Novel multi‐spectral short‐wave infrared imaging for assessment of human burn wound depth

Abstract

Burn depth determination is critical for patient care but is currently lacking accuracy. Recent animal studies showed that Short Wave Infrared (SWIR) imaging can distinguish between superficial and deep burns. This is a first human study correlating reflectance of multiple SWIR bands using a SWIR assessment tool (SWAT) with burn depth classifications by surgeons and histology. Burns and adjacent normal skin in 11 patients with thermal injuries were imaged with visual and narrow bands centred at 1200, 1650, 1940 and 2250 nm and biopsies were taken from select areas. Reflectance intensities for each band in 273 regions of interest (ROI) were divided by the normal skin reflectance and combined into three Reflectance Indices (RIs). In addition, burns in ROIs and biopsies were classified by five surgeons and three pathologists, respectively, as superficial partial, deep partial, or full thickness. Results show that for burn depth increase classified by the surgeons, reflectance increased at 1200 and 2250, decreased at 1940, and didn't change at 1650 nm. In contrast, all three RIs increase with burn depth and predict the deep and full depths ROIs representing operable regions (Area Under Curve >0.6507, p < 0.0001). Pathologists' classification matched surgeons' classification of burn category only in eight of 21 biopsies (38.1%), but reflectance at all bands and one RI for all deep partial and full thickness biopsies were larger than in non‐biopsy normal and superficial partial thickness ROIs (p < 0.0118). In conclusion, multi‐spectral imaging with a new SWAT is a promising approach for evaluation of burn wound depth.

Abbreviations

AUC

area under the curve

H&E

haematoxylin and eosin

I

Light intensity

I ₀

Intensity of light source

I _i

Intensity of light incidence on burn area

I _r

Intensity of reflected light

I _λ‐raw

intensity of reflected light detected by the SWIR camera at band centered at wavelength λ (1200, 1650, 1940, or 2250 nm)

I _λ

intensity of reflected light detected by the SWIR camera and normalised by the normal skin reflected intensity at band centered at wavelength λ (1200, 1650, 1940, or 2250 nm)

ICG

indocyanine green

LDI

laser Doppler imaging

RI

reflectance index

RI₁

1st reflectance index calculated as I ₁₂₀₀ × I ₂₂₅₀

RI₂

2nd reflectance index calculated as I ₁₂₀₀ × I ₂₂₅₀/I ₁₉₄₀

RI₃

3rd reflectance index calculated as (I ₁₂₀₀ × I ₂₂₅₀)/(I ₁₆₅₀ × I ₁₉₄₀)

ROC

receiver operating characteristic

ROI

region of interest

SEM

standard error of the mean

SWAT

short‐wave infrared assessment tool

SWIR

short‐wave infrared

1. INTRODUCTION

In the United States, an estimated 1.2 million patients are annually affected by burn wounds. ¹ , ² Not only do these wounds detrimentally affect patients' lives, they also result in an enormous financial burden to the health care industry of more than $25 billion per year. ¹ U.S. military personnel have an increased incidence of injuries from burns compared to the general population. Overall prevalence of burns for military personnel was 19.5 burns/100,000 persons compared to a U.S. population that had an accidental burn prevalence of only 7.1/100,000 from 2003 to 2007. ³ These military burn injuries account for 5%–20% of wartime casualties, and burns sustained as a result of combat have increased rates of mortality and morbidity. ³ To avoid delayed wound healing, infections, and other adverse complications caused by deep partial or full‐thickness burns, practice guidelines call for early debridement and skin grafting procedures. ⁴ Currently, clinician assessment is the most commonly utilised method to evaluate burn depth and severity of injury; however, this only correlates with objective measures of tissue viability 60% of the time, leading to longer hospital stays and unfavourable long‐term functional outcomes. ¹ , ⁵ , ⁶ , ⁷ This current standard of care leads to differences in initial burn resuscitation and subsequent surgical and wound treatment by both experienced and inexperienced burn surgeons.

To improve patient outcomes, a more objective and standard method of burn depth evaluation is required. A standardised method to evaluate burn depth based on pathological assessment of burn tissue has been proposed based on animal model studies, ⁸ however recent discussion panels from the American and European burn associations recognise that inspection by surgeons is the most common practice for burn evaluation. ⁹ , ¹⁰ Technologies such as laser Doppler imaging (LDI), indocyanine green (ICG) videoangiography, thermography, and near‐infrared imaging have been used by surgeons for burn inspection but vary in target features, accuracy, sensitivity, and accessibility. ¹¹ , ¹² , ¹³ In a recent systematic review, LDI, which probes blood flow, was found to be the most favourable technology for assessing the potential healing of burn wounds, ¹¹ but LDI is not recommended for use during the first 48 h or after the initial 5 days of presentation, and is not used in patients with other comorbidities such as anemia, cellulitis, or vascular disease. ¹² , ¹³ ICG videoangiography is beneficial for assessing dynamic tissue perfusion but requires invasive injections, is expensive, and has potential side effects such as headache, rash, and anaphylaxis. ¹³ , ¹⁴ , ¹⁵ Thermography measures hemodynamic flow to the burn wound using the wound's temperature, ¹³ , ¹⁶ however it is influenced by environmental factors such as ambient room temperature, humidity, and wound evaporation. ¹³ , ¹⁷ Near‐infrared spectroscopy can be used to quantify edema in deeper burn tissues but lacks the ability to quantify more superficial tissues and also is dependent on blood flow, similar to LDI. ¹⁸ , ¹⁹ Thus, the most suitable technologies currently used to detect burn depths have restrictions that limit clinical applicability and demonstrate a need for improved technology to assess skin viability after burn injury.

The present study explores for the first time human burn wounds assessment by a new technology of short wave infrared (SWIR; light between 1000 and 2500 nm) imaging that has been proposed for characterisation biological tissue characteristics. ²⁰ The SWIR range includes prominent absorption peaks of water and lipids, that comprise up to 80% of all cells in the human body, as well as of collagen. ²¹ For example, alterations in water concentration are a hallmark of cell injury and death and the absorption of water at 1940 nm SWIR peak is 260‐fold greater than at the 970 nm near‐infrared peak. ²² Indeed SWIR imaging has been shown to map tissue water and lipid content in models of edema, inflammation, and tumour heterogeneity. ²³ In a previous study, we demonstrated that a SWIR imaging tool (SWAT) was able to quantify water content and differentiate superficial versus deep burn depths by the differing tissue reflectance at various wavelengths in both a proven mouse and porcine burn model. ²⁴ We therefore hypothesize that measurements at various SWIR wavelengths will help to improve burn assessment also in patients undergoing surgery for debridement of partial‐ and full‐thickness burns. In the present human study we provide for the time correlations of reflectance of burns at multiple SWIR narrow bands and their combinations, with the burn categories classified by surgeons and by histological characterisation. The study provides a proof‐of‐concept demonstration and motivates further investigation for the utility of the multi‐spectral SWAT approach for an objective measurement of wound burn characteristics.

2. MATERIALS AND METHODS

2.1. Patient enrolment

The study cohort consisted of patients presenting to the Parkland Hospital Trauma Center with recent burn injuries. This study complied with the Declaration of Helsinki, and study procedures underwent a full board review and were approved by the institutional review board at the University of Texas Southwestern and Parkland Hospital (STU‐2020‐0481). The inclusion criteria for the study specified male and female burn patients over the age of 18 with a total body surface area burn of 1%–80%. Patients were excluded if they were treated non‐operatively or had a history of keloids. Patients who met the inclusion criteria and agreed to participate in the study provided written informed consent allowing visual images, SWIR images, and tissue samples to be taken at the time of surgery.

2.2. Visual and SWIR imaging

Once patients were screened successfully and informed consent was obtained, they were enrolled in the study. SWAT was utilised to obtain high‐resolution images of burn wound areas in the operating room before burn debridement and excision. Prior to the imaging, the burn area was cleaned with a chlorohexidine scrub and tapped dry. Several sterilised 0 silk ETHICON sutures were placed on the burn area to precisely mark the location of the post‐imaging biopsy that would be taken for histological investigation and to enable a precise co‐alignment of the various images taken. After the images were taken, the areas marked by the silk suture were marked out with permanent marker so biopsies could be accurately collected after the skin was cleaned.

The SWAT used for imaging the burns is illustrated in Figure 1 and consists of the following main components: (i) a DC‐powered 0.5 kW light source that generates a broadband light illuminating via a diffuser collimator and an infrared dichroic mirror the skin and burn area, (ii) a SWIR camera capturing images of light reflected by the skin and burnt tissue at four different narrow wavelength bands. The four bands were separated and selected to span approximately even the SWIR range, including absorption peaks for lipids, collagen, and in particular the prominent water absorbance band at around 1940 nm, ²⁰ , ²⁵ and (iii) a visual camera that captured a standard view of the burn area using ambient light.

FIGURE 1.

The SWIR multispectral imaging unit used for the human burns study. (A) Diagram of visual and SWIR imaging of patient skin. (B) Two pictures of the SWAT development platform prototype setup in the operating room from two different angles with some components marked. The diagram in panel A and the pictures in panel B illustrate the following components and optical path of the developed system: A tungsten filament DC‐powered 0.5 kW lamp (1) that generates a broadband light I ₀ that is transmitted via a diffuser collimator (2) onto a dichroic mirror (3) that reflects the infrared incident light I _i (4) to illuminate the burn area on the patient skin (5). The skin tissue (5) reflects light diffusively, and rays of intensity I _r (6) that are a certain proportion of I _i (4) propagate toward the SWIR camera components (7–9). In its path to the SWIR camera (9), the reflected light I _r (6) passes through narrow wavelength band filters mounted on a filter wheel (7) for rapid exchange of bands. Afterwards, the filtered light passes through an adjustable iris aperture (8) and lens to form an image of raw intensities I _λ‐raw on the SWIR camera at each particular wavelength band (9). Finally, a visual camera (10) captures a standard view of the burn area under ambient illumination.

The light source was custom‐made and equipped with a Tungsten‐Halogen filament bulb (OSRAM XENOPHOT® NAED 54259; HLX 64663; 400 W‐36 V‐G6,35). The cutoff wavelength of the infrared dichroic mirror was 700 nm (Model CFW6, ThorLabs). The four different SWIR band filters mounted on a manual filter wheel for rapid exchange were centered at (with corresponding Full Width at Half Maximum; FWHM) 1200 (10), 1650 (12), and 2250 (500) nm (Thorlabs, Inc.), and 1940 (55) nm (Spectrogon). The increased FWHM with filter wavelength compensates for reduced illumination and SWIR camera sensitivity at longer wavelengths. Filtered images were captured via a Navitar f25 mm F/1.4 lens by the SWIR camera (Xeva‐2.35‐320, Xenics, Leuven, Belgium) that consisted of a type 2 strained sensor layer and was sensitive in the 1–2.35 μm wavelength range. For each of the four filters at different wavelengths, images of targets were acquired in 320 × 256 14‐bit pixel arrays collected as 10 frame long videos at 90 frames per second with an integration time of 6 ms per frame using Xeneth v2.6 software (Xenics, Leuven, Belgium). The camera output yielded pixel intensity values in analog‐to‐digital converter units. The recorded images were pre‐processed before analysis with a factory‐supplied software‐based filter compensating for aberrant pixel values and fixed pattern noise of the imaging sensor.

2.3. Processing and analysis of SWIR reflectance

Following iris and focus adjustments of the camera lens for each of the four filters, four videos of the area reflectance intensities (I _λ‐raw) were obtained at the four SWIR bands for each burn area and its surrounding normal skin a few centimetres outside of the burned skin. Subsequently, four images of I_λ‐raw were created from the time‐averaged videos for each of the wavelength bands and were co‐aligned based on the silk sutures markers. Thereafter regions of interest (ROIs, size >10 × 10 pixels) were digitally marked across the images, including areas designated for biopsies and area of normal skin, and excluding areas with saturated signals in any of the four bands. The average reflectance intensity signals I _λ‐raw at each wavelength was noted for each ROIs. Thereafter the reflectance in all ROIs imaged at each burn area of each patient were divided by the averaged reflectance of all ROIs at normal skin regions surrounding that burn area at their corresponding SWIR wavelength to create normalised reflectance intensity datasets I _λ (λ corresponding to either 1200, 1650, 1940 or 2250 nm centred bands).

To enhance the performance of the multi‐spectral SWIR reflectance analysis we generalised the approach of ratiometry in imaging. ²⁵ , ²⁶ Accordingly, analysis of the four wavelengths' reflectance also included product and ratio parametrization of the normalised reflectance intensities in three new Reflectance Indices (RIs). The RIs were defined as the following combinations: RI₁ = I ₁₂₀₀ × I ₁₂₅₀, RI₂ = I ₁₂₀₀ × I ₁₂₅₀/I ₁₉₄₀, and RI₃ = (I ₁₂₀₀ × I ₁₂₅₀)/(I ₁₆₅₀ × I ₁₉₄₀). The combinations were selected based on general direct and inverse trends of the normalised reflectance intensities as a function of the posteriorly observed burn depths.

2.4. Histology

The SWIR reflectance at ROIs were correlated with histological analyses of samples collected from the wounds. As a part of the standard of care, the surgeons debrided and excised tissue that was determined to be nonviable. The biopsies were 6 mm in diameter and taken with a punch biopsy. The areas were chosen from tissue that was determined to be deep partial or full‐thickness by the operating surgeon. Biopsies were obtained from areas that appeared to be visually homogenous for the full 6 mm biopsy area. The selected areas were marked with silk suture which was then outlined with permanent surgical marker so it could be cleaned. The tissues were then fixed using 10% formalin for 24 h and transferred to 70% ETOH for storage until they were removed and underwent paraffin embedding. The samples were then paraffin‐embedded and sectioned into 5 μm slices and stained with haematoxylin and eosin (H&E), trichrome, and picrosirius stains. A Hamamatsu NanoZoomer digital slide imager and software were used to scan the slides into digital images. Using the H&E staining, the level of tissue necrosis was assessed by three blinded pathologists. The pathologists determined the depth of burn by assessing the deepest level of irreversible burn necrosis through assessment of collagen, hair follicles, glands, and blood vessels. They then categorised the depths as either superficial partial thickness (extension into the papillary dermis), deep partial thickness burns (extension through the papillary dermis and into the reticular dermis), or full‐thickness burns (extension throughout the entire dermis into the subcutaneous tissue). ²⁷ After the three pathologists independently assessed the samples, any samples that did not have full consensus were discussed amongst the three pathologists. If all came to a consensus agreement on the sample, then it was used for analysis at the agreed upon burn depth. The non‐consensus burn depth samples were excluded from analysis. ²⁸

2.5. The surgeon burn survey

Burn wounds were assessed using visual light photographs obtained by Iphone Xr (iOS 17.3.1, Apple Inc., Cupertino, CA) from imaged areas that were collected at the same time as the SWIR images. De‐identified visual pictures of the burn areas were rescaled and aligned with SWIR images based on silk sutures placements on the burn area. Thereafter ROIs marked on the SWIR images, including biopsied areas, were marked on the visual picture of the burn and presented to five blinded board‐certified surgeons who filled out a survey to assess burn depth. The survey included the following categories: normal skin, superficial burn, superficial partial‐thickness burn, deep partial‐thickness burn, full‐thickness burn, or unable to determine from the picture. Each individual ROI was classified as a certain burn depth category if >60% of the surgeons agreed on the same burn depth. If there was not a >60% consensus, the areas were eliminated from further SWAT evaluation. The classification data were then compared to both the SWIR image findings and histology.

2.6. Operational and non‐operational burn categories

For a relevant surgical application of the burn classification by the SWAT, the ROIs with the burn categories above were re‐grouped into two categories: the ROIs with normal skin, superficial or superficial partial‐thickness burns were grouped under the non‐operational category (i.e., ROIs not requiring debridement), and the ROIs with deep partial‐thickness or full‐thickness burns were grouped under the operational category (i.e., ROIs requiring debridement). ²⁹

2.7. Statistical analysis

Pearson correlation was used to test the independence between the SWIR reflectance data at the different bands and the different burn depth categories. Following Shapiro–Wilk and Kolmogorov–Smirnov normality tests (p < 0.05), pairwise comparisons within burn conditions and nonpaired comparisons between burn depths (including histology) were conducted with Wilcoxon signed‐rank tests and Mann–Whitney U tests, respectively, or Student's t‐test where appropriate. Receiver operating characteristic (ROC) curve and area under the curve (AUC) analyses were used to quantify SWIR‐based separation of burn depths. AUCs were compared with the Delong test. ³⁰ Statistical significance was evaluated using two‐sided tests with an alpha of 0.05. Statistical analyses were conducted utilising SAS 9.4 and GraphPad Prism. A p < 0.05 was considered significant in all tests.

3. RESULTS

3.1. Patients

Eleven consecutive patients (2 female, 9 male) admitted to Parkland Hospital between June 2021 and January 2022 with thermal burns covering 3%–37% of total body area were included in the study. Of those, three identified as Black, three as Hispanic, and five as White. A total of 22 burn areas on the hands, arms, legs, torsos, and shoulders of the patients were imaged with visual and SWIR light.

3.2. Surgeons assessment of burn wounds

Visual light images were taken at the time of the SWIR imaging with approximately the same optical axis. Afterward, these images were marked with ROIs for surgeon assessment (Figure 2A) of the burn depth across the imaged burn areas in the 11 patients. These ROIs were then assessed by five blinded surgeons. Based on the ROI evaluation explained in the Section 2, 46% of ROIs were determined to be full‐thickness, 32% were deep partial‐thickness, 15% were superficial or superficial partial‐thickness, and 6% had no consensus (<60% agreement; Figure 2B). Excluding the ROIs with no consensus, a total of 273 ROIs, including 129 full‐thickness, 97 deep partial‐thickness, and 47 superficial or superficial partial‐thickness areas (Figure 2C) were selected for further evaluation using SWIR imaging.

FIGURE 2.

Surgeon analysis of collected burn data. The assessment of blinded surgeons of burn regions of interest (ROIs). (A) Representative images presented to surgeons for evaluation (by survey) of burn category in individual ROIs. Sample ROIs are marked by circles. (B) Overall composition of survey answers with burn level of each ROI determined by ≥60% surgeon answer congruence. (C) Composition of burn ROIs categories used for SWIR imaging analysis.

3.3. SWIR reflectance for different surgeons burns categories

SWIR and visual images were collected for 22 burn areas of 11 patients with 1–4 images per patient. Figure 3 shows representative images and collective SWIR reflectance intensities at the four SWIR bands for the three burn category groups, as determined by expert surgeon evaluation, and a comparison to normal skin from non‐burned areas of the same patients. Panel A shows a representative burn area imaged using visual light (left) as well as for 1200, 1650, 1940, and 2250 nm SWIR wavelengths. Though the images appear different, the silk sutures are visible in all and were used for spatial co‐alignments. The ROIs evaluated by surgeons were superimposed on the visual light and the 1200 nm images (ROIs were at the same locations for all 4 SWIR reflectance images). The reflectance value at each wavelength was noted for each of the total 273 ROIs along with the consensus classification of the burn depth category by the surgeons. Panel B shows the reflectance (I _λ) at those 273 burn ROIs for the four SWIR bands normalised by the average reflectance of normal skin ROIs, at each of the 22 imaged burn areas. A Pearson correlation value of 0.0026 confirmed independency between the reflectance and the wavelength bands for all burn categories. The reflectance however depended on the burn depth category. The most visible effect of burn depth on reflectance was at 1200 and 2250 nm wavelengths. The average normalised reflectance at 1200 nm was 1.0652 ± 0.0852 (mean ± SEM) for full‐thickness, 1.0375 ± 0.1144 for deep partial‐thickness, and 1.0206 ± 0.0852 for superficial and superficial partial‐thickness burns, respectively. The mean reflectance at 2250 nm was 1.0501 ± 0.0737 for full‐thickness, 1.0326 ± 0.0914 for deep partial‐thickness, and 1.0293 ± 0.0895 for superficial and superficial partial‐thickness burns. Using Mann–Whitney U tests, there were significant differences between deep partial‐ and full‐thickness burns at 1200 nm (p = 0.0012) and 2250 nm (p = 0.0027), and between full‐ and superficial‐thickness at 1200 nm (p = 0.0056). When assessing the 1940 nm wavelength, superficial and superficial partial thickness burns were significantly different than the deep partial‐thickness burns (p = 0.05) and significantly different than full thickness burns (p = 0.009).

FIGURE 3.

Reflectance intensity at different wavelengths and different burn depths. (A) Representative burn area imaged at visual light as well as at 1200, 1650, 1940, and 2250 nm light. The ROIs evaluated by surgeons are superimposed on the visual image, and the same ROIs used for the SWIR reflectance analysis are also shown on the 1200 nm image (ROIs are not shown on other SWIR images for the sake of clarity). (B) SWIR reflectance intensity normalised by normal skin average reflectance intensity at the four SWIR bands for the three burn depth categories as determined by expert surgeon evaluation. (C) Reflectance Indices based on combining information from the four wavelengths analysed in B are shown for the normal and three burn depth categories (see Section 2). Boxes show medians and interquartile. Symbols show means ± SE. Numbers inside boxes are the numbers of ROIs. Asterisks indicate normal distributions and horizontal bars indicate comparisons in which p < 0.05.

To further quantify burn ROIs to identify those requiring debridement (i.e., superficial partial‐versus deep partial‐thickness burns), we developed a novel parametrization of reflectance based on the independent reflectance intensities I ₁₂₀₀, I ₂₂₅₀, I ₁₆₅₀, and I ₁₉₄₀ of the four corresponding SWIR wavelengths. Based on the approximate trends of the normalised intensities observed in Figure 3B, we defined three new Reflectance Indices (RIs, see Section 2) and calculated their values for each ROI in the images, in addition to I _λ. Figure 3C shows a general increase in the three RI ³¹ values from normal skin to superficial partial‐, deep partial‐, and full‐thickness burn depths, with most cases, but not all, showing significant differences. RI₃ shows a consistent increase of the median and mean values with burn severity categories whereby the mean difference between full‐thickness burns and of normal skin for was 0.15 (p < 0.0001). At (mean ± SEM) 1.1136 ± 0.0184, the RI₃ for the operational deep‐thickness ROIs was larger than the 1.0642 ± 0.0192 for non‐operational superficial thickness ROIs by about 0.05 (p = 0.0328), which is about 3‐fold the difference observed between the mean values of I ₁₂₀₀ of 1.0375 ± 0.0116 and 1.0206 ± 0.0124, respectively, for ROIs with these two burn categories (Figure 3B).

3.4. Operational vs. non‐operational ROIs based on SWIR reflectance

We proceed with analysing the sensitivity and specificity of SWIR reflectance parameters in distinguishing between the two groups of operational and non‐operational ROIs (see Section 2). In Figure 4, we present the ROC analysis for all SWIR reflectance parameters presented in Panels B and C of Figure 3. Overall, all the parameters, except I ₁₆₅₀, showed an ability to detect operational ROIs based on the surgeons' classification with AUCs significantly larger than the non‐discriminatory diagonal line with AUC = 0.5. In general, the AUC for the three reflectance indices were larger than for the reflectance of the four individual wavelengths. In particular, the reflectance index RI₃, which is a combination of all four reflectance parameters at the four individual SWIR bands in our SWAT, presented the highest AUC of 0.7565, which was higher than the highest AUC of the individual wavelength band I ₁₂₀₀ of 0.6513 (p = 0.0109). Thus, combining the reflectance at the multiple SWIR band into a single reflectance index enhanced the detection of burns ROIs determined to be debrided.

FIGURE 4.

ROC and AUC analysis. Sensitivity and specificity of four I _λ and three RIs for detection of operational deep partial and full thickness ROIs as classified by surgeons. Probability of AUC parameters is reported against the no discrimination AUC = 0.5.

3.5. Pathologists assessment of burns categories

For the 11 patients, the reflectance parameters in the multi‐wavelength SWIR imaging were correlated with histological analyses of the biopsies collected from burn wounds as part of standard‐of‐care excision and debridement. Before patient tissue was collected, circular areas of silk suture were placed on the skin so that the biopsied areas could be identified in both the visual images and the multiple SWIR images. Thirty‐one tissue samples excised from the burns following their multi‐wavelength SWIR imaging were processed and analysed histologically.

Figure 5 shows representative examples of the process. In Figure 5A, visual images show burn areas with marked silk loops of deep partial‐thickness (left) and full‐thickness (right) where punch biopsies were to be collected. Figure 5B shows the RI₃ images of the two corresponding burn areas based on the multiple SWIR wavelength imaging prior to the biopsy collection. The RI values at the ROI at the center of the silk loop were 1.064 and 1.189 for the deep partial (left) and full (right) burns, respectively, consistent with the average RI₃ values in Figure 3C. Figure 5C shows two samples in which the necrosis borders were marked and the depth category of the burn was determined by three pathologist (see Section 2) to be deep partial‐ (left) and full‐thickness (right).

FIGURE 5.

Histological assessment of burn wounds. (A) Visual images of burns with superimposed silk sutures. Encircled silk loops mark regions for biopsies to determine burn depths by histology. These regions were classified by surgeons to be deep partial thickness (Orange circle, left) and full thickness (Red circle, right) burns. (B) A reflectance index RI₃ images processed from the SWIR wavelengths images taken prior to the biopsies in the burn areas illustrated in A. The RI₃ values at the center of the biopsied regions are superimposed, showing smaller values for deep‐partial thickness (left) than for full‐thickness (right). Thicker silk lines in right versus left RI₃ image is due to higher magnification of imaging (camera closer to patient). (C) Representative H&E‐stained biopsy sections with blinded pathologists' markings of the border of the burn injury. Examples of deep partial‐thickness (left), and full‐thickness (right) burn categories are shown.

A similar analysis was performed on 31 samples, of which seven were determined to be too shallow for histologically based burn category classification, and in three the pathologists did not reach a classification consensus. The remaining 21 samples were classified by the pathologists as superficial partial thickness (n = 3), deep partial thickness (n = 13), or full thickness (n = 5). Comparing the pathologists' classification of samples from a particular ROI to the surgeons' classification, only eight out of the 21 ROIs showed similar deep partial and full thickness classifications (38.1%). Of those eight similarly classified ROIs, five were deep partial burns and three were full burns. Overall, the category classification by the pathologists did not depend on whether there was or there was not an agreement with the surgeons' classification (χ² test: p = 0.4101).

3.6. SWIR reflectance for pathologists and surgeons operational burns

The most sensitive and specific reflectance parameter for the surgeons' ROIs classification, RI₃, did not reach a significant distinction between the five deep partial and three full thickness ROIs agreed by both surgeons and pathologists, and in general the number of biopsy samples was restricted and low, preventing a rigorous investigation. Thus, we focus on characterising correlations and differences in reflectance parameters grouped by the two classifications. Figure 6 presents correlations in all reflectance parameters for normal and superficial partial thickness (i.e., non‐operational) vs. the deep partial and full thickness (i.e., operational), and for the non‐biopsy vs. biopsy ROIs. Panel A shows that ROIs at biopsies classified as either deep partial of full thickness had their reflectance I_λ for 1200, 1650, 1940, and 2250 nm higher than for the non‐biopsy ROIs classified as normal or superficial partial thickness (1.08 ± 0.0298 vs. 1.0 ± 0.0338, 1.03 ± 0.0105 vs. 0.99 ± 0.0028, 1.03 ± 0.0102 vs. 0.99 ± 0.0028, and 1.08 ± 0.0227 vs. 1.0 ± 0.0045, respectively; p < 0.0118). I_λ for 1200 and 2250 nm did not differ between the biopsy and non‐biopsy ROIs with deep partial and full thickness. Panel B shows RI values for those ROIs equal to 1.1731 ± 0.057, 1.1358 ± 0.057, and 1.0941 ± 0.0483 for RI₁, RI₂, and RI₃, respectively, but only RI₁ showed difference relative to its value at the non‐biopsy normal and superficial partial ROIs (1.0113 ± 0.0082, p = 0.0041). Values for the three RIs of the ROIs with deep partial and full thickness biopsies did not differ from the non‐biopsy ROIs with the same surgeons' classification. Characteristics of the eight ROIs for which pathologists and surgeons agree on deep partial and full thickness classifications were inconclusive with non‐distinction in most reflectance parameters, and only higher I ₁₉₄₀ = 1.04 ± 0.0199 compared to non‐biopsy normal and superficial ROIs (0.99 ± 0.0028, p = 0.0133) and higher than non‐biopsy deep partial and full thickness ROIs (1.0 ± 0.0045, p = 0.0007), as well as lower RI₃ in biopsy compared with non‐biopsy ROIs (1.0092 ± 0.0507 vs. 1.1406 ± 0.0101, p = 0.0102). Overall, various reflectance parameters presented in Figure 6 demonstrate correlations and incompatibilities between operational ROIs in biopsies and non‐operational burns in non‐biopsied ROIs data groups. It should be emphasised that despite the incongruousness between the surgeons and pathologists in classifying operational deep partial and full thickness burns in biopsy ROIs, all four bands and RI₁ demonstrated reflectance that is significantly different from the non‐biopsy and non‐operational ROIs.

FIGURE 6.

Comparison of reflectance parameters for biopsy and non‐biopsy ROIs. ROIs are grouped in four data sets: (i) non‐biopsies classified by surgeons as normal or superficial partial thickness (non‐operational), (ii) non‐biopsies classified by surgeons as deep partial or full thickness (operational), (iii) biopsies classified by pathologists as deep partial or full thickness (operational) regardless of agreement with classification by surgeons, and (iv) biopsies classified by pathologists as deep partial or full thickness (operational) in agreement with classification by surgeons. (A) Reflectance intensities (I _λ) at four bands. (B) Values for three RIs of the same ROI groups as in A. Boxes show medians and interquartile. Symbols show means ± SE. Numbers inside boxes are the numbers of ROIs. Asterisks indicate normal distributions and horizontal bars indicate comparisons in which p < 0.05.

4. DISCUSSION

The study presents the first in‐human imaging and analysis of a multi‐spectral SWIR assessment tool (SWAT) for burn depth evaluation. The main findings demonstrate distinct reflectance intensities for narrow SWIR bands centred at 1200, 1650, 1940 and 2250 nm with various degrees of correlations to burn categories classified by surgeons and pathologists. When considering the surgeons classification, superficial partial‐thickness burn depths had different reflectance than deep partial‐ and full‐thickness wavelengths at 1200, 1940 and 2250 nm as well as reflectance indices comprising combinations of the bands. Reflectance at all bands, except 1650 nm, as well as their indices demonstrated ability to predict burns with either deep partial or full thickness that are clinically targeted for debridement. Although burn depths classification based on histological evaluation by pathologists matched the surgeons classification only partially, differentiation between operational and non‐operational burns was also attainable, albeit by a reduced number of reflectance parameters as compared to the classification by surgeons. In general, our data show that the multi‐spectral SWIR images provide reflectance measures associated with distinct burn wound depths and motivate additional research and developments for determining physiological mechanisms and clinical applications.

4.1. SWIR imaging and tissue characteristics

Imaging in the SWIR range offers unique opportunities in burn depth characterisation in‐vivo. Our study focuses on the quantification of the SWIR reflectance intensity and the analyses presented in Figures 3, 4 and 6 support the utilisation of narrow SWIR bands to distinguish between operational and non‐operation burn areas. Considering the light‐tissue interactions should give insight into factors contributing to our observations. The minimal autofluorescence of biological tissue in the SWIR range leads to increased specificity in molecular probing, ³² , ³³ while the significantly reduced light attenuation from scattering and absorption by the blood and other structures enables imaging with high spatiotemporal resolution and millimetre‐to‐centimetre scale penetration depth. ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ The SWIR optical band has been shown to reveal the presence and condition of lipids, fats, glucose, collagen, and many other molecules, owing to their distinct vibrational overtone levels and absorption spectral profiles. ⁴⁰ , ⁴¹ , ⁴² Ions, water, and compartmental volume shifts between spaces are also thought to affect intrinsic optical signals. ⁴³ , ⁴⁴ Most relevant to our study, blood absorbance in the SWIR range has several peaks, the most prominent one being near 1940 nm related to water absorbance, and troughs near 1200 and 1650 nm. ⁴⁵ Overall, the reflectance at the bands explored in our study to burn depths has been shown to be heterogenous and likely to reflect the complex transformations of the tissue characteristics in burns. For example, Figure 3 shows I₁₂₀₀ and I₂₂₅₀ increase with burn depth, while I₁₉₄₀ is decreasing, reflecting a multi‐factorial and complex processes with possibly overwhelming water content increase associated with edema at 1940 nm. ⁴⁶ Nevertheless, as the water‐air barrier in burns is compromised, patients can also undergo dehydration following a transdermal fluid loss. ⁴⁷ Thus, the precise origin of the increase or decrease in reflectance with burn depth at the various SWIR bands studied in this study are not known and additional research, including the utilisation of additional SWIR bands, machine learning algorithms, ⁴⁸ as well as modalities such as THz imaging which has shown the formation and progression of edema, ⁴⁹ is required.

4.2. Clinical implications

The standard of care for distinguishing wound depths is currently based on clinical physical examinations. Physical examination has limited accuracy, as no clear visual markers exist and wounds often evolve, making it difficult to decide when and if to intervene and operate. The crux of decision‐making for wound care surgeons lies in whether the tissue is viable, and currently, no portable, accurate, objective modality exists to help clinicians make this decision. Furthermore, once the decision is made that the patient needs an operation, it is difficult to assess the point at which all nonviable tissues have been removed. Objective measurements of tissue viability, such as histology from tissue biopsies, are impractical, expensive, invasive, and subject to sampling errors. Most tissue optics research to date, such as for the previously discussed LDI, has employed wavelengths in the visible and near‐infrared regions of the spectrum (400–1000 nm). ¹³ , ²⁰ Although these technologies detect deep vascularity, such data do not always correlate with superficial tissue viability. ¹⁴ , ¹⁵ Thus, the clinical applicability and accuracy of existing wound care diagnostic tools remain limited.

Our SWAT device has several advantages over other methods that are currently used to quantify burn depth and tissue moisture. LDI and thermography cannot be used intraoperatively on extremity burns that are debrided under a tourniquet given that they require blood flow. In our study surgeon‐determined matched only partially histologically determined deep partial and full‐thickness reflectance intensities consistent with a previous study ⁵⁰ however, the SWAT device's measurement of real‐time tissue characteristic, including moisture, gave readings consistent with assessments for debridement by both surgeons and histology under these conditions. ²⁴ , ⁵⁰ As opposed to ICG videoangiography, the SWAT device is non‐invasive and allows for immediate measurement of burn depth without the use of injectable fluorescent dyes that may cause allergic reactions. ¹⁴ , ¹⁵ In comparison to near‐infrared spectroscopy, ¹⁸ , ¹⁹ the SWIR wavelengths are more sensitive to tissue water content, ⁴⁵ allowing the evaluation of both the superficial and deeper dermis for a more complete analysis of the burn. Additionally, since SWIR is not in the visible range, overhead lighting and skin pigmentation that can affect other technologies have significantly less effect on the SWIR readings. ⁵¹ , ⁵²

4.3. Limitations

This study analysing SWIR imaging data as a method for burn depth determination has several limitations. Recent studies have shown that SWIR technology can only assess the wound to a depth of 4 mm depending on the wavelength and capture methodology. ²³ While this is a limitation for visceral tissue injury, we are primarily interested in surface tissue viability. An additional technology limitation of evaluating tissue moisture through reflectance intensity is that the SWAT device is sensitive to artifactual reflection from the skin surface, which is also dependent on the angle of the illumination and camera. The latter can be mitigated by considering ratios of the images at the different wavelengths, as for example in RI₃. However a potential artifactual issue is the effect of surgical cleansing materials on the wound bed. Thus, it is important to use a standardised wound preparation protocol as described in the Section 2 as excess external moisture could alter the tissue reflectance. A final limitation of this current study is the small sample size. By increasing the number of patients and biopsies in the study, we will be able to determine if these findings hold over a larger and more diverse patient population.

5. CONCLUSIONS

In summary, our study demonstrates that multi‐spectral SWIR imaging yields different reflectance intensities depending on the band wavelength and the depth of the burn. The results of this study provide evidence in support of using SWAT to provide an objective measurement of tissue viability and burn depth. Visual assessment of burn depth varies widely and the technology studied here can be further developed to potentially be utilised by surgeons to provide a more accurate assessment regardless of experience levels. Using this technology, surgeons will potentially be able to optimise the excision and debridement of damaged tissue and maximise the preservation of viable tissue. These preliminary results motivate further studies on SWIR imaging of burns in the hope of non‐invasively and accurately identifying operative versus non‐operative burns.

FUNDING INFORMATION

This study was supported in part by Department of Defence grant W81XWH‐18‐2‐0038 (Omer Berenfeld and Benjamin Levi), the Coulter Translational Research Partnership at the University of Michigan (Omer Berenfeld and Benjamin Levi), and National Institutes of Health grants R21‐HL153694, R21‐EB032661, and R01‐HL156961 (Omer Berenfeld).

CONFLICT OF INTEREST STATEMENT

Dr. Berenfeld is a co‐founder of Cor‐Dx LLC. The other authors declare no conflict of interest.

Nunez J, Mironov S, Wan B, et al. Novel multi‐spectral short‐wave infrared imaging for assessment of human burn wound depth. Wound Rep Reg. 2024;32(6):979‐991. doi: 10.1111/wrr.13221