1.1.

State of the Art in Assessment of Burn Injuries

The accuracy rate of current clinical assessment technique to differentiate between burn grades, based mainly on visual inspection by experienced surgeons, is only about 65% to 70%.^2,5 Highly accurate differentiation and delineation of burn wounds can potentially alter management, reduce length of hospital stay, and improve overall recovery for the burn patient. For instance, a noninvasive clinical diagnostic modality that could guide the treatment plan by predicting the healing outcome of second-degree burns and guide surgical delineation to minimize the scar formation would be of significant value. This idea has motivated the development of a wide variety of invasive and noninvasive diagnostic tools,² which are usually benchmarked against histology as the gold standard. These techniques include, for example, indocyanine green dye fluorescence imaging,^3,6 nuclear magnetic resonance imaging,⁷ ultrasonography,⁸ contact dielectric measurement at radio frequencies,⁹ multispectral optical reflection imaging,⁵ laser Doppler imaging,¹⁰ polarization-sensitive optical coherence tomography,¹¹ and near-infrared spectroscopy.^12,13 Although some of these techniques have shown promising results, they still have not been able to achieve sufficient specificity and sensitivity, as compared with histological assessment of wound biopsies. Moreover, significant problems have limited the implementation of such modalities in clinical environments, including cost effectiveness and compatibility to routine patient care.²

1.2.

Terahertz Modality for Characterization of Burn Wounds

The terahertz part of the electromagnetic spectrum, known as the so-called “THz gap,” is usually defined by the frequencies between 100 GHz and 10 THz (wavelengths from 3 mm to 30 μm).¹⁴ The high absorption of terahertz radiation by both bound and free water molecules provides a sensitive signal contrast for imaging applications.¹⁵ Difference in the water content of the tissue has been recently proposed as the main basis for many biomedical and biological applications of terahertz radiation.^16–18 Pickwell et al. furthermore showed that a double Debye dielectric relaxation model can be used to describe the terahertz response of healthy human skin,^19,20 which is very similar to the model often used for polar liquids.²¹ Pulsed-terahertz emissions have also been used to image and delineate basal cell carcinoma^22–24 and human breast cancer tumor margins,^25,26 based on the differences in the water content of the cancerous tumor versus healthy tissue.²⁷

Terahertz reflectometry and imaging has recently been proposed for the diagnosis of burn injuries ex vivo^28–30 and in vivo.^31–33 We have previously shown that the double Debye model, proposed by Pickwell et al.,¹⁹ can be extended to explain the higher terahertz reflectivity of moderate burn samples immediately postinjury.³⁰ In accordance with the general burn literature,² an increase in the water content of the burned tissue, due to formation of interstitial edema, has been experimentally observed using THz spectroscopy.^30–33 Furthermore, we proposed that the sensitivity of THz waves to the postburn reduction in the number and size of discrete normal skin structures, such as microvasculature, sweat glands, and hair follicles can be used as a source of signal contrast in THz burn imaging.^30,33

In this paper, we formulate a hypothesis for a burn diagnosis criterion based on in vivo terahertz reflection spectroscopy results obtained from second- and third-degree burns over a 72-hour period after injury. In order to do so, we first introduce an image processing approach to objectively quantify the density of skin structures (DOS) in histological sections of tissue samples. We then show that a strong statistical correlation exists between the THz observations and the DOS metric.

2.1.

THz-TDS Technique

The THz-TDS setup used in this study was custom-made and consisted of an 800 nm 50 fs Ti:Saphire laser (Micra, Coherent Inc., Santa Clara, California), which by impinging upon a biased photoconductive antenna (Zomega Terahertz Corporation, East Greenbush, New York), built with 100 μm gap on low-temperature GaAs, generated the terahertz waves. Terahertz radiation was first collimated and then focused, using a pair of off-axis parabolic gold mirrors, on the samples placed on a fused-silica imaging window. Upon reflection from the burned and normal skin samples, the terahertz waves were detected via the electro-optic sampling method in a 1 mm thick ZnTe crystal. The entire THz-TDS system was enclosed in a box and purged with dry ${\mathrm{N}}_2$ to eliminate the absorption features of ambient humidity from the experimental results.

2.2.

Self-Calibration and Signal Processing

Before employing the fast Fourier transform (FFT) to investigate the spectral dependence of normal and burned skin samples, a split cosine taper was multiplied by the first and the last 20 to 25 points of the time series, and additional zeros were padded to both ends. The FFT amplitudes of the samples were normalized by the FFT of the differential reference to eliminate the effect of the intrinsic system response function. Moreover, to account for a typical terahertz signal amplitude drift in between measurements, the signal peak from the first interface of the slab, between air and fused silica, was used to self-calibrate each measurement with that of the references.³⁰

3.1.

Density of Skin Structures

Histopathological study of wound biopsies remains the gold standard of the burn depth assessment despite its invasive and time-consuming nature. Even though there still does not exist any universally accepted benchmark for inference of burn depth from histological sections,³⁶ the accuracy of all other technological aids are usually measured against this method. The complexity of the dynamic molecular and cellular level changes, which skin constituents experience postburn, gives rise to most of the discrepancies in this field. For instance, while some studies indicate that the patency of microvasculature at the burn sites is the most critical factor in recovery,³⁶ other methods emphasize cellular and epithelial intactness as the main predictor for wound healing.² The presence of discrete skin structures, which are also known to be determinants of the healing likelihood of the burned skin,^{11–12,34,36} can potentially give rise to scattering of electromagnetic waves as they propagate through skin layers. In this section, we introduce a simple image processing method to objectively quantify the DOS in an attempt to relate our histopathological findings to the THz observations.

Figure 3(a) shows a flowchart diagram of the steps involved in the image processing of the histology sections to count each pixel associated with the discrete structures in the burned and normal skin samples. The algorithm first identifies the skin edge, and subsequently defines a search area approximately 400 μm deep in the tissue. It then recognizes tissue structures such as microvascular capillaries, hair follicles, sweat glands, and their skin duct based on the ratio of the red and blue components of the acquired image in comparison to the two detection thresholds, $t_1$ and $t_2$. To determine these threshold values, we used two groups of the histological sections as learning subsets: (1) deeply coagulated tissue with only minor superficial skin cracks (usually of lighter colors) and (2) burned tissue including only dark-colored hair follicles and shafts. Each group included two skin biopsy samples and four cross section images per biopsy. We varied the values of $t_1$ and $t_2$ within the search domain for the learning subsets until all pixels associated with white skin cracks or dark hair structures were correctly identified by the algorithm. The determined threshold values were then used for all remaining histological images that presented a more complex collection of skin structures. Finally, the program calculates the total areal density of such structures within the search area. This density value is the largest ($\sim 20\%$ to 30%) for healthy skin, and the more severely the tissue is damaged, the smaller the DOS will be. Figure 3(b) and 3(c) shows typical outcomes of the image processing routine for a full-thickness and partial-thickness burn sample, respectively.

Fig. 3

(a) Cross-functional flowchart of the image processing and density of skin structures (DOS) calculation routine. (b, c) Typical pseudo-colored figures of the burned skin cross-sections show the thresholds $t_1$ and $t_2$ used in detection of skin structures for a representative third- and second-degree burn, corresponding to Fig. 1(e) and 1(c), respectively. The edge of the skin is previously detected in the image processing algorithm, and therefore pixels outside the tissue margins are not considered.

Figure 4 maps out our experimental results in the form of a scatter plot, where the $x$-axis shows the average reflectivity of skin samples between 0.2 and 1 THz, while the $y$-axis shows the spectral slope or rate of change in reflectivity with respect to frequency at 72 h postinjury. The color code is determined by averaging the DOS metric from the image processing analysis of up to four histological sections of each tissue biopsy. Individual symbol types correspond to individual rats. The dotted $y=-x$ lines, which are vertically offset for clarity, are drawn here as a visual aid. It can be seen that while all normal samples are aggregated in the lower left quadrant of the figure, the burned tissue data are spread out according to the degree of deviation of the THz reflectivity from their respective normal controls. In this plot, the excess reflectivity along the $x$-axis is generally attributed to the formation of interstitial edema at the burned sites.^30–33 However, the change in the spectral slope of the THz reflectivity (THz color of the burns), which is shown in the $y$-axis, is not well understood. One possible explanation suggests that a potential change in the dispersive properties of normal skin as a result of the thermal insult as well as absorption by the water content of the tissue may be responsible for this observation. It may also be consistent with a reduction in electromagnetic scattering of terahertz waves due to a decrease in the density of discrete scatterers, i.e., the DOS value, as we observe here. Specifically, in all normal tissue samples, where the DOS is the largest, the normalized THz reflectivity showed a steep roll-off with higher frequencies, resulting in negative slope values along the $y$-axis for the blue symbols in Fig. 4. However, the burned samples demonstrated near-zero or positive spectral slopes, suggesting reduced scattering levels as the DOS value decreases.

Fig. 4

The severity of burn injuries can be mapped out based on the mean and spectral slope of their THz reflectivity at 72 h postinjury. The more severe burns (lower DOS values) are further apart from their normal (control) tissue (blue). The dotted lines ($y=-x$) are guide to the eye and color denotes DOS.

From this map, we infer that some of the second-degree injuries were severe enough to reach DOS values comparable to third-degree burns within 72 hours. For example in Fig. 4, the burn sample marked with $\mathrm{DOS}=6.7\%$ was created with a second-degree protocol, but in both THz spectroscopy response and DOS, it has reached values corresponding to a full-thickness burn. We can also confirm from histological sections that other samples showed signs of healing within the same period. For instance, the second-degree burn with DOS estimated at 15%, placed very closely to its normal (control) tissue on the scatter plot, illustrates this effect. Of the nine rats in our survival studies, we found only one outlier to this general trend (a nominally second-degree burn with DOS value of 7.7%).

These results suggest that while discrimination of burn severity cannot be ascertained solely based on the absolute value of the terahertz reflectivity, a combined measure of reflectivity ($R$) and spectral slope ($S$) can differentiate among them based on the DOS metric. We define this combined measure, $Z$, with a linear combination of $R$ and $S$, as given in Eq. (1),

Fig. 5

The anticorrelation between the DOS and $Z$ metrics ($a$ and $b$ coefficients are optimized for maximum specificity, resulting in $a=b=1$) are shown. The green-colored data point refers to the partial-thickness sample that exhibited signs of healing within 72 h postinjury.

Large scale experimental studies, complete with statistical models, are necessary before the usefulness of the $Z$-value criterion in predicting the healing outcome of partial-thickness burns can be verified. In order to develop such statistical models, the progress of different burn grades should be monitored over a 2- to 3-week period to determine which values of DOS and $Z$ correspond to the samples that naturally heal, while others require surgical intervention.