2.1.

Spatial Frequency Domain Imaging Method

SFDI is practiced by projecting spatially modulated illumination (typically periodic) patterns of various spatial frequencies over a large (many ${\mathrm{cm}}^2$) area of a sample and capturing the reflected images with a camera.^23,24 The demodulation of these spatially modulated waves characterizes the sample modulation transfer function (MTF), which encodes the optical property information-absorption and reduced scattering. Accelerated Monte Carlo-based analysis of MTF data results in two-dimensional maps of the quantitative absorption (${\mu }_a$) and reduced scattering (${{\mu }_s}^{\prime }$) optical properties via the least-squares fitting or look-up table approaches.^23,32 Height maps are calculated using the spatially varying phase to accommodate the measurement of curved surfaces in analysis.³³ When spectroscopy is desired, the SFDI measurement is repeated at multiple wavelengths, which results in a sparse spectrum of ${\mu }_a$, and ${{\mu }_s}^{\prime }$ optical properties versus a wavelength at each pixel. Recovered absorption data at an optimal combination of wavelengths can be used to fit for functional chromophore concentrations (Fig. 1), such as deoxy-hemoglobin (ctHHb), oxy-hemoglobin (${\mathrm{ctO}}_2\mathrm{Hb}$), hemoglobin (ctHbT), tissue oxygen saturation, and water (${\mathrm{ctH}}_2\mathrm{O}$).³⁴ The wavelength-dependent reduced scattering coefficient at each wavelength can be fit to the power-law decay:

Fig. 1

(a) Prototype imaging head, (b) imaging head taking data on portable cart, (c) brass burning tool.

In this equation, $A$ is proportional to the density of the scattering particles and $b$ is related to the average scatter size (collagen, nuclei, and mitochondria).³⁵ In skin, the collagen fibrils in the dermis are major contributors to the scattering signal.³⁶

2.2.

Spatial Frequency Domain Imaging Instrumentation

The SFDI device used in this study was a compact, clinic friendly, light emitting diode (LED)-based system developed by Modulated Imaging Incorporated (MI Inc., Irvine, California) for research purposes [Figs. 1(a) and 1(b)]. This general purpose SFDI system was designed for rapid imaging of a $15\times 20\mathrm{cm}$ field of view (FOV. Microcontroller electronics synchronized LED pulses with a digital micromirror device projection (Discovery 3000 $\mathrm{w}/0.7$” XGA LVDS DMD, Texas Instruments Inc., Dallas, Texas) and camera acquisition (LU160m, Lumenera Corp., Ottawa, Canada) to enable rapid image sequence and capture ($\sim 5\mathrm{ms}$ integration times) for all spatial patterns and illumination wavelengths. A custom thermoelectric board is used to maintain LED stability and temperature. Polarizers are placed in cross-polarized orientation after the source and before the detector to reject specularly reflected light from the tissue surface. The system simultaneously collected color photography with calibrated, color-balanced white light illumination, which enabled a standardized method of comparing SFDI results with visual clinical impressions. Data were acquired using 10 wavelengths in the system (400 to 850 nm) and at five spatial frequencies (0 to $0.2{\mathrm{mm}}^{-1}$). Height profile data at a single spatial frequency ($f_x=0.15$) were also acquired to correct for height and angle-dependent reflectance changes. Data were analyzed by fitting for optical properties at each wavelength, using a homogenous Monte Carlo SFD light transport model. A nonlinear least-squares fit to tissue reflectance was carried out for each pixel and wavelength using all five frequencies via a Levenberg–Marquardt algorithm implemented in C# software, where the forward model was based on precomputed Monte Carlo simulations. Subsequently, a linear least-squares fit was applied to transform the absorption maps to a tissue chromophore concentration. On a quad-core 3.3 GHz computer (Intel Core i5), optical property and tissue chromophore recovery took $\sim 54\mathrm{s}$ per dataset (or $14\mu \mathrm{s}/\text{pixel}$). For this analysis, chromophores were fit using only the NIR data (650 to 850 nm). Scatter cross-section and wavelength-dependent parameters were fit using full visible and NIR data.

2.3.

Preclinical Burn Model

All animal work was performed under Vanderbilt Institutional Animal Care and Use Committee (IACUC) and United States Army Animal Care and Use Review Office (ACURO) approved study protocols. Yorkshire pigs (mean weight = 65 lb) were anesthetized with an intramuscular injection of ketamine plus xylazine. The dorsal surface of the pig was clipped, shaved, and washed with mild soap and disinfected with povidone-iodine. Contact thermal burns were created on the pig skin using a square brass rod [Fig. 1(c)] heated to 100°C in boiling water.^37,38 The burn wounds in this study were imparted on the pig using a “reverse time course” model. On day 0, a total of six burn injuries were imparted on an animal; two wounds of each severity on either side of the animal (Fig. 2). The severity of the injury was controlled by altering the contact time; a 12 s contact time led to a superficial partial thickness thermal injury (green), 24 s to a deep partial thickness injury (yellow), and 75 s to a full thickness injury (red). Three wounds were placed on either side (six total) of the animal during a single session with one of each burn type located on a defined grid on the animal. Six additional burns (two of each severity) were imparted in a similar fashion for each subsequent day to achieve the time points of 24, 48, and 72 h. Burns made on the final day were allowed to progress for 2 h after each wound was introduced, including those of the final day. Thus, multiple burn wounds at all four desired time points (2, 24, 48, 72 h postinjury) could be harvested from the animal on the final day. Burns were placed on either side of the animals and a total of two animals were used in this study in order to obtain a total of 48 wounds. The “reverse time-course” model eliminates the need for intermediate punch biopsies at each time point, reducing the risk of wound complications due to biopsy caused infection and inflammation.^37,38 Additional SFDI measurements were taken immediately postburn and after 1 h, but there was no histology to represent this time point. SFDI data were collected using the system described above (Fig. 1) for each measurement time point. Histology procedures included two immunostains such as vimentin and Masson’s Trichrome. Vimentin is an immunomarker that identifies viable mesenchymal cells in burned skin. These cells represent the sites of activity where the wound still has viable cells beginning the wound healing steps. In a burn wound, a visible junction between positively stained cells and unstained cells can be used to assess the burn depth in a pig burn.⁷ Masson’s trichrome was used as a way to look at the morphology of the dermal collagen, which is a key contributor to the SFDI scattering signals.

Fig. 2

Diagram of reverse time course wound model employed in this study. Colors correlate to burn severity and arrows indicate new wounds. This sequence was repeated on both sides of the animal. After day 3, the animal was sacrificed and the tissue was salvaged for histology. Histology time points included $t=2$, 24, 48, and 72 h.

3.1.

Optical Property Maps of Burn Wounds

A total of 48 wounds were analyzed for this study, 16 of each severity (superficial partial, deep partial, and full thickness). The representative color and SFDI images of each wound type highlight the diverse nature of contrast for different burn depths. Color images captured the visual evolution of the burn wound over 72 h. As shown in Fig. 3(a), deep partial thickness and full thickness burns are nearly indistinguishable over the first 48 h to the visible eye. In Fig. 3(b), a calibrated planar (flood illumination) diffuse reflectance at a single NIR wavelength (658 nm) shows some graded contrast for each wound type over the course of the first 48 to 72 h. The wavelength of 658 nm was chosen as a representative in the middle of the visible and NIR range. We observed that the trends in contrast were similar at all wavelengths (see Sec. 3.2, Fig. 5 and related discussion for quantitative region of interest (ROI) values across all wavelengths). Separating reflectivity into absorption [Fig. 3(c)] and scattering [Fig. 3(d)] components using SFDI reveals that scattering appears to have earlier time sensitivity to burn severity that remained constant over 72 h. In the case of the most superficial burns, scattering values increase compared to surrounding normal tissue, but decrease in the case of the two more severe burns [Fig. 3(d)]. Multiwavelength SFDI analysis yielded NIR absorption spectra at each pixel that can be used to measure the tissue functionality, resulting in maps of ${\mathrm{stO}}_2$ [Fig. 3(e)] and tissue hemoglobin (images not shown for brevity but mean values are presented in Fig. 6). ${\mathrm{stO}}_2$ maps show a decrease in tissue saturation at early time points for the two most severe wounds. After 24 h, ${\mathrm{stO}}_2$ for the deep partial wound levels out while full thickness burns continue to decline [Fig. 3(e)]. As described in Sec. 2.1, the wavelength dependence of the tissue reduced scattering coefficient can be quantitatively modeled as a polydisperse collection of Mie scattering sphere particles, where the scattering coefficient is related to the size and density of the scattering objects in tissue. At each pixel, wavelength-dependent scattering data were fitted to a power law Eq. (1) to derive A (scattering cross section) and b (wavelength dependence) parameters using a center wavelength ${\lambda }_o=800$. As shown in Fig. 3(f), wavelength-dependent scatter power parameter (b) slightly increases compared to the surrounding normal tissue for superficial wounds and conversely decreases for the two more severe wounds. This result was consistent with the scattering results presented above.

Fig. 3

Derived SFDI maps for three burn severities highlighted the potential to monitor wounds over time. (a) Color images showed the similarity between wound types during acute stages of the injury. Single wavelength (b) reflectance images and (c) absorption (658 nm) differentiated among all three burn types after 72 h, (d) single wavelength scattering images showed differential contrast immediately postburn and was significant for each burn type after 24 h, multiwavelength absorption was used to quantify (e) tissue oxygen saturation (${\mathrm{stO}}_2$) and differentiated between burn types after 72 h. Finally, multiwavelength scattering was used to fit for the “b parameter” which showed differential contrast after 24 h.

3.2.

Quantitative Structural and Functional Tracking of Burned Tissue

To quantify the differential contrast among superficial partial, deep partial, and full thickness burns across all the burns, an ROI was chosen for each wound and the mean values for each recovered parameter were tracked over time. The ROI chosen was a $50\times 50\text{pixel}$ area ($\sim 15\times 15\mathrm{mm}$) in the middle of the wound. If there was a blister in the wound, this area was avoided in both histology and SFDI ROI selection. This analysis resulted in four mean ROI values per wound severity at each time point. A one-tailed paired t-test was performed for all mean ROI values at each time point to determine the significance for each recovered parameter. Single wavelength reflectance [Fig. 4(a)] at 658 nm differentiated among all three burn types after 72 h ($p<0.01$). Figure 4(a) shows that the reflectance values for superficial wounds are significantly different from those of the two most severe wounds after the initial burn injury. Figure 4(b) shows that by separating tissue reflectance into absorption and scattering parameters, 658 nm absorption does not differentiate any of the burn severities from each other except for the case of the most severe burn [Fig. 4(b)]. In this case, the most severe burn has a significant increase in absorption compared to the other wounds after 72 h ($p<0.01$). In contrast, Fig. 4(c) demonstrates that scattering values differentiate between all wound types 24 h ($p<0.01$) after the injury as opposed to 72 h. These results suggest that the sensitivity observed in single wavelength reflectance was likely due to scattering changes. Interestingly, scattering values clearly identified superficial partial thickness wounds from the deep partial and full thickness immediately postburn independent of the evolving functional status of the tissue. As we discuss below, this differentiation could be an early indicator of outcome and, hence, treatment plan.

Fig. 4

Region of interest (ROI) time course data for three different burn severities show that single wavelength (a) planar reflectance, SFDI-derived (b) absorption (${\mu }_a$) and (c) reduced scattering (e.g., ${{\mu }_s}^{\prime }$) measures are sensitive to burn depth. Time points with asterisk represent burn types that are significantly different compared to other burn types ($p<0.01$). Reflectance was significantly different for all three burn types after 72 h. Absorption was unable to differentiate among all three burn types after the first 72 h. In contrast, recovered scattering was significantly different for all three burn types after 24 h. Error bars represent the standard error between animals.

Other wavelengths show similar trends as those derived from the 658-nm optical properties. In Fig. 5, we plot the mean multiwavelength absorption and scattering [Figs. 5(a) and 5(b)] spectra at 24 h to show the wavelength-dependent sensitivity to burn severity, as well as the improved sensitivity of the scattering spectrum to burn injury severity as compared to that of the absorption spectrum. Wavelength-dependent absorption does not differentiate the three wound severities from each other at any wavelength after 24 h. However, recovered scattering values at four wavelengths (470, 658, 690, and 730 nm) differentiated all burn types from each other ($p<0.01$). Data acquired at 400 nm were excluded during analysis due to concerns about tissue autofluorescence.³⁹ Multiwavelength absorption data were fit to tissue chromophores (ctHbT, ${\mathrm{stO}}_2$) to describe the functional status of the tissue, and scattering was fit to describe the structural status of the tissue (collagen fibril diameter, density). Tissue chromophore values were able to differentiate between superficial burns and the deep partial/full thickness burns in the first 72 h ($p<0.01$). ${\mathrm{stO}}_2$ values differentiate the superficial wounds ($p<0.01$) prior to total blood volume values [Figs. 6(a) and 6(b)]. It is important to note that these functional parameters were not able to differentiate between the deep partial thickness and full thickness wounds in the first 72 h, possibly due to uncertainty in the final wound outcome. Wavelength-dependent scattering parameters showed a similar sensitivity to burn severity as that of a single wavelength scattering [Figs. 7(a) and 7(b)]. All wounds were separable ($p<0.01$) after 24 h for both A and b scattering-derived parameters. Superficial burns were distinguishable from the other severities immediately after the injury.

Fig. 5

Average wavelength dependent (a) absorption and (b) reduced scattering parameters 24 h postinjury for all three burn types. Near-infrared (NIR) absorption values were used to fit for tissue physiological parameters (tissue hemoglobin, tissue oxygen saturation) and visible/NIR scattering values were used to fit for power law. Error bars represent the standard error between animals.

Fig. 6

Tissue functional parameters were fit using NIR absorption values. Recovered values for (a) tissue-level hemoglobin were significantly different for full-thickness burns only after 72 h. Recovered values for (b) tissue oxygen saturation (${\mathrm{stO}}_2$) were significantly different for superficial partial thickness wounds within 2 h. Error bars represent the standard error for all animals.

Fig. 7

Wavelength-dependent scattering values for each burn type and at each time point were fit to a power law to approximate (a) scattering cross section and (b) wavelength-dependent parameter. Superficial partial thickness burns were immediately identifiable postburn from other burn types. However, deep partial and full thickness burns only became separable after 24 h. Error bars represent the standard error for all animals.

3.3.

Spatial Frequency Domain Imaging Scattering Parameters Correlated with Histological Burn Depth

Tissue histology was used as a gold standard metric of burn depth to validate the correlation between SFDI scattering values and burn severity. The vimentin immunostain was used to determine the burn depth at each time point (2, 24, 48, and 72 h) and for each type of burn (superficial partial, deep partial, and full). For this study, this junction was assessed at six locations across the center part of each wound and averaged to determine the average wound depth [Fig. 8(a)]. SFDI scattering values, averaged in the center region for each wound at a single wavelength (658 nm), showed a strong correlation ($r^2=0.94$) to measured burn depth for time points greater than 24 h and for all burn severities [Fig. 8(b)]. All time points and burn types for the corresponding recovered burn depths were plotted against A and b. A and b parameters were also strongly correlated ($r^2=0.89$ and 0.95 respectively) to burn depth [Fig. 8(b)]. We hypothesized that the differential contrast in scattering in burns was primarily due to the phase changes in tissue collagen. Thermal burns have been previously shown to cause denaturation of collagen, which disrupts the normal structure and distribution of fibrils.^12,40 As shown in Fig. 9, a Masson’s trichrome immunostain positively identified the changes in collagen structure for different burn severities, in support of our hypothesis. Normal tissue has a clear basket-weave structure [Fig. 9(a)] in a partial thickness burn [Fig. 9(b)]. In contrast, this basket-weave structure disappears as the tissue is damaged by the burn injury [Fig. 9(c)].

Fig. 8

(a) The burn depth was measured at multiple sites in the center of a superficial partial thickness wound and then averaged. (i) Healthy tissue was identified by positive vimentin stain and burn depth by determining (ii) margin between positive and negative stain. (b) Correlation between histology-based burn depth measurements and SFDI derived scattering signals at a single wavelength shows a strong negative correlation ($r^2>0.89$) for all measurements 24 h after injury. The same is true for wavelength-dependent scattering parameters (A, b).

Fig. 9

Representative changes in collagen structure due to a burn injury as seen by Masson’s trichrome stain. (a) The classic “basket-weave” pattern disappears as a (b) function of depth due to thermal denaturation depending on severity of burn, (c) burned tissue lacks this structure and appears to be a more homogenous structure.

3.4.

Wide-Field Snapshot Imaging of Burn Extent

A major utility of SFDI is wide-field mapping of both the functional and structural changes in tissue architecture. To demonstrate this utility, we created a wide-field ($28\times 10\mathrm{cm}$) map of reduced scattering (Fig. 10) for three different severities (superficial, deep partial, and full thickness) and four different time points (2, 24, 48, and 72 h) immediately before the animal was sacrificed. This larger FOV was achieved by stitching together two $15\times 10\mathrm{cm}$ SFDI images acquired prior to sacrifice. In Fig. 10(a), color images of burns of the same severity appear very different to the eye depending on the age of the burn injury. However, reduced scattering images of the burn appear to separate each burn type independent of the wound age (Fig. 10). The darker areas represent the most severe wounds and are consistent independent of the burn age. This capability highlights the ability to measure a number of wound severities in a single snapshot to characterize the distribution of wound severity.

Fig. 10

Overlay of color image and reduced scattering map ($\lambda =658\mathrm{nm}$) highlights the ability of SFDI to perform snapshot noninvasive wide-field quantitative mapping of burn severity and extent no matter age of the injury.