1. Introduction

Optical imaging systems are highly desirable in medical imaging since optical methods are inherently non-ionizing. Also, tissue characteristics are wavelength dependent, allowing unique information to be gained at different wavelengths. For example, these methods offer the potential to measure blood content and oxygenation, which are surrogates for angiogenesis and hypermetabolism, respectively [1]. The basis for these measurements is the strong optical absorption of hemoglobin in blood and its associated oxygen-dependent spectral changes [2]. However, optical methods suffer from optical scatter, a serious limitation that prevents features below the surface from being visualized at high resolutions. For instance, confocal microscopy fails to provide high resolution images when imaging is deeper than a few hundred micrometers.

To overcome the limitation of optical scatter, several approaches have been investigated, including time-domain imaging, and spatial filtering. Time-domain imaging, which observes the first arriving photons which have shorter path lengths than the scattered light, is a technique that was first shown to be applicable to small animal imaging more than a decade ago [3], but has suffered from lengthy image acquisition times due to hardware inadequacies. Moreover, there has been much interest in developing a sub-millimeter resolution optical imaging method with wavelength independency, which can be useful for multispectral or hyperspectral imaging.

Another way of rejecting multiple scattered photons can be achieved by employing collimated detection by spatial filtering, which uses a lens to create the Fourier spectrum of the spatial distribution of light on the exit surface of a transilluminated object and then removes high-order frequencies using a small aperture placed at the focal point of the lens [4]. Through spatial filtering, an image can be formed using only the small angle deviated light emitted from the exit surface of the object. Image contrast deteriorates as the scattering level of the medium is increased due to the decreasing proportion of non-scattered light, which is a limiting factor to this method’s use in real tissue imaging applications. Moreover, the lens and aperture system has a spectral bandwidth limitation due to the diffraction dependency of multiple wavelengths sources, which makes it more difficult to use in multispectral/hyperspectral imaging systems.

Angular domain imaging [5] using a micro-tunnel array is a good candidate for such multi-spectral imaging since it is independent of the wavelength and coherency of the source. In fact, the ADI optical method can ultimately be useful not only for detection of absorbing targets in the scattering medium but also for bioluminescence and fluorescence imaging of biological tissue, such as in whole animal scanning of genetic reporters and molecular contrast agents. In this paper we are exploring angular domain imaging beginning with test phantoms using homogenous scattering fluids of µ′_s in the range of 0.8 to 4 cm^-1 with the thicknesses decreasing from 5 cm to 1 cm respectively. These will be compared to tests with chicken breast tissue, an inhomogeneous muscle tissue with a higher µ′_s in the range of 5 – 8cm^-1, and thicknesses up to 5 mm. This should be a good first test for the complexities of dense tissue biomedical imaging. All these experiments are using transmission of illumination through the test samples. ADI can be applicable for observing optical properties which can help histologists to test thicker tissue samples with a relatively larger field of view. Small animal imaging e.g. mouse pup (µ′_s=5–15 cm^-1) [11] is within the capabilities of angular domain imaging is one of the major goals of the authors.

In this paper, an explanation of the angular behavior of photons in tissue and an overview of the angular domain imaging principle are given, followed by an investigation into the multispectral feasibility of angular domain imaging. Experimental results are provided to confirm sub-millimeter spatial resolution for chicken breast tissue samples at different wavelengths.

2. Light tissue interaction

Within the last thirty years, a better understanding of light distribution in turbid media has been attained, providing an accurate prediction of light-tissue interaction. The fundamental mechanism of light transport in absorbing and scattering media was adapted from nuclear physics, where Chandrasekar calculated the fluencies of neutrons through different materials [6]. Hence, in most medical laser applications, light is regarded as a particle and not as a wave. The integro-differential equation describing the stationary photon transport for monochromatic light is given by the diffusion theory as follows [7, 8]:

where I(x,Ω) is the angular energy flux density at position x and angle Ω in Wm^-2sr^-1 and S(Ω’→ Ω) denotes the angular distribution of light scattered from angle Ω’ to Ω. In addition, µ_a in cm-1 denotes the absorption coefficient and µ_s in cm^-1 denotes the scattering coefficient.

S(Ω’→ Ω) also represents the scattering phase function and is normalized to unity when integrated over a 4π solid angle. The net anisotropy of scattering for an arbitrary phase function is named the g-factor, which is defined as follows:

A phase function S(Ω’→ Ω) that describes the experimentally observed distribution of scattering angles of photons after a single scattering event is the Henyey-Greenstein (HG) function [9]:

It has been found by Jacques et al. [10] that the HG function S(θ) also accurately describes the scattering of light in biological tissue. The important parameter of this phase function is the g-factor. Many common biological tissues have a high degree of forward scattering, with g values ranging from 0.79 to 0.98 for 633 nm wavelength illumination [11]. This forward scattering result in a much lower effective scattering quantity that can be calculated for tissue, known as the reduced scattering anisotropic coefficient, µ′_s. This reduced scattering coefficient takes into account the so-called “quasi-ballistic” or “snake photons” (i.e. photons which follow paths close to that of ballistic photons), and is defined according to the formula:

Described simply, a collimated beam of light decays exponentially along its path through a tissue layer of thickness d in accordance with the Beer-Lambert law [12]:

where I(d) is the intensity of transmitted light measured using a distance photodetector with a small aperture, I_ο, is the incident light intensity.

While the reduced scattering coefficient µ′_s characterizes a material for a given thickness of medium it is useful to recognize the relative levels of scattered light and information carrying photons. In order to quantify the scattering level compatible with ADI method, a metric used in this paper to describe the scattering level of a medium is termed the Scattering Ratio, SR. The SR value is defined as the number of photons that are scattered for every non-scattered photon that passes through the medium as the following.

For example, for SR=10⁶:1, we have 10⁶ photons that are scattered in the medium for every ballistic and quasi-ballistic (information carrying) photon that passes through the medium. Our measurement results in section 4 shows the SR value can be fitted to the exponential trend to get approximation for scattering coefficient of our test samples.

3. Angular domain imaging principle

It can be assumed that quasi-ballistic photons are closely confined within a small angle of their initial light source trajectory (i.e. they have a forward directed distribution) due to the anisotropy factor in tissue. However, highly scattered light, which has undergone numerous scatterings, has a nearly uniform or isotropic angular distribution. A Monte Carlo model of steady-state light transport in multi-layered tissues (MCML) by Jacques et al. [13] has demonstrated through simulation that non-scattered photons exit an ideal isotropic scattering medium with very small angular deviations from their initial entry trajectories. Extending this work, we have previously [5] used Monte Carlo simulations to show that exiting scattered photons with small angular deviations also follow the shortest path, and hence are quasi-ballistic in nature. Hence, a detection system using angular filtration with a small acceptance angle and an aligned, collimated source of light that is passing through a turbid media can be a practical method to reject most of the uniformly distributed scattered light while accepting quasi-ballistic photons with small angular deviations.

In conditions where the scattering of light within a medium is not too large, many researchers report filtering non-scattered photons by employing a collimator as an angular filter aligned co-axially with an incident beam of light. Jarry et al [14] in 1984 designed a scanning collimated transillumination system to image small metal objects embedded in 1.5 cm thick mammalian tissue. A more sophisticated computerized tomographic device described by Wist et al in 1993 [15] has been investigated employing thin collimated light beams (1.5 mm cross section) synchronized with a similarly collimated detector to increase contrast in lesions normally lost due to the detection of diffuse light. The study demonstrates that detection of a 1.5 mm wide target is possible in phantoms simulating breast tissues 6 mm thick, regardless of depth. Wang et al in 1993 [16] have investigated a similar technique that employs a lens and aperture system to produce the Fourier spectrum of the spatial distribution of light on the exit surface of a transilluminated object. In this technique, an image can be produced using only the light emitted at angles close to the normal to the exit surface of the object with a maximum 0.2° angular deviation.

The angular domain imaging method, introduced previously by the authors [5], operates by filtering out highly scattered light in a medium based on the observation that light tends to scatter in a nearly uniform angular distribution, while non-scattered ballistic and quasi-ballistic light remain closely confined within a small angle of the initial light source. In the implementation described in this paper ADI utilizes the creation of a micro-tunnel array with a sufficiently high aspect ratio, as determined by tunnel length over width and height, to form the basis of angular discrimination.

Figure 1 illustrates a basic experiment setup for ADI in transillumination mode, consisting of a laser source, a scattering medium with resolution target, the angular filter micro-tunnels with a small acceptance angle, 0.29o, and an image sensor. The angular filter (see Fig. 2) constructed for ADI experiments at high scattering ratios (SR ≥10⁶:1) consists of a parallel array of tunnels, 51 microns wide on 102 micron centers passing along a 1 cm long plate to obtain an aspect ratios of approximately 200:1. The tunnel size was selected to match an even number of pixels in the detector for best imaging, while the aspect ratio was around the limit for alignment of the system with the current setup. Because the tunnels are semicircular in geometry, there exists an angular acceptance angle variation from 0.11° to 0.29°. The angular filter acceptance angle must be tolerant enough to collect quasi-ballistic photons, which are believed to deviate by up to 0.2° [25]. The standard configuration for the angular filter (see Fig. 2) is a 2 cm wide linear array consisting of 1 cm long tunnels with widths of 51 µm, an approximate depth of 20 µm, and a pitch spacing of 102 µm. The micro-tunnel cross-sectional area is large enough such that each tunnel illuminates a consistent number of image sensor pixels. Current angular filter array do not collect the photons that hit the wall areas between the tunnels. This area represents a loss of approximately 60%. Switching to square cross section collimators would reduce this but create other complications. For example making the tunnel spacing narrower will allow overlapping of the accepted light from adjacent tunnel. This is being studied in other experiments underway. Also it is important to confine the light to the micro-tunnel line so as not to waste photons in areas not collected.

The linear array of micro-tunnels has a limited vertical field of view. Therefore, the scattering sample and test phantoms are incrementally raised by a computer controlled precision z-axis stage (with 0.05 µm repeatability). As the test sample is raised, one horizontal section of the sample is imaged through the angular filter at each step, and a final image is assembled using all the horizontal sections from each step. Hence, an entire region of the sample can be passed through the angular filter’s field of view and imaged.

 

Fig. 1 Basic ADI experiment setup in transillumination mode.

Download Full Size | PDF

To test the spatial resolution of our scans, resolution targets consisting of L-shaped patterned aluminum thin film on a glass slide, as shown in Fig. 3, have been fabricated for imaging. The resolution target is placed either within the scattering medium or at the front (i.e. closest to the laser). To test the resolution performance of our imaging technique, we have two sets of differently sized resolution targets, with each target composed of lines and spaces of equal width.

 

Fig. 2. Angular filter array.

Download Full Size | PDF

 

Fig. 3. Resolution target to measure the spatial resolution of ADI.

Download Full Size | PDF

The small resolution target slide with line and space widths of 200 µm, 150 µm, 100 µm, and 50 µm is used in our homogeneous scattering medium (e.g. milk/water solution) experiments, while the large resolution target slide with the line and space widths of 520 µm, 420 µm, 320 µm, and 220 µm is used in our heterogeneous scattering medium (e.g. chicken breast) experiments.

One important limitation of angular domain imaging is that near the detection limits a fraction of the highly scattered photons creates a background noise in the image. As highly scattered light is nearly uniformly distributed across all angles, there will be a small fraction that happens to exit the tissue within the micro-tunnel acceptance angle in addition to the ballistic and quasi-ballistic photons. This results in “background scattered light” that reduces the effective image contrast. Figure 4 shows two ADI images using an 808 nm laser. As the scattering level of the medium is increased from a modest SR=10³:1 (Fig. 4)) to a high SR=10⁷:1 (Fig. 4(b)), the fraction of the light that is ballistic or quasi-ballistic decreases significantly (by almost 10⁴), while the background scattered light stays nearly constant. This results in a decrease in the signal relative to the background noise. As a result, image contrast decreases (Fig. 4(b) has less contrast compared to Fig. 4(a)) as the scattering level increases to the point where the object patterns are submerged in the background scattered light level.

 

Fig. 4. Image of small resolution targets placed in a 2 cm optical cell filled with homogenous scattering milk/water solution using an 808 nm laser diode at (a) SR ≈ 10³:1 [µ’_s=1 cm^-1, µ’_a=0.01cm^-1]and (b) SR ≈ 10⁷:1 [µ’_s=2.5 cm^-1, µ’_a=0.01cm^-1]. (Total scan area is 6.6 mm by 6.5 mm)

Download Full Size | PDF

Most recently other authors have been exploring areas related to this ADI work. K. Shimizui et al in 2000 [18] reported employing collimators with a 0.57° acceptance angle to capture so-called “Near Axis Scattered Light” (NASL) using an optical computer tomography (CT) technique. With this technique, 1 cm thick samples can be detected within a 5 cm thick container filled with an aqueous solution of non-fat milk as the scattering medium (µ’_s=0.44 mm^-1).

The ADI technique works successfully in the transmission geometry, where the laser source is aligned to the angular filter to trans-illuminate the turbid medium from front to back (Fig. 1). ADI can also be employed using an illumination source on the same side as the angular filter to capture non-scattered photons generated deep from within the scattering medium. In this geometry, collimated light is injected into the turbid medium, thus generating a ball of illumination from inside the medium and behind the imaging target object. This newly formed light source in the scattering medium emits non-scattered and scattered light relative to the angular filter array micro-tunnels. When these back reflected photons pass an imaging target and reach the angular filter, the relative non-scattered photons are accepted through the micro-tunnels and reach the camera, whereas scattered photons are rejected by the filter. Reflection geometry ADI has been thoroughly investigated in [24] by the authors.

In the present paper, transmission geometry ADI is investigated to achieve multi-spectral sub-millimeter resolution optical imaging for relatively thick scattering media as compared to other high resolution optical imaging methods. The long term goal of this work is to allow imaging through moderately thick samples around 2- 3 cm in thickness of biological tissue with sub-millimeter spatial resolution, thus enabling imaging through a small animal target.

Although the authors have been investigating [26] how to improve image contrast in ADI, discussion about such contrast enhancement techniques is outside the scope of this paper, which instead focuses on the multi-spectral feasibility of the ADI technique.

4. Methodology

The basic ADI setup used in previous research by the authors [5–21] involves using a large-frame 488–514 nm Argon ion laser source due to the very high quality beam characteristics of that system. In our new experiments, three sets of experiments using several diode lasers with a collimation lens system have been employed in place of the Argon laser, thus allowing us to utilize longer wavelength diode sources in the red and near infrared spectrum. Fig. 5 shows our experiment setup with a scattering sample and submerged resolution targets resting on a computer-controlled vertical linear stage (with ±0.05µm resolution), where it is illuminated by a collimated line of light and imaged through the angular filter array by a CMOS camera detector.

 

Fig. 5. ADI setup with diode laser and a collimation system.

Download Full Size | PDF

Diode lasers are available in many visible to near infrared wavelengths, take much less space, have lower power supply requirements, and much lower costs compared to large frame lasers. However collimation is an area that presents challenges for diode lasers due to their small and asymmetric emission areas (on the order of one micron along the vertical axis by tens or hundreds of microns along the horizontal axis). As a result, the beam diverges rapidly along the horizontal axis and even more rapidly along the vertical axis. However, this asymmetric divergence can be corrected using an aspheric/cylindrical lens system, (shown in Fig. 5) that bends the light with a different power along each axis, and is thus able to shape the beam into a collimated line of light. The collimated beam then passes through an iris approximately 8 mm wide (slightly larger than the image sensor width) to restrict the width of the beam before it illuminates the sample. Although the 808 nm and 975 nm laser diodes have different beam divergences, the same experiment setup is used for both except with modification to the beam collimation optics, where the positions of the aspheric lens followed by two cylindrical lenses have been slightly changed to achieve proper beam collimation. The diode laser coherence length is much shorter (few mm) than the Argon laser (23 m). However, this is not a problem for ADI since a coherent source is not required.

A resolution target slide can be inserted at any position along the 5 cm cuvette, ranging from the front position (5 cm) facing the laser, to the back position (0 cm) facing the angular filter and camera. The angular filter is placed in front of a high-resolution (1280×1024 pixel) CMOS detector with a square pixel size (5.2 µm×5.2 µm) and is mounted on a 6 degree-offreedom jig that provides alignment to the laser source. Because the angular filter is a linear array of tunnels, only one horizontal line of the sample can be imaged through the angular filter at a given time. Thus, a two-dimensional image is captured by scanning line-by-line as the scattering medium is moved vertically while the illumination and detection equipment are kept stationary.

Calibrated scattering samples were prepared for ADI experiments made from 2% fat partially skimmed milk diluted in water. Milk was chosen as a scattering element because it exhibits similar properties to tissue and is highly scattering but with a low absorption coefficient.

The scattering level of the solution is measured by a collimated transmission measurement with a small acceptance angle. A collimated laser beam (1 mm diameter) is passed through a 1 cm thick glass cell filled with the milk/water solution. After this, the collimated transmittance is filtered through two 1 mm pinholes with an approximately 50 cm apart from each other, creating an angular filter with an approximately 0.11° acceptance angle and allowing straight or slightly deviated photons to pass through the pinholes and reach the photodetector. We have chosen this method of scattering sample measurement since it employs the same angular filtration concept as ADI with micromachined tunnels. Transmitted light from the medium is attenuated by the small acceptance angle formed by the pinholes and also by the reflection and refraction that occurs due to refractive index mismatch between the sample, glass, and air. These two effects may explain why the measured light intensity is lower than other collimated transmission measurements in the literature [17, 18].

 

Fig. 6. Collimated transmission measurement results; Scattering ratio versus concentration of 2% fat partly skimmed milk (in percentage) diluted by water in 1 cm container.

Download Full Size | PDF

Figure 6 plots out the relationship between the scattering ratio and the concentration of 2% fat partly skimmed milk in water (in percentage) for a 1 cm deep solution. Because the scattering coefficient for milk decreases at higher wavelengths, a higher concentration of milk at higher wavelengths is required to achieve the same level of scattering compared to the lower wavelengths. Exponential trends for all four wavelengths are visible in the graph across a broad range of SR values, where the 514 nm line [µ′_s ~115.2 cm^-1] has an exponential coefficient 2.3 times that of the 670 nm diode laser line [µ′_s ~50 cm^-1], 3.6 times that of the 808 nm line [µ′_s ~32 cm^-1], and roughly 5.5 times that of 975 nm line [µ′_s ~21 cm^-1].

Scattering ratios were measured based on the collimated transmission measurement through chicken breast samples sandwiched between two microscopic glass slides. These experiments allow us to investigate the behavior of attenuated light through real tissue at red and near infrared wavelengths. The transmittance light intensity to the detector is measured using the same setup as for the milk/water solutions. Fig. 7 shows different scattering ratio measurements at three different wavelengths with a vertical log scale and at different chicken breast sample thicknesses.

 

Fig. 7. Chicken breast scattering ratio measurement results with vertical log scale.

Download Full Size | PDF

As shown in Fig. 7, total light attenuation decreases as wavelength is increased. This is anticipated from previous research works [26, 28]. Fig. 7 shows the measured data fit to the exponential trend, which is in agreement with the Beer-Lambert law. Our extracted values for reduced scattering coefficient (µ′_s) of chicken breast tissue are 7.2 cm^-1 for 670 nm, 6.2 cm^-1 for 808 nm, and 5.1 cm^-1 for 975 nm, assuming the absorption coefficient is much less than the reduced scattering coefficient [26, 28].

5. Results and discussions

Expanding the operating wavelength of our ADI experiments from an Argon ion (488- 514 nm) laser to longer wavelengths (670, 808 and 975 nm) is an important step towards imaging biological tissues, which exhibit lower scattering coefficients at NIR wavelengths. Furthermore, utilizing laser sources of differing wavelengths confirms that ADI is capable of multi-spectral imaging and is not dependent on a particular wavelength for operation.

An imaging performance comparison has been previously made between the 670 nm wavelength diode laser source and the Argon laser in [23]. Results with all the diode lasers (see Fig. 8) are similar up to higher scattering ratios such as SR ~ 10⁶:1 (µ’_s=0.8 cm^-1, µ’_a=0.01cm^-1) with imaging results consistent with water, but with increasingly degraded image contrast as SR is increased. The 200, 150 and 100 micron line/space targets in this readily identified. These results appear to be on par with or better than previous imaging results with the Argon laser source, as shown in [23]. Fig. 8 shows scans at all three wavelengths, each for a sample at SR ~10⁶:1, with test patterns at the middle and front positions after image processing using histogram equalization and gamma curve correction. Improvement in scans for the 670 nm diode laser may be attributed to a better shaped and collimated beam produced by the collimation optics system, as compared with the 808 nm and 975 nm laser beams.

In order to progress from imaging synthetic/organic scattering samples (e.g. milk solutions) towards being able to image human tissues, it is useful to experiment with other biological tissues (e.g. chicken breast). Tissue samples were prepared from commercially available fresh chicken breast meat. Sections of tissue were cut out using a scalpel and then compressed by hand with firm pressure and secured between two glass microscope slides. Compression appears to reduce the thickness of the chicken sample on the order of a millimeter, only. Scans were conducted using the 670 nm, 808 nm, and 975 nm diode laser sources with an aspheric/cylindrical lens collimation system (shown in Fig. 5) for compressed chicken samples of different thicknesses.

 

Fig. 8. Resolution targets placed in diluted milk sample with SR ~10⁶:1 scattering media [µ’_s=0.8 cm^-1, µ’_a=0.01cm^-1]: at 2.5 cm depth in a 5 cm optical path cuvette with the (a) 670 nm, (b) 808 nm, or (c) 975 nm diode laser system; and at 5 cm depth with the (d) 670 nm, (e) 808 nm, or (f) 975 nm diode laser system. (All images are histogram equalized and gamma curve corrected)

Download Full Size | PDF

 

Fig. 9. Large resolution target slide in front of 2.2 ± 0.2 mm chicken breast with the (a) 670 nm, (b) 808 nm, or (c) 975 nm diode laser system. (Total scan area is approximately 7.8 mm×8.8 mm; white scale bar is 500 µm).

Download Full Size | PDF

Scans of the large resolution target slide placed in front of the 2.2 ± 0.2 mm chicken breast sample (closest to the laser source) at all three wavelengths are presented in Fig. 9. The scattering ratio measurement for the three wavelengths are SR=4.9×10³:1 for 670 nm, 2.7×10³:1 for 808 nm, and 1.8×10³:1 for 975 nm as taken from Fig. 7. These SR values are well below the detection limit (around 10⁷:1) reached in previous milk sample experiments. These scans indicate that the tissue samples are quite heterogeneous (especially when compared to the milk solution scans), with many uneven illuminations due to features in the image intrinsic to the tissue itself. These non-uniformities make it more difficult to distinguish the small resolution targets introduced in front of the sample for imaging. For all three wavelengths in Fig. 9, the three largest resolution targets (composed of 520, 420, or 320 µm wide lines and spaces) are resolvable in the image with their lines and spaces distinguishable. The 220 µm resolution target is only faintly detectable in the 670 nm image, and is not visible at the longer two wavelengths. These results demonstrate sub-millimeter resolution performance (320 µm or better) with the ADI system at all three wavelengths for a 2.2 mm thick chicken sample.

 

Fig. 10. Large resolution targets in front of a chicken breast sample of 4 mm thickness with the (a) 670 nm, (b) 808 nm, or (c) 975 nm diode laser system, and for 5 mm thick chicken breast with the (d) 670 nm, (e) 808 nm, or (f) 975 nm diode laser system. (Total scan area is approximately 7.8 mm×8.8 mm; white scale bar is 500 µm)

Download Full Size | PDF

Figure 10 shows scans at all three wavelengths of thicker chicken breast samples of approximately 4 mm and 5 mm. According to the measurements shown in Fig. 7, SR values of SR=1.8×10⁴:1, 8.3×10³:1, and 4.5×10³:1 for the 4 mm thick chicken breast sample and SR=3.7×10⁴:1, 1.5×10⁴:1, and 7.5×10³:1 for the 5 mm thick chicken breast sample were measured for 670, 808, and 975 nm wavelengths, respectively. Again, these SR values are well below the detection limit (around 10⁷:1) reached in previous milk sample experiments. Both samples exhibit better uniformity in the image, as compared to the 2.2 mm sample. For the 4 mm thick sample (Fig. 10(a)–10(c), only the two largest resolution targets (520 µm and 420 µm) are resolvable, while the smallest two resolution targets (320 µm and 220 µm) are not clearly visible. For the 5 mm thick sample (Fig. 10(d)–10(f), the largest resolution target (520 µm) is resolvable at 808 nm and 975 nm, but is barely distinguishable at 670 nm. The second largest target (420 µm) is no longer clearly detectable at 670 nm and is only partially resolvable at the 808 nm and 975 nm wavelengths. The two smallest resolution targets are not visible for the 5 mm thick scans. These results demonstrate sub-millimeter resolution for all three wavelengths in 4 mm thick chicken tissue and for 808 nm and 975 nm in 5 mm thick chicken tissue, with the largest resolution target visible but not clearly resolvable at 670 nm. In addition, these scans demonstrate that image contrast and detectability is improved as wavelength is increased from 670 nm to 808 nm and 975 nm. This trend is expected since Fig. 7 shows that light undergoes less scattering as wavelengths are increased from 670 nm to 975 nm.

Various image processing techniques can be employed to improve ADI results. The following image processing techniques are based on previous image processing work [24]. One operation that can be performed is to fill in the missing image information that is lost due to the approximate 51 µm gap between the angular filter tunnels. The approach taken to accomplish this involves vertically scanning the chicken tissue and resolution target slide followed by a second scan with the entire sample shifted laterally by 50 µm. The reconstruction process involves comparing the mean intensity value for a particular column from both the normal and laterally shifted images, and then selecting the column with the highest value and inserting it into the new reconstructed image.

 

Fig. 11. Large resolution targets in front of 3 mm chicken breast (a) without digital image processing, and (b) after digital image processing. (Total scan area is approximately 7.8 mm×8.8 mm)

Download Full Size | PDF

In order to enhance the definition, contrast, and detectability of the resolution targets, homomorphic filtering background subtraction using morphological opening and adaptive histogram equalization have been employed. The digital image processing details have been investigated in further details in [24]. As an example, this image processing has been performed on a 3 mm chicken breast sample ADI scan using the 808 nm diode laser system. The original scan (Fig. 11(a)) is processed and the final image (Fig. 11(b)) shows some improvement in contrast and detectability of the resolution target lines and spaces, especially for the 320 µm resolution target slide (top right in both images). Further exploration into different parameters and additional techniques for image processing should be conducted in order to further improve ADI scan results.

6. Conclusion

ADI with an angular filter array has been shown to be successful at multiple wavelengths with sub-millimeter resolution or better using different laser sources ranging from a 670 nm to a 975 nm diode laser. Note that the same micro-collimator angular filter worked over this wide wavelength range. Performance with the 670, 808, and 975 nm diode lasers is on par with or better than previous results with an Argon ion (488–514 nm) laser source for calibrated milk scattering solutions. Chicken breast tissue samples with a compressed thickness ranging from approximately 2 mm to 5 mm were imaged using 670, 808, and 975 nm wavelengths with 520 µm resolution or better. Resolution target lines and spaces of 520 µm were successfully resolved when imaged in front of 5 mm thick chicken breast tissue. Enhancement of the angular filter tunnels to improve scattered noise rejection and imaging performance can be investigated in future research, along with alternate designs for the angular filter array (e.g. new tunnel dimensions and aspect ratios). Since angular filter arrays are wavelength independent, different wavelength diode lasers can be introduced to create a true hyperspectral imaging, along with improved shaping of the beam into a thinner line of collimated light to enhance the signal to noise ratio.