression, and B is a matrix of errors associated with (and minimized by) the MLR model. But real data includes much irrelevant or useless
information which should not be given equal weight.
PCR explicitly deletes irrelevant noise and thereby reduces the dimensionality of the problem, which is helpful mathematically and practically in obtaining a solution. PCR is a factor-based modeling procedure, a factor being any linear
combination of the original variables in C or R. Factors can be used to decrease the weighting given to irrelevant data, which can include redundant or collinear data as well as noise. PCR starts by determining the factors for the data matrix D,
beginning with the linear combination or factor most correlated to the original variables, and proceeding to less and less correlated factors. The data matrix can then be re-expressed in terms of the factors as a score matrix F, in the process possibly
deleting some factors and reducing the dimensions of the problem. Finally, MLR is used to solve the equation C=AF.
PLSR is also factor-based but further improves upon PCR by limiting the types of acceptable solutions to the problem. This is useful in obtaining a best fit to the data. The factors are determined for both the reference or control matrix C and
the data matrix D at the same time. The factor model for PLSR thus is a compromise that describes both C and D, and incorporates more information in the model-building phase.
In practice for PCR or PLSR some optimal number of factors will exist which produces the best fit to the data. More factors than this optimal number will tend to decrease the predictive ability of the model, by overweighting the effects of noise
in the data. Fewer factors may ignore crucial features of the sample or relevant information.
(The particular wavelengths and factors best suited for a particular analysis would usually be identified through research and development, using a continuous spectrum. Neural nets implemented in hardware or software may help to identify more
quickly the most significant wavelengths and best factors, for particular analytical problems. The set of wavelengths necessary for the analysis then could be implemented using discrete Fourier transform spectroscopy and imaging as described in the
present invention.)
Near infrared spectroscopy (NIRS) is useful in examining aqueous solutions and mixtures, as well as in biological studies. While the chief application of absorption spectrometry is to quantitative measurements, such measurements are less
sensitive in the infrared (IR) regime than in the visible or ultraviolet. This is because vibrational and other transitions responsible for infrared absorption are less intense than the electronic transitions in the visible and ultraviolet absorption.
Consequently IR is not as sensitive in analyses of species present at low or trace concentrations. Yet electronic transitions have the drawback of often resulting in chemical changes, an undesirable side-effect when examining living tissue. The energy
absorbed in vibrational transitions, however, is converted more directly into heat with less likelihood of harmful side-effects.
The less intense absorption of near-infrared wavelengths also allows them to penetrate sufficiently to be useful in the analysis of thicker samples, such as body tissues. Water, proteins, nucleic acids, and other molecules absorb strongly in the
ultraviolet, visible, and mid-infrared region of the spectrum but only weakly in the near-infrared. Qualitative analysis in the IR regime is excellent, and very good quantitative results can be obtained for species present in moderate concentrations.
Beefs Law describes the nature of optical absorption in mixtures of linearly independent absorbing species:
where I.sub.0 is the incident light, N is the concentration of absorbers, A is the absorptivity, and d is the path length. For a mixture of j absorbing species,
where N.sub.j is the concentration of absorbers of the j.sup.th type, and A.sub.j is the absorptivity of the j.sup.th type. For the same mixture enclosed in a container with k absorbing barriers,
where N.sub.k is the number of absorbing barriers of the k.sup.th type, and A.sub.k is the absorption of the k.sup.th type of barrier.
Beer's Law suggests a simple means of estimating constant absorption due to skin or other membranes in the light path. If the sample or tissue is compressed and released in order to alter its optical thickness d, constant contributions from less
elastic parts or tissues may be subtracted from the absorption. This may help address problems such as differing levels of melanin in the skin, and better distinguish bone from fat. Also it is worth noting that the action of the heart introduces a
natural frequency (the pulse) with which changes in blood pressure alter the amount of blood in the optical path. This can, for example, rhythmically change the concentration of oxygen by about 2%. Stroboscopic and schlieren methods can be used to
select different parts of this cardiovascular cycle, or take difference measurements at the maximum and minimum blood pressures in order to better characterize the blood.
Differential photometry or the transmittance ratio method can be used to obtain better accuracy and precision at very low (or very high) concentrations. This involves the study of small changes in a predefined range of concentrations, seeking
ways to amplify the effects of these changes and so boost the sensitivity of the measurement. A small range of concentrations is bracketed and a calibration curve is prepared using reference samples.
To the extent that the absorption behavior is linear, improved results may also be obtained by use of a dual-wavelength approach. Two wavelengths are chosen for each species-an isobestic point and a distinctive absorption peak. An isobestic
point is a wavelength at which the system absorbance remains constant although the concentration of a given species is changing. An absorption peak often can be found that distinguish the species of interest; if necessary, more than one such peak and
wavelength can be used per species. Thus at least two wavelengths, and possibly more, may be chosen for each species in the system. The use of multiple wavelengths per species also provides an important means to confirm the identification of a species
to be quantified. For example, slight increases in temperature shift the water absorption bands to higher frequencies, lowering the transmission in at least one band of wavelengths useful for measuring glucose.
However, non-linearities are often present due to the interactions of various molecules and chemical or physical structures. Beefs Law does not fully describe many samples of interest, and multivariate calibration is required to predict and
interpret the results. Moreover, samples which are relatively opaque or turbid may produce a great deal of light scattering, which further complicates analytical spectroscopy. The precise position of the spectrometer with respect to the sample can also
introduce variations or changes in the data, which must be taken into account in the calibration algorithm.
The multivariate calibration method must take into account various other non-linearities that may be present in practice. These include high opacity or turbidity, due to scattering by the sample. Since instrumental noise becomes more
significant in differential measurements, precautions must also be taken to optimize the signal-to-noise ratio. Despite all these complications, the measurement problem remains unchanged-the identification of the relative concentrations in a mixture.
Fourier transform spectroscopy has certain advantages over traditional spectroscopy, in which the response of a sample to light is measured by scanning sequentially over a range of wavelengths. Fourier transform spectroscopy measures the
response of the sample to all the wavelengths of interest simultaneously, by measuring the light after it interacts with the sample and recording the entire spectrum at once. The signal is recorded as the Fourier transform of the wavelength, as a
function of distance.
The Michelson interferometer is one of the simplest designs described in the prior art. A spectral interferogram is obtained with a Michelson interferometer by systematically varying the path length of one lightbeam with respect to a second
lightbeam. The resolution (in wavenumbers or cm.sup.-1) is defined by 1/d where d is the change in path length, while the sampling interval is 1/2w where w is the maximum wavenumber of interest.
One advantage of Fourier transform spectroscopy is a better signal-to-noise (S/N) ratio. All X distinguishable wavelengths are measured simultaneously. Since the signal increases linearly and the noise with the square root of the measurement
time, the S/N ratio increases by X.sup.1/2 (the so-called Fellgett advantage). A suitable detector must be chosen, of course, so that the detector noise does not increase in proportion to the signal level.
Another advantage is that all the wavelengths are combined in a single beam of light, which can be adapted to microscopy and imaging applications.
The S/N ratio can be further improved by rejecting scattered or background light. Use of a single bandpass filter to filter out unwanted wavelengths in a continuous spectrum has been shown to decrease the root-mean-square noise level about
3-fold for a particular absorption band, with a more than 10-fold increase in sensitivity. Thus if FTIR measures only the wavelengths of interest, the dynamic range of the detector can be more fully utilized. A more efficient approach than filtering a
continuous spectrum is taken in the present invention-only the wavelengths of interest for the analytical problem are generated.
Assume that a specific analytical problem requires a set of twenty or fewer wavelengths in the lightbeam. The spectral resolution requirements may not be very stringent to distinguish these discrete wavelengths from one another. If a resolution
of 100 cm.sup.-1 is sufficient, the path difference required is on the order of 1 millimeter. If the maximum wavenumber is 10,000 cm.sup.-1 (a wavelength of about 1 micron), then the sampling interval is 0.5 micron and 2000 sampling intervals are
required.
The digitized spectrum then can be processed with a "fast Fourier transform" or FFT algorithm. For a spectrum of N points FFT methods require about N*logN operations, whereas older matrix methods required N*N operations. The FFT works best when
N is a power of 2, so that it is worthwhile to add null values to the spectrum to increase N until it is a power of 2 (otherwise the algorithm may work much more slowly). Thus a 2000-step spectrum should be increased to a 2048-step spectrum, and the FFT
processing will be over 250 times faster than matrix methods.
Digital filtering also is of great use, in order to pass only that part of the signal that is at the frequencies or wavelengths of interest. It is generally desirable to remove from the data both high-frequency noise and low-frequency drift, or
baseline variations. For the present invention, it would be desirable to pass only the frequencies corresponding to the original set of wavelengths. It is often more convenient (and faster) to filter in the Fourier domain, which simply involves
multiplying the FFT by a filter function. One reasonable filter function is a set of Gaussian functions centered on the wavelengths of interest, with full-widths at half maxima corresponding to the linewidths of the lightsources.
The FFT of the signal on a detector produces a spectrum of intensity versus wavelength or frequency, from which absorption behavior can be obtained. This data is then evaluated with the PLSR calibration scheme, and the corresponding pixel in an
image or display is colored so as to convey the pertinent information.
Process analysis and control is an important industrial and practical application of chemometry. Near-infrared wavelengths of light are absorbed due to distinctive molecular vibrations and low-level electronic excitations. Many molecules,
particularly molecules of biochemical interest, have characteristic "fingerprint" absorption spectra in the near infrared. Consequently applications may exist for monitoring processes in biotechnology and diverse other industries. A common engineering
problem in the mass production of a desired substance has to do with the dynamics of flow, or rheology. Remote or non-contact spectroscopic sensing avoids disruption of the flow patterns, while offering rapid feedback for process control.
Multivariate calibration methods can be adapted to statistical process control (SPC) techniques, which are in widespread use to maintain and improve product quality.
Clinical chemistry is another important practical application of chemometry. Clinical chemistry covers a very wide area, including the identification of cost-effective and reliable means of securing accurate diagnoses. (Appropriate regimes of
therapy are chosen by other means.) Near infrared spectroscopy or NIRS was applied to human skin in the 1950's and has since been developed for transcutaneous measurements of body fat composition, oxygen levels in blood and tissue, and breast cancer
screening as well as for many in vitro biochemical measurements.
Transillumination or diaphanography irradiates a body part such as the breast with near-infrared light, and records the image formed by transmitted light.
In traditional methods of transillumination, a broad lightbeam is directed into the body. Either reflected or transmitted light is used to project an image of internal body parts onto the skin. Light is reflected well from the interface of
tissues with air or fluid so that this method has been used to image arteries and veins, seminal vesicles, intestines, and so forth. Other applications have been in specialties such as pediatrics of infants, ophthalmology, urology, venipuncture, and
dentistry, which are not confronted with the problem of examining small organs deep within large bodies.
Radiologists have shown that transillumination can be used in mammography to distinguish benign tumors, malignant tumors, and cysts from each other. However, in clinical trials mammography using traditional methods of transillumination was no
more effective then manual or physical examination, since the image resolution for transmitted light is limited to about 2 cm or so.
Image resolution can be improved using collimation and time-of-flight methods that have been recently developed for the examination of turbid samples.
In addition, near-infrared wavelengths penetrate biological tissues more deeply than visible light. Studies of brain function in fetuses and infants have used the absorption of infrared light to quantify levels of oxyhemoglobin, deoxyhemoglobin,
and oxidized mitochondrial cytochrome oxidase. The concentrations of these molecules indicate cerebral blood flow and volume and change in response to external perturbations such as increased oxygen, the onset of contractions in labor, or the
administration of various therapeutic drugs.
Physical optics concerns the properties of light and experimental means to generate, control, and measure light. This body of knowledge is necessary to the design of the present invention.
Optical coatings influence how much light is reflected, transmitted, or absorbed by each optical element in an instrument and so can help improve the overall performance. Anti-reflective coatings are suitable for prisms and lenses, while
reflective coatings are desirable for mirrors and the cladding of fiberoptic lightguides. Coatings can form Fabry-Perot interference filters which transmit or reflect only desired wavelengths of light.
Spectrographs and spectrometers use prisms or gratings to disperse light according to wavelength. Spectrographs record the entire spectrum at once and spectrometers employ a slit to record only a narrow range of wavelengths at any given moment.
A spectrograph can measure the absorption of all wavelengths in an infrared lightbeam simultaneously, but requires X detectors to record each of X separate wavelengths. Thus the S/N ratio decreases by X.sup.0.5 compared to FTIR which uses a single
detector. The number of optical components per lightbeam is also lower for FTIR, which therefore provides a more economical approach to imaging with arrays of lightbeams.
Materials for infrared optical applications are well-known. Metals or metallic coatings serve as good broadband infrared reflectors (with efficiencies on the order of 99% ). Even higher reflectivities may be obtained over selected wavelength
bands using all-dielectric or dielectric- enhanced-metal mirrors. Silica glass, aluminum oxide, and magnesium oxide transmit near-infrared light adequately. Germanium, germanium-arsenic-selenium glass, or arsenic trisulphide function well in the mid-
and far-infrared regime. Infrared fiberoptics can be fabricated with chalcogenide glass (3-10 microns transmission), fluoride glass (0.5 to 4.3 microns), arsenic trisulphide (1 to 8 microns), AgClBr (3.3 to 15 micron), and sapphire (0.3 to 3.5 microns). Alternatively, hollow metal fibers can be effective broadband lightguides. Materials for lenses and prisms must be transparent in the spectral region of interest. The refractive index and spectral dispersion (rate of change in refractive index with
wavelength) should both be large for a prism, while for a lens the spectral dispersion should be low to minimize chromatic aberration. For example, the angular dispersion is eight times lower for quartz than for heavy flint glass at a wavelength of 0.4
micron, making it a good choice for a lens but not a prism.
Mercury cadmium telluride (MCT) detectors are preferred for fast and sensitive infrared measurements. Both narrowband and wideband versions are available, the former having a cut-off at 750 cm.sup.-1 and the latter at 400 cm.sup.-1. Liquid
nitrogen cooling is recommended for the best results, to reduce thermal noise to the background-limited infrared photodetector (BLIP) limit. (This limit decreases with decreasing wavelength and sample temperature; the infrared spectrum of a normal human
being has a thermal maximum at a wavelength close to 10 microns.) A single liquid nitrogen reservoir may cool an array of MCT detectors. Detectors for spectroscopy in the visible to the very near infrared may be selected from a wide variety:
photomultipliers, or semiconductor (Si, Ge or AlGaAsSb) photovoltaic or photoconductive designs. Photodiodes are photovoltaic or photoconductive designs operated at a large reverse-bias voltage, which offer high amplification and speed. Some Ge-based
detectors are among the fastest broadband detectors, but do not reach as far into the IR as do HgCdTe detectors.
Optical multichannel analyzers (vidicons) use a silicon target with a microscopic array of up to 10,000,000 photodiodes to provide excellent spatial resolution. In combination with an image intensifier, a vidicon can provide quantum efficiencies
on the order of 15%.
The image intensifier can also be gated in order to provide high time resolution. Optical, electo-optical, or electrical gating may be used either in the time domain or the frequency domain. Homodyne detection makes use of phase modulation and
lock-in techniques, which are well-known and relatively easy to implement. (Heterodyne detection must deal with non-linear effects in somewhat sophisticated and expensive ways to attain good signal-to-noise ratios, but is less sensitive to imperfections
and noise in the detector.) If an optical switch based on a Kerr or Pockels cell were used in front of the detectors to lock-in on a given phase, the dynamic range of the detector would be somewhat improved. An array of Kerr cells could be fabricated in
the form of Ti-diffused LiNbO.sub.3 channels. Whether the phase modulation takes place between the sample and detector, or after the detector, a single electronic phase-locked loop can be used to gate the entire array synchronously.
Light scattering occurs in turbid samples. When a lightbeam passes through a clear sample with negligible scattering, the photon trajectories are ballistic. A two-dimensional image maps directly onto its projection. The photon trajectory is no
longer ballistic in a turbid sample, but diffuses away from the central axis of the trajectory as a result of multiple scattering events. The projection of an image will be blurred and attenuated, as indicated by the results of Monte Carlo simulations
summarized in FIG. 1.
The trajectories through a sample can be described for photons of a given wavelength in terms of the mean free path and phase angle for scattering, and the mean free path for absorption. A given point in the original image will spread out onto a
distribution of points in the projected image. In an isotropic sample, the diffusive spread can be described in terms of a Gaussian or normal distribution.
This Gaussian function is actually the product of several separate Gaussian distributions, each with its own characteristic halfwidth. The halfwidth is a quantity defined as half the width of the symmetric distribution, at a point that is half
the maximum height or amplitude. The halfwidth for the product of two (or more) Gaussians is the sum of the individual halfwidths. It seems most instructive to conceive of the original image propagating through the sample, each point "blooming" or
spreading according to the following distributions.
A. First, photons will tend to spread slightly from the axis due to divergence in the original lightbeam.
B. Second, scattering events will alter the exit angle or direction of the photons. Initially in the "ballistic" region the halfwidth of angles will be very small, but after many scattering events the halfwidth will become very large in a
"diffusive" region.
C. Third, absorption will preferentially remove photons that are scattered far from the central axis or that have very long pathlengths. Absorptivities vary with wavelength, so this effect will also vary.
D. Fourth, the pathlengths or flight times of the photons have a characteristic distribution. The mean pathlength increases with the sample thickness.
E. If the original light was polarized, the degree of depolarization defines a fifth variable over which a distribution can be measured.
For extremely thick or turbid samples, all the halfwidths will be comparable to the sample size in which case no images can be resolved.
However in many practical cases the image resolution can be improved by selecting only the central portions of each of these distributions. This improves the spatial resolution of an image at the least cost in terms of the signal intensity or
S/N ratio. The halfwidths for each distribution vary as a function of sample thickness (defined in units of the mean free path for scattering in FIG. 1). The mean free path for scattering is about 10-100 microns in biological tissues and bone, due in
large pan to interactions with cells of size similar to the wavelengths of visible and near infrared light. (The mean free path for absorption is somewhat longer, especially in a spectral window of low absorption between 0.5 and 1.4 micron.)
In practice, how much can image resolution be improved by selecting the central portion of the "bloom"?
Image resolution better than 4 mm can probably be achieved for mammography in a direct imaging mode, scanning with a single lightbeam. Time-averaging and multiple lightbeams should permit even better resolution. Consider the following.
A. First, the position at which a photon exits the sample can be recorded precisely by use of apertures and a scanning detector or detector array. Traditional methods of medical transillumination illuminate the body with a broad beam of light,
and inspect the distribution of light reflected from or transmitted through internal organs and tissues. The spatial resolution is limited to about 2 cm for a sample thickness of about 6-8 cm (e.g. mammography). Small veins near the surface of the skin
can be imaged with better resolution-the worst case is for an object halfway through the sample for which no improvement can be obtained by reversing the direction of illumination.
B. Second, the angle at which a photon exits the sample can be selected by the use of two or more apertures placed in front of a detector to form a collimator. Collimation is used to discard light that has been scattered into directions far
off-axis. Each scattering event usually changes a photon's direction only slightly. Collimation improves image resolution by recovering photons with ballistic trajectories, at the cost of much intensity. If a collimator accepts only light traveling
within about 3 degrees of the original axis (a solid angle of about 0.01 steradian), well over 99% of diffuse light would be filtered out. Collimation has been used to achieve a resolution of about 1 mm in biological samples 3 cm thick, but about ten
minutes were required to obtain sufficient signal to form an image. FIG. 1 suggests that collimation is most effective in thin samples for which ballistic photons are plentiful.
C. Third, the absorption of light will be greater for those photons which diffuse further outwards from the central axis and undergo longer trajectories. The graph in FIG. 1 indicates that discrimination by means of absorption is more effective
than collimation in thicker samples. The absolute amount of absorption must be taken into account in the identification of the most suitable MLR protocol and factors for an analysis. If two absorption peaks are equally acceptable for quantitative
analysis, then the one which rides on top of a broad absorption band might be preferable in order to improve the image resolution. But good contrast is necessary to achieve the best image quality, and may often impose the opposite choice
D. Fourth, pathlength is a highly effective means of discriminating against off-axis photons. The shortest pathlength is just the thickness of the sample, for which the transit time is determined by the speed of light in the sample. Scattered
and diffuse light will travel longer trajectories that take more time. The distance from the central axis increases as the square root of the pathlength or transit time, as expected for diffusion by means of a random walk. Shorter pathlengths can be
selected by means of gating the signal for a very short interval on the time axis, or more cheaply by use of a modulated signal and a phase offset in the frequency domain to select the mean pathlength. The graph in FIG. 1 indicates that pathlength
selection is the most effective means of improving image resolution. Modulation techniques are essential for chemical analysis as well, since accurate quantitative analysis in turbid solutions should ratio wavelengths of photons that have similar or
identical pathlengths.
E. The fifth consideration mentioned was polarization, for which little relevant experimental dam is available. It seems reasonable to assume that light scattering will affect angles and polarization similarly, so that this will be of most use
for ballistic rather than diffusive light.
How should the lightsource and detector be configured?
Three possible arrangements for imaging are as follows: rastering a single lightbeam in two dimensions, rastering a linear array of lightbeams in one dimension, or using a planar array of lightbeams with no rastering.
Rastering of a single lightbeam (e.g. in a left-to-right, top-to-bottom pattern like the electron beam in a television set) or a linear array (e.g. top-to-bottom only) may be accomplished in at least two ways. The first way is to actually move
the lightsource or an optical aperture connected with fiberoptics to the lightsource, in a plane perpendicular to the direction of the light. The detector, or an optical aperture connected with fiberoptics to the detector, would then be moved
synchronously in a second plane on the other side of the sample. A second, low-inertia way seems cheaper and quicker: use a set of stationary mirrors and lenses to amplify small motions of a directional mirror or a fiberoptic lightguide. Scanning with
a directional mirror typically makes use of an oscillating plane mino | | |
