Coagulation depth estimation using a line scanner for depth-resolved laser speckle contrast imaging

Abstract

Partial-thickness burn wounds extend partially through the dermis, leaving many pain receptors intact and making the injuries very painful. Due to the painfulness, quick assessment of the burn depth is important to not delay surgery of the wound if needed. Laser speckle imaging (LSI) of skin blood flow can be helpful in finding severe coagulation zones with impaired blood flow. However, LSI measurements are typically too superficial to properly reach the full depth of the adult dermis and cannot resolve the flow in depth. Diffuse correlation spectroscopy (DCS) uses varying source-detector separations to allow differentiation of flow depths but requires time-consuming 2D scanning to form an image of the burn area. We here present a prototype for a hybrid DCS and LSI technique called speckle contrast diffuse correlation spectroscopy (scDCS) with the novel approach of using a laser line as a source and using the speckle contrast of averaged images to obtain an estimate of static scattering in the tissue. This will allow for fast non-contact 1D scanning to perform 3D tomographic imaging, making quantitative estimates of the depth and area of the coagulation zone from burn wounds. Simulations and experimental results from a volumetric flow phantom and a gelatin wedge phantom show promise to determine coagulation depth. The aim is to develop a method that, in the future, could provide more quantitative estimates of coagulation depth in partial thickness burn wounds to better estimate when surgery is needed.

1. Introduction

Partial thickness burn wounds are a particularly painful form of burn wound where the dermis is partially damaged, but pain receptors still are intact and able to sense the tissue damage. Proper diagnosis of burn wounds is important for optimal treatment as they may or may not heal on their own depending on burn depth and area in the dermis of the skin. Superficial burn wounds that do not extend into the dermal layer as well as deep burn wounds that extend fully through the dermal layer are easy to diagnose by eye, but partial thickness burn wounds are more difficult to assess. As partial thickness burn wounds are very painful, it is of great interest to determine whether skin resection and transplantation is needed or if they will heal on their own as soon as possible. Laser Speckle Contrast Imaging (LSI) is a non-contact method to measure relative blood perfusion, which can be used to detect coagulated areas with reduced blood perfusion. LSI has shown promise for diagnosis of scald wounds in children to determine whether surgical resection and skin transplantation is needed or if the wound will heal on its own [1]. However, LSI has limited measurement depth, typically within 700 µm [2], which is a limiting factor for burn wounds in adults with thicker skin, where the dermis usually is about 1-2 mm thick [3]. Further, it cannot resolve the burn wound in depth. Diffuse Correlation Spectroscopy (DCS) [4] allows for deeper, and potentially depth-resolved, measurements by using larger source-detector separations. However, it only gives point measurements and requires time-consuming scanning to form images of the blood perfusion. In recent years, a hybrid approach of LSI and DCS named Speckle Contrast Optical Spectroscopy (SCOS) has been developed [5]. In SCOS, the spatial speckles in an image are used to estimate the perfusion as in LSI but with a point source as in DCS. This allows spatially resolved information over a larger area at once but still requires scanning in 2D to form a complete image of the blood perfusion. The combined imaging and depth resolution can be used for tomographic imaging, speckle contrast Diffuse Correlation Tomography (scDCT) [6]. We are currently developing a novel system using a laser line as the source instead of point sources, which will allow for depth-resolved imaging with a single 1D scan [7,8].

In this paper, we present a novel method to utilize the static scattering component to estimate the coagulation depth with this system. Laser speckle techniques are based on the phenomenon that coherent laser light generate speckle patterns that move faster with increased motion in the structures that scatter light, e.g. moving red blood cells. In LSI, single or multi-exposure [9,10] short acquisition times are used to estimate the relative perfusion in the tissue where faster motion of the speckle pattern translates into more blurred images. However, if a long exposure time is used instead, only the fraction of light that has been scattered in static tissue will remain in the speckle contrast and has thus been proposed to provide valuable information about the amount of coagulated tissue where the blood flow has ceased completely [11]. A long exposure time can be generated artificially by averaging laser speckle images before calculating the speckle contrast, instead of the usual LSI method of calculating the speckle contrast from the individual images before averaging.

2. Material and methods

2.1. System set-up

A red 633 nm collimated laser with a beam width of 0.8 mm (Melles Griot 05-LHR-151) was directed through a Powell lens with 30° fan angle (Thorlabs Inc., USA) to generate a laser line with a length of 4 cm in the image plane. The line source is then used to illuminate two distinct tissue-like phantoms with geometries emulating burn wounds at an angle of 55°. Wide-field images were collected with a 12-bit CCD camera, with the line source at the center of the field of view, at 30 frames per second (Blackfly BFLY-U3-13S2M9, Teledyne FLIR LLC, USA) (Fig. 1(a)). Images were taken at a distance of 16 cm over an area of 72 × 54 mm giving an image pixel size of 56 µm/pixel. A relatively small f-number = 4 was used as multiple speckles per pixel have been found to improve signal-to-noise ratio (SNR) [12–15], with a resulting estimated speckle diameter of 3.3 µm, slightly smaller than the physical detector pixel edge length of 3.75 µm. Polarizing filters with orthogonal orientation were placed in front of the laser and the camera to reduce the risk of surface glares that can create artefacts in the LSI.

Fig. 1.

System set-up and flow phantom. (a) Laser light is shined on a line and speckle contrast is measured at different distances from it, corresponding to increasing measurement depth with distance. The system was tested on (b) an optical phantom with flow tubes at different depths from the surface and (c) a liquid phantom with a gelatin wedge on top to simulate coagulation from a burn wound with varying thickness. The red line marks the laser line orientation.

2.2. Burn wound simulating optical phantoms

Two tissue simulating phantom variants were made to test the system in the context of burn wounds: 1) a solid phantom with tubes for controlled flow at discrete depths and 2) a homogeneous liquid phantom with a static gelatin wedge on top to emulate coagulated tissue. The first phantom variant was made of the silicone Polydimethylsiloxane (PDMS) with titanium oxide, TiO₂, added for scattering and red food dye (E 120, Dr. Oetker, Germany) for a simple approximation of hemoglobin absorption. The optical properties of the phantom was measured with a Spatial Frequency Domain Spectroscopy (SFDS) system [16,17] to be an absorption coefficient of µ_a = 0.002 mm^-1 and a reduced scattering coefficient of µ_s^′ = 1.15 mm^-1 at 633 nm. The tube phantom contains a transparent tube with an outer diameter of 1 mm and an inner diameter of 0.5 mm bent to pass in straight lines at four different depths, starting from the tube edge in line with the phantom surface and the deeper in steps of 0.5 mm (Fig. 1(b)) to emulate discrete thicknesses of static (coagulated) tissue above intact, flowing vasculature. The analogue with a burn wound here is that a deeper wound will have a thicker static layer above the flow area, corresponding to a deeper tube in the phantom. Intralipid (Fresenius Kabi, Sweden) diluted with water to a lipid concentration of 1% was used as a scattering liquid in the tubes as it has similar reduced scattering, µ_s^′ = 1.5 mm^-1 at 633 nm (measured with SFDS), as blood, µ_s^′ = 2.1 mm^-1 [18], at 670 nm

In the second phantom variant, a small container with a thin plastic sheet top was also used to contain the same scattering liquid as was used at discrete, directed flow in the first phantom. On top of the plastic layer, a wedge of 0.1 g/ml gelatin and Intralipid at a lipid concentration of 3% was placed to simulate a burn wound with a continuously varying coagulation depth from 0 to 4.7 mm (Fig. 1(c)), which exceeds relevant depths for partial thickness burn wounds. Mixtures with 30 and 50% glycerol in the liquid layer were also used to reduce the Brownian motion in the liquid by the higher viscosity of the glycerol as emulation for different tissue blood perfusions. Since the difference in index of refraction between lipids and glycerol is smaller than between lipids and water [19], the Intralipid concentration was increased to 3 and 7% to maintain similar reduced scattering as in the liquid without glycerol. For the gel phantom, the higher concentration was also due to the reduction in scattering that occurs during the heating process [20]. Measured scattering values at 633 nm for a larger reference slab of the gelatin phantom was µ_s^′ = 0.82 mm⁻¹. For the 3% intralipid / 30% glycerol liquid µ_s^′ = 1.61 mm^-1, and for the 7% intralipid / 50% glycerol liquid µ_s^′ = 1.32 mm^-1.

2.3. Phantom measurements – flow phantom with discrete steps

The intralipid was pushed through the tube using a syringe pump (Orion Sage model M362) at a flow speed of 1 ml/minute. 25 images were taken with an exposure time of T = 1 ms. The speckle contrast, K (-) with adjustment for dark noise, σ_dark, and shot noise, σ_shot, was calculated as [5]

$$K^2=\beta \left(\frac{{\sigma }_{\text{measured}}^2-{\sigma }_{\text{shot}}^2-{\sigma }_{\text{dark}}^2}{⟨I^2⟩}\right)=\beta \left(\frac{{\sigma }_{\text{measured}}^2-⟨I⟩/\gamma -{\sigma }_{\text{dark}}^2}{⟨I^2⟩}\right)$$ (1)

Where β (-) is the spatial correlation coefficient of the system, I (photon counts) the light intensity, σ (photon counts) the standard deviation of the light intensity, and γ (-) the full well capacity of the CCD camera divided by the number of digital bits in the image. In practice, the dark and shot noise factors were estimated simultaneously by removing the lenses from the camera and measure unfocussed non-coherent white light at intensities spanning the intensity range of the camera. The average variance along a sliding window of 25 pixels was then fitted with linear regression to the average light intensity $⟨I⟩$ with the constant term corresponding to the dark noise, ${\sigma }_{\text{dark}}^2$ , and the linear component corresponding to the shot noise, ${\sigma }_{\text{shot}}^2$ . As the laser line gives an image with rapidly decreasing light intensity from the line, the speckle contrast was calculated along lines with 25 pixels parallel with the laser line so that the intensity gradient would not increase the speckle contrast. The speckle contrast was finally averaged over the 25 images and spatially smoothed with a square kernel of 25 × 25 pixels. Finally, a simple relative perfusion estimate was calculated as 1/K² - 1 and a static scattering fraction, K_S (-), was estimated by the speckle contrast for 25 averaged images,

$$K_{\text{s}}=K\left(⟨I⟩\right)$$ (2)

2.4. Phantom measurements – liquid phantoms with gelatin wedges

For the wedge phantom, the varying height causes the laser line to be displaced due to the angle of the incoming light. To compensate for this, a moving average of the intensity over 50 pixels (2.8 mm) with a step size of 1 pixel was done in the orthogonal direction of the laser line and the position of the peak intensity in the averaged signal was tracked to shift the columns of the image in the postprocessing. Measurements with the laser line at the center of the wedge along the direction of the slope were done for the liquid phantoms with 0, 30, and 50% glycerol. Finally, scanning was done over a 1.8 mm wedge phantom in steps of 2 mm to generate an image.

2.5. Estimation of coagulation depth from static scattering

Based on the results for the wedge phantom (Fig. 4), it was hypothesized that the distance of isolevels of the static scattering should be useful to estimate the coagulation depth. Isolevel distances, L (mm), of K_S = 15, 20, and 30% were calculated and their distances to the center of the laser line were used to make a piecewise regression model to fit the thickness, h (mm), of the gelatin wedge in the measurement on the 4.7 mm wedge on top of the liquid phantom without glycerol. The regression fit was limited to the height 3.7 mm to avoid edge effects and averaging with the pure liquid beyond the wedge edge from kernel size and smoothing. This range of phantom thicknesses exceeds all possible thicknesses of the dermis and hence all clinically relevant burn depths, from superficial to full thickness burns. The isolevel values were set to a distance far away from the laser line in case no values for the isolevel existed. From the data, it was determined that the regression should be between L² and h. Fitting was also done between the static scattering, K_S, at the center line of the laser, corresponding to what would be obtained from traditional LSI, and the wedge height with a logarithmic function $\hat{h}=-a\ln (b-S)+c$ . Here, a, b, and c are the fitted parameters.

Fig. 4.

(a) Estimated perfusion from a gelatin wedge phantom over a liquid phantom with 1% fat intralipid as scattering and moving media thresholded at a perfusion level of 5. The perfusion isolevels can be seen to be further away from the laser line at y = 0 mm with increasing wedge thickness. (b) Comparisons of perfusion estimates at different thicknesses of the wedge compared with reduced Brownian motion from added glycerol to the liquid intralipid phantom. The perfusion estimates from 30% and 50% glycerol are similar to the estimates for 1 and 2 mm layer thickness respectively but the slope increases more slowly with distance from the laser line in the glycerol cases without gelatin layer on top compared to the varying gelatin thickness. (c-d) Estimated static scattering, K_S, in the same measurement. The traditional perfusion estimate is not only affected by the thickness of the coagulation layer but also greatly affected by the underlying perfusion, here modelled with varying viscosity form varying glycerol concentrations. The static scattering estimate, however, is mostly affected by the thickness of the coagulation layer. Note that in the ideal case, the perfusion estimates should increase with distance from the laser line while the static scattering estimate should decrease. However, opposite trends appear as artefacts once the distance is so large that the signal-to-noise becomes poor.

2.6. Light transport simulations

In addition to the experimental measurements of the two coagulation depth phantoms, simulations of perfusion estimates for the corresponding phantom geometries were also done. Finite Element Method (FEM) simulations of the electric field autocorrelation, G₁, were set up in COMSOL Multiphysics 5.6 (COMSOL, Sweden) according to the diffusion approximation [21]

$$\mathrm{\nabla }\cdot \left(-\frac{1}{3{\mu }_s^{\mathrm{\prime }}}\mathrm{\nabla }{G}_1\right)=S-{\mu }_a{G}_1-2{\left(\frac{2\pi n}{{\lambda }_0}\right)}^2{\mu }_s^{\mathrm{\prime }}{D}_{\text{B}}\tau G_1$$ (3)

where S is a source term in the form of the scattered light from the exponentially decaying incoming light from the laser line, D_B (mm²/s) the Brownian diffusion coefficient, assumed to be 10⁻⁵ mm²/s in the simulation, n (-) the index of refraction, and λ₀ (mm) the vacuum wavelength of the light.

$$S={\mu }_s^{\mathrm{\prime }}{I}_0{e}^{-({\mu }_s^{\mathrm{\prime }}+{\mu }_a)z}$$ (4)

Here, I₀ is the intensity of the laser line, which was removed by normalization, and z (mm) the depth from the surface in the phantom. The boundary conditions of the phantom was set to

$$\hat{n}\cdot \frac{1}{3{\mu }_s^{\mathrm{\prime }}}\mathrm{\nabla }{G}_1=-\frac{1}{2}\left[\frac{1-R_{\text{eff}}}{1+R_{\text{eff}}}\right]G_1$$ (5)

where $\hat{n}$ is the normal vector of the surface and R_eff (-) the effective reflection coefficient, set to a typical value of R_eff = 0.47 for air over tissue [22].

G₁ was normalized to the normalized electric field autocorrelation g₁(τ) = G₁(τ)/G(0). Finally, g₁ was converted to the corresponding speckle contrast K, according to [23]

$$K^2(T)=\frac{2\beta }{T}{\int }_{\tau =0}^T|g_1(\tau ){|}^2\left(1-\frac{\tau }{T}\right)\text{d}\tau$$ (6)

where the spatial correlation factor, β, was assumed to have the ideal value forpolarized light, β = 1. This does not take the real speckles per pixel on the detector into account, but the actual value is removed by the static area normalization. The integral was done by exporting the results from simulation to MATLAB and summing them in steps of 0.1 ms up to an exposure time of T = 1 ms.

3. Results

Measured and simulated perfusion estimates from the speckle contrast for the flow phantom are presented in Fig. 2. Perfusion from the tubes could be seen up to a distance of approximately 10 mm from the central laser line at y = 0 mm. Flow from the most superficial tube is visible directly beneath the laser line while flow from deeper tubes become visible at increasing distances from the line, corresponding to deeper measurement depths. There is not a distinct appearance but e.g. the isoline of the measured perfusion estimate = 1 is seen at -1.0, -1.8, -3.6, and -4.4 mm for the tubes at depths of 0.5, 1.0, 1.5, and 2.0 mm respectively (Fig. 2(a)). Corresponding distances for the simulated perfusion estimates are -1.0, -1.7, -3.3, and 3.8 mm (Fig. 2(b))

Fig. 2.

(a) Measurement and (b) simulation of the perfusion estimate. The rightmost tube is the most superficial and can be seen in level with the laser line source at y = 0 mm while the deeper tubes appear further away from the laser line. The speckle contrast has been normalized to be 1 in the center of the images (around x = 12.5, y = 0 mm) where only static scattering is expected. Dashed blue lines mark approximate tube locations (c) Measurement and (d) simulation of the static scattering estimate K_S. Dashed red lines mark the laser line illumination.

Results for the liquid phantom with the wedge on top are presented in Fig. 3 to 7. The shift of the laser line due to the height of the wedge is presented in Fig. 3 As the light comes in at an angle, it is displaced in the y direction and this needs to be compensated for in the image to get the correct distances from the laser line in the image. Perfusion and static scattering estimates are presented in Fig. 4. The viscosity of the liquid phantom increases with the increased glycerol concentration, resulting in reduced perfusion estimates. The static scattering estimates are mainly affected by the thickness of the gelatin wedge on top, however (Fig. 4(b)). Fitting of static scattering to wedge height in the phantom with 0% glycerol is presented in Fig. 5. Using the fits to estimate the wedge thickness for different glycerol concentrations is presented in Fig. 6. It can be seen that the height can be recovered even with greatly reduced perfusion from the added glycerol. The scanning over the wedge is presented in Fig. 7, showing the kind of images that would be generated for a burn would case in a patient in the future, and corresponding thickness estimates are shown in Fig. 8.

Fig. 3.

(a) The angle of the laser light is displaced by varying phantom height. (b) Average intensity over 50 pixels in the x direction over the pure liquid phantom part and over the gelatin wedge at 4 mm thickness. (c) Diffuse reflectance light intensity from the laser line at y = 0 mm. (d) Displacement of the peak position of the laser line due to the height of the gelatin wedge. This displacement was the used to deskew the image so that the laser line is centered on y = 0 in Fig. 4.

Fig. 7.

Scanning in steps of 2 mm over a 1.8 mm solid wedge phantom over a liquid phantom of 1% intralipid with perfusion estimates at 0–4 mm from the laser line normalized with the perfusion estimates from the liquid phantom only (red area).

Fig. 5.

Isolevel distances, L, from the laser line. (b) Piecewise linear regression between the coagulation depth modelled with the wedge thickness, h, and static isolevels squared, L². When estimating h, The 30% isolevel is used as long as ${\text{L}}_{30\mathrm{\%}}^2$ ≥ 3 mm², then 20% isolevel as long as ${\text{L}}_{20\mathrm{\%}}^2$ ≥ 0.8 mm², and finally the 15% when ${\text{L}}_{20\mathrm{\%}}^2$ < 0.8 mm². The curves end when the wedge is so thick that the static scattering estimate never becomes as low as the corresponding isolevel at any distance. (c) Least squares fit between h and the static scattering S at the center of the laser line using a logarithmic function $\hat{\mathrm{h}}=-\text{a}\ln (\text{b}-{\text{K}}_{\text{s}})+\text{c}=-1.04\ln (0.87-{\text{K}}_{\text{s}})+0.96$ .

Fig. 6.

(a) Estimation of coagulation thickness through piecewise linear regression against static scattering isolevels, L, utilizing the increasing depth penetration of the light with increasing distance from the laser line. (b) Estimation of coagulation thickness from least-squares fitting to a logarithmic function of the static scattering, S, at the center of the laser line, corresponding to a traditional LSI measurement without depth resolution. The fitting works well for a wedge thickness up to about 1.5 mm (at x around 25 mm), but when the wedge gets thicker, the fitting becomes much less accurate than when using the longer distances from the laser line where the signal comes from light that has travelled deeper.

Fig. 8.

(a) Thickness estimate from static scattering isolevels L (b) Error in Thickness estimate, limited to the range 0 to 1 mm.. (c) Thickness estimate from static scattering S at the center of the laser line., (d) Error in Thickness estimate from center line, limited to the range 0 to 1 mm..

4. Discussion

In this paper, a proof of concept for fast burn wound coagulation depth estimation with a line laser source is presented. The system shows promise in estimation of coagulation depth through the use of estimates of the static scattering, analyzed by calculating distances of static scattering fraction isolevels from the central laser line. Another benefit of the system compared to previous scDCT set-ups is that it will allow the forming of depth-resolved images with a simple 1D scan and should thus speed up the imaging considerably. The ability to form images quickly is very important for the system to be of practical use in a clinical setting. While traditional perfusion estimates are not used in the estimates of coagulation depth here, they are still simultaneously obtained. The perfusion estimate will provide information about blood flow beneath the coagulation layer that may be valuable for assessing the individual vascular response over time and the ability of the tissue to heal. One particular problematic topic for burn wound assessment is that the wounds progress with a growing necrotic coagulation zone over time in a manner that is difficult to predict. Important processes for the wound progression are inflammation and ischemia [24] and it is possible that the perfusion measurements can provide valuable information about those processes.

We are currently not using the static scattering estimate in the perfusion estimate itself, but this is something that could be done in the future, especially if using a multi-exposure scheme [9] which would allow to implement static scattering correction [25].

While the coagulation thickness estimate seems robust in regard to varying underlying perfusion (Fig. 6), it is most likely that the thickness estimates depend on the optical properties of the tissue and the fitting model thus must be extended to take varying properties into account. Individual optical tissue properties can be obtained with spatial frequency domain spectroscopy [26] or imaging (SFDI) [27], which also are non-contact measurement techniques and as such suitable for the sensitive burn wounds. Chromophore content and scattering properties obtained with SFDI could also be valuable by themselves for assessing the burn wounds as they can reflect damage to collagen and hemoglobin in the tissue [28]. Spatial frequency illumination is also a possible alternative to the laser line that has been shown to provide depth resolution [29,30].

Ideally, the calculated perfusion from the speckle contrast should increase indefinitely while the static scattering estimate should decrease with distance from the laser line as the light travels further and experience more Doppler shifts before reaching the detector. However, the inherent limits to signal-to-noise ratio makes the perfusion estimate “peak and the static scattering estimate starts to increase at some distance. Interestingly, this peak is further away for the lower Brownian motion of the increased glycerol concentration or the thicker gelatin layer (Fig. 4(b)). The noise compensation [5] (Eq. (1)) improves the distance the perfusion can be measured at but cannot improve the signal infinitively. The camera used is a relatively cheap one with 12 bits and better results can be expected with a higher end model with more bits. This could especially be the case if the system is to be used in another setting where deeper measurements are desired, and the quantization noise may be too high with only 12 bits. SNR limitations can, for example, be seen in Fig. 4(c), where the static scattering estimate ideally should be lower further away from the laser line but at distances beyond 5 mm it starts to increase due to low SNR. It can be interesting to note that the increase begins closer the higher the flow is as can be seen over the pure liquid area to the right side of the wedge.

Estimating the coagulation thickness from the center line gets unstable for thicker coagulation depths than using the isolevels at distances from the laser line but for thin coagulation depths it still gives valuable information. This is thus a measure that could be incorporated with ease in already existing LSI systems with full-field illumination.

The displacement of the laser line due to the illumination angle (Fig. 3) also allows estimating a height profile of the tissue [22]. This could, for example, be of interest when tracking the healing process of resected burn wounds and also other types of wounds. Today there are dedicated topographic 3D cameras developed for the purpose of estimating the depth of wounds [31]. Our proposed system could be designed to obtain similar information at the same time as obtaining information about coagulation depth and blood perfusion that the 3D cameras can’t provide, which could make the clinic work faster and cheaper.

Using a red laser is convenient during the prototype development as it is well visible. Eventually, it will be switched to the commonly used near-infrared wavelength of 780 nm which is only weakly visible to the human eye but allow better light penetration due to generally lower absorption and scattering in the tissue.

5. Conclusions

In conclusion, the proof-of-concept system shows promise to determine coagulation depth and should be improved with better hardware and models to take varying optical properties of the tissue into account. The usage of the static scattering estimate also shows promise for compact, low-cost implementations with wide field illumination, although with more limited measurement depth.

Funding

Linköping University10.13039/501100003945 Area of Strength CircM; Knut and Alice Wallenberg10.13039/501100004063 Centre for Molecular Medicine.

Disclosures

The authors declare no conflicts of interest.

Data Availability

Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.