2.1.

Compound-Eye Burn Imager

To address the requirements for a portable low-cost burn assessment device, we developed an extremely compact compound-eye multispectral sensing platform based on a thin observational module by bound optics (TOMBO) approach.^23–26 In the TOMBO configuration, a lens array and mask are placed over a single image sensor. This array forms multiple images located in distinct areas of the image sensor array. By positioning bandpass filters in front of the lenslets in the array, the images of an object at different spectral windows can be acquired simultaneously on one sensor.

A schematic of the compound-eye multispectral is shown in Fig. 1. A commercial CMOS image sensor (XIMIA, MQ042RG-CM, $2048\times 2048\text{pixels}$, 10 bit) was divided into multiple imaging regions, each having a 6-mm focal length $f\#4$ objective lens (Edmund optics #63-714, original diameter 4 mm, ground to enable assembly with a pitch of 3.7 mm). Bandpass filters were then positioned above each lens. We employed five apertures having filters with transmission bands centered at 546, 677, 736, 856, and 966 nm. The measured transmission of the filters that include the transmission of the optics and the camera sensitivity is shown in Fig. 2, along with their locations on the camera sensor. These wavelengths were selected to provide reduced scattering coefficient maps suitable for burn severity assessment and also to enable quantification of oxy- and deoxyhemoglobin concentrations and tissue water fraction.

Fig. 1

(a) Schematic of the compound-eye design and (b) photograph of a completed device.

Fig. 2

(a) Normalized transmission sensitivity of the compound-eye camera for the five spectral channels. The normalization is such that the integrated sensitivity for each channel is 1. (b) Location of the subapertures on the image sensor for the five spectral channels.

The working distance of the camera was $\sim 280\mathrm{mm}$ and the image size for each aperture was $630\times 630\text{pixels}$, corresponding to a field of view of $\sim 165\times 165\mathrm{mm}$. The short baseline of the image array enables the slightly displaced spectral images to be coregistered by translating the images. This was achieved using the MATLAB function “imregtform.” Since there is no rotation among the apertures in the compound–eye camera, we do not expect any rotation. For in-plane object features, we do not expect any distortion due to binocular disparity; however, for out of plane features, there will be a small amount of disparity. Based on the subaperture spacing of $(\surd 2\times 3.7)=5.2\mathrm{mm}$, object distance of 280 mm, objective focal length of 6 mm, and an angle between the projector and camera of 15 deg, we calculate this disparity to be of the order of 1 pixel ($5.5\mu \mathrm{m}$) over a field of view of $\sim 55\mathrm{mm}$ (the size of the burn area imaged in these experiments). The dimensions of the complete compound-eye multispectral camera were $26\mathrm{mm}\times 26\mathrm{mm}\times 24\mathrm{mm}$ depth (along optic axis).

2.2.

Preclinical Rat Burn Experiment

Two male Sprague–Dawley rats weighing $\sim 350\mathrm{g}$ (Charles River Laboratories Inc., San Diego, California) were used in this study. Housing and care for animals were in accordance with UC Irvine’s University Laboratory Animal Resources. The burn model and study protocol were approved by the UC Irvine Institutional Animal Care and Use Committee (IACUC #AUP-17-137). Before the experiments, each rat was shaved along the lateral dorsal region of the body using electric clippers and depilated with Nair (Church and Dwight, Princeton, New Jersey). During experiments, the rats were anesthetized using a chamber with 5% isoflurane and then transferred to a heated measurement plate having a nosecone that delivered isoflurane (2%). This allowed the rat to be moved between imaging devices. At the completion of imaging, the rats were euthanized with sodium pentobarbital ($150\mathrm{mg}.{\mathrm{kg}}^{-1}$).

To generate repeatable burns, we followed the method outlined in our previous publications^17,18 using a brass burn “comb” comprising four $10\times 20\mathrm{mm}$ tines separated by 5 mm gaps (Fig. 3). The comb was heated to 100°C in an isotemp dry bath incubator (Thermo Fisher Scientific Inc., Pittsburgh, Pennsylvania). The heated comb was applied to the dorsum of the rats using its own weight only (313 g) for a duration of 8 s. From previous experience, this contact duration produces deep-partial to full-thickness burns.¹⁸

Fig. 3

(a) The brass burn tool used to create contact burns. (b) Location of burn placement on rat dorsum.

The rats were imaged using SFDI instruments before application of the burns and $\sim 1\mathrm{h}$ post burn injury. At the end of the experiment, the dorsal skin encompassing the burned area was excised and a punch biopsy (4 mm diameter) was taken. The tissue was fixed in 10% neutral buffered formalin then stored in 70% ethanol before being embedded in paraffin. Sections, $6\mu \mathrm{m}$ in thickness, were taken and stained with hematoxylin and eosin (H&E) for histological determination of burn severity assessed by collagen denaturation and hair follicle damage.

SFDI measurements were performed using the compound-eye imager, a benchtop laboratory–based image based on a liquid crystal tunable filter (LCTF) that has been used previously in our rat burn studies¹⁸ and a commercial SFDI device (Reflect ${\mathrm{RS}}^{(\mathrm{R})}$, modulim, Inc., Irvine, California) used in our porcine burn studies.²⁰ For the compound-eye and LCTF systems, the structured illumination was generated using a digital micromirror device (DMD) projector (Texas Instruments, Dallas, Texas) with a liquid light guide coupled to a 250-W quartz tungsten halogen lamp source (Newport Oriel, Irvine, California). This produced white light illumination at five equally spaced spatial frequencies from 0 to $0.20{\mathrm{mm}}^{-1}$ over a $90\times 60\mathrm{mm}$ field. To minimize specular reflection from the region of interest (ROI), the projected light was linearly polarized (Meadowlark Optics, Frederick, Colorado), and crossed polarizers were used in the detection light paths.

Images were captured using the compound-eye imager with an exposure duration of 31 ms. While this provided good signal to noise for the 677, 736, and 856 nm channels, the low sensor sensitivity at the extremes of the spectral range (546 and 966 nm) required a second exposure of 124 ms to obtain adequate signal to noise levels for these wavelengths. The requirement for two exposures could be removed by adding the appropriate neutral density filters to balance the sensitivity of the spectral channels or by adjusting the spectral irradiance of the illumination.

The LCTF imager (Nuance™, Perkin Elmer, Massachusetts) combines a 12-bit CCD camera and an LCTF tunable from 650 to 1100 nm, with an FWHM of $\sim 10\mathrm{nm}$. This system was used to sequentially collect images at (3 phases × 5 spatial frequencies × 17 equally spaced wavelengths) between 650 and 970 nm. The exposure duration varied from 36 ms at the peak of the camera sensitivity to 1100 ms at 970 nm where the sensitivity is low.

The Reflect ${\mathrm{RS}}^{(\mathrm{R})}$ SFDI is capable of imaging optical properties (${{\mu }_s}^{\prime }$, ${\mu }_a$,) of tissue over a $20\times 15\mathrm{cm}$ field of view. In contrast to the other two devices used here, the illumination is provided by LEDs and a broadband detection is used. The instrument was controlled using MI Acquire v1.34.00 software provided with the commercial device. Sinusoidal two-dimensional patterns are projected and images were captured at eight wavelengths centered at: 471, 526, 591, 621, 659, 691, 730, and 851 nm and at five spatial frequencies evenly spaced between 0 and $0.2{\mathrm{mm}}^{-1}$ according to a protocol that we have previously employed.¹⁹ In addition, a planar ($0{\mathrm{mm}}^{-1}$) measurement is also taken at a wavelength of 971 nm. This wavelength is not transmitted efficiently through the DMD projection system, therefore, to maximize the signal, the illumination beam path bypassed the DMD. The 659-, 731-, and 851-nm reduced scattering coefficients were used to fit power law scattering spectra of the form:

This relationship was then used to extrapolate a value for the scattering at 971 nm. We have described this approach previously.²⁷ The values of the absorption coefficients were then calculated using the planar reflectance and the extrapolated reduced scattering coefficient. Each ROI was imaged three consecutive times requiring a total of $\sim 90\mathrm{s}$.

For all SFDI systems, a measurement of a polydimethylsiloxane (PDMS) phantom having known optical properties (${{\mu }_s}^{\prime }$ and ${\mu }_a$) was measured under the same background lighting conditions and geometry as the rat measurements to provide a reference calibration. The same calibration phantom was used for all systems.

Color images were taken using a 14-megapixel digital camera (NEX-3, Sony Corporation of America, New York, New York) after each SFDI measurement.

3.1.

SFDI Imaging System Validation

Prior to commencing the burn experiment, a homogeneous PDMS reference phantom, having ${\mu }_a\sim 0.025{\mathrm{mm}}^{-1}$ and ${{\mu }_s}^{\prime }\sim 1.0{\mathrm{mm}}^{-1}$ at a wavelength of 700 nm (typical of tissue), was measured using all three systems. The optical property spectra from these measurements are shown in Fig. 4. The error bars are ± the standard deviations of the measured optical properties over the central 80% of the phantom. The feature around 910 nm in the ${\mu }_a$ spectrum resolved by the LCTF SFDI imaging system is due to an absorption peak of the PDMS. The measurements from compound-eye imager agreed to within $\pm 3\%$ for ${{\mu }_s}^{\prime }$ and $\pm 6\%$ for ${\mu }_a$ of the LCTF and Reflect ${\mathrm{RS}}^{(\mathrm{R})}$ measured optical properties. The measurements for the compound-eye system show a slightly larger standard deviation, which may be a result of the higher noise of the camera compared with the detectors used in the other two systems. Nevertheless, it is sufficient for tissue optical property measurements.

Fig. 4

Optical property spectra (${{\mu }_s}^{\prime }$ and ${\mu }_a$) of a PDMS reference phantom measured using the three SFDI systems used in this work, showing good agreement between the instruments. The error bars are the standard deviations of the measured optical properties over the central 80% of the phantom. The feature around 910 nm resolved by the LCTF SFDI imaging system is due to an absorption coefficient peak of PDMS.

3.2.

SFDI Imaging of Burns

The images of the H&E-stained biopsies are shown as Fig. 5. For each rat, the biopsies were taken from the most caudal burn (burn #4) centered on the ROI shown by the black circles in Fig. 6. For rat #1, the H&E histology shows that the collagen matrix has been degraded, and the hair follicles have been damaged through the entire dermis, indicating a full-thickness burn. For rat #2, the damage extends almost throughout the dermis; however, some hair follicles at the bottom of the dermis and close to the subcutaneous fat appear intact, indicating a deep partial-thickness burn, although the difference in burn severity for these burns is small.

Fig. 5

H&E histology (a) unburned skin, (b) rat #1 burn tissue, and (c) rat #2 burn tissue. The dotted line shows the depth of hyalinized collagen, and the circles indicate damaged hair follicles and lumen. For the burned tissue, the histology was taken from the ROI in the most caudal burn (position #4) for each rat.

Fig. 6

(a) Color images of the rat before and 1 h after application of the burn. (b) ${{\mu }_s}^{\prime }$ maps of the burn region before (upper) and after (lower) the burn for a wavelength of $\sim 690\mathrm{nm}$ for the LCTF and Reflect RS instruments and a center wavelength of 677 nm for the compound-eye imager. (c) Bar charts of ${{\mu }_s}^{\prime }$ pre- and 1 h postburn at the same wavelength as the maps, above for the ROIs indicated on the ${{\mu }_s}^{\prime }$ maps.

Figure 6 shows data obtained from rat #1 before and 1 h after application of the burn. Figure 6(a) presents the color images of the rat dorsum before and 1 h after the application of the burns. The burns are labeled from 1 to 4, where 1 is the most rostral and 4 is the most caudal. Figure 6(b) shows the maps of the SFDI-measured reduced scattering at a wavelength of $\sim 690\mathrm{nm}$. This wavelength was chosen as both the Reflect RS^® and LCTF systems report optical properties at this wavelength, and it is encompassed by compound-eye channel having center wavelength of 677 nm. This wavelength is also close to the wavelength that optical properties in our previous rat studies were reported (650 nm).¹⁸

The optical property maps from the Reflect ${\mathrm{RS}}^{(\mathrm{R})}$ system are more uniform across the rat dorsum. In this system, a profile measurement is taken, and the height and angle of the sample are then accounted for in the calculation of the reflected signal. This was not the case for the compound-eye and LCTF systems. However, this is not an issue for the burned region that was created along the spine of the rat and was positioned parallel to the measurement plane. Moving laterally away from the spine, there is a curvature of the rat dorsum that leads to a roll off in the scattering values. This can be seen clearly in the top of the scattering maps for these systems. For all three devices, the maps show a reduction in ${{\mu }_s}^{\prime }$ compared with the baseline measurement before the burn. Burns 2 and 3 are less homogeneous, which may be a result of poorer contact of the burn tool with the rat during the application of the burn tool due to the rat anatomy. In particular, burn 2 shows a smaller reduction in ${{\mu }_s}^{\prime }$ that may indicate a more superficial burn, although this would require verification with histology.

Figure 6(c) shows bar plots of ${{\mu }_s}^{\prime }$ for the 6-mm-diameter ROIs shown on the reduced scattering maps. These ROIs were selected in the center of the burns where the burn appeared most homogeneous, and these locations were then was colocated on the baseline ${{\mu }_s}^{\prime }$ maps to provide comparison. The mean of ${{\mu }_s}^{\prime }$ over the ROI is plotted along with the standard deviation.

In Fig. 7, we present the ${{\mu }_s}^{\prime }$ spectra for burns #4 for both rats for all three instruments. For legibility, the error bars are not plotted as they are of the order of the plot symbol size. For both rats, the spectra are very similar and show a reduction in ${{\mu }_s}^{\prime }$ at all wavelengths compared with baseline (preburn). The reduction in scattering is slightly less for rat #2, which may support the histological evidence that the burn is marginally less severe, although the changes are so small that it is difficult to draw this conclusion from a single burn.

Fig. 7

(a) and (c) Spectra of ${{\mu }_s}^{\prime }$ for ROI location #4 before (blue symbols) and 1 h after application of the burn (red symbols) for rats #1 and #2 for all three SFDI instruments. (b), (d) The ratio of in ${{\mu }_s}^{\prime }$ postburn to baseline (also for rats #1 and #2, burn #4).

At a wavelength of 650 nm, the baseline ${{\mu }_s}^{\prime }$ for rat #1 was measured to be $2.04{\mathrm{mm}}^{-1}$ (standard deviation of $0.08{\mathrm{mm}}^{-1}$), which decreased to $1.65{\mathrm{mm}}^{-1}$ (standard deviation of $0.06{\mathrm{mm}}^{-1}$) postburn. This agrees well with previously reported values of $2.10{\mathrm{mm}}^{-1}$, reducing to $1.66{\mathrm{mm}}^{-1}$ after a deep burn.¹⁸ At wavelengths shorter than $\sim 650\mathrm{nm}$, the reduction in ${{\mu }_s}^{\prime }$ for the burned tissue appears much greater. However, this corresponds to a wavelength range where ${{\mu }_s}^{\prime }$ is increasing rapidly. To clarify the changes in scattering, the ratio of the mean ${{\mu }_s}^{\prime }$ for the burned skin to the mean ${{\mu }_s}^{\prime }$ of the baseline ROI ${{\mu }_s}^{\prime }$ is also plotted for both burns. The plots [Figs. 7(b) and 7(d)] show that there is indeed a much larger fractional decrease in ${{\mu }_s}^{\prime }$ for shorter wavelengths. The reduction is $\sim 50\%$ at 470 nm that decreases to 70% at 600 nm before plateauing at 85% to 90% at $\sim 800\mathrm{nm}$. This behavior is likely due to the shallower penetration depth of the shorter wavelength light due to increased scattering and absorption. At 470 nm, most of the light samples the denatured collagen in the dermis, whereas at longer wavelengths the light penetrates beyond the dermis, sampling undamaged subcutaneous tissue. Of course, rat skin is significantly thinner than human skin, and so this variation with wavelength will be different in humans.