1.

Introduction

Wound healing after burn injury involves the sequential phases of inflammation, tissue formation, and tissue remodeling.¹ Scar formation during this process is a natural facet of healing, and hypertrophic scars are one pathology arising from an excessive response to injury during healing.² Such scars are characterized by prolific neovascularization; and excessive content^3–5 and abnormal arrangement of collagen.⁶ In normal skin, collagen is arranged in a “basket-weave”-like structure, whereas collagen in hypertrophic scars is more unidirectionally aligned.^6,7 Previous studies have shown that, compared with normal skin, hypertrophic scars contain collagen fibrils with reduced diameter^6,8 and have higher water content.^9,10

Various techniques are available or under investigation for the treatment of pathological scars including surgery, pressure therapy, corticosteroid injections, silicone gel sheeting, radiotherapy, and laser therapy.¹¹ Evaluating the efficacy of these clinical treatments typically requires the longitudinal assessment of a scar’s severity. Current scar assessment methods are primarily based on the direct observation and palpation of the features of scars. For example, the Vancouver Scar Scale uses observer assessment of pigmentation, vascularity, pliability, and height to characterize the scar.¹² Assessment with such methods is inevitably somewhat subjective, leading to high inter-observer variability. There is reported research on alternative, less subjective techniques. Electron microscopy has been used to study the characteristic morphological profiles of pathological scars but requires histological slices of excised scar tissue.^6,13 Multiphoton^14,15 and confocal microscopies¹⁶ have been applied to scar tissue or wounds, ex vivo and in vivo, including for high-resolution imaging of collagen and elastic fiber components. Ultrasound imaging has been investigated as a means of assessing the thickness of burn scars in pediatric patients noninvasively.¹⁷ Other techniques, such as laser Doppler perfusion imaging and laser speckle perfusion imaging, have used the degree of vascularization as a surrogate indicator of scar status.¹⁸

Changes in tissue morphology in pathological scarring can result in changes in the optical properties of the tissue in the near-infrared wavelength range. This has driven a growing body of work exploring the use of optical coherence tomography (OCT) in scar assessment.^19–26 OCT is a noninvasive imaging modality capable of acquiring three-dimensional (3-D) scans of tissue at resolutions from 1 to 20 μm from superficial tissues at depths of up to 1 to 2 mm.¹⁹ The morphological features of normal and diseased skin have been studied in vivo with OCT imaging.^20,21 Wound healing has been evaluated noninvasively by high-resolution OCT,²² and polarization-sensitive (PS) OCT has been used to measure the birefringence of both normal and burn-damaged human skin.^23,24 OCT has also been used to noninvasively assess vascularity as a measure of normal skin and scar status.^25,26

For objective scar assessment, absolute quantities that characterize the tissue are required to monitor the progression of healing. One such quantity that may be extracted from OCT scans is the attenuation rate (parameterized by the attenuation coefficient), which is the rate at which the OCT signal decreases with depth in the tissue. This rate can be extracted by fitting the measured OCT signal versus depth (an A-scan) to a specific model.^27–29 Previous studies have shown that the attenuation coefficient measured with OCT can be used to distinguish between different types of tissue.^30–33 Previous research has also explored the attenuation coefficient of normal human skin tissue.^27,34,35 A very effective means of visualizing attenuation coefficients is through the presentation of en face maps, i.e., parametric images of the attenuation coefficient.^30,31 We hypothesize that differences in tissue microstructure between scarred and normal skin give rise to differences in the attenuation coefficient of near-infrared light. In this article, we extend the measurement of OCT attenuation coefficients and their presentation in parametric images to the case of pathological scarred tissue.

A difficulty arises when applying the attenuation coefficient method to in vivo OCT scans of skin. The vasculature of scarred and normal skin generates strong forward scattering at near-infrared wavelengths due, in part, to the high anisotropy and high scattering cross-section of red blood cells,³⁶ causing artificially high attenuation of the OCT signal that is not representative of the surrounding nonvascular component of the tissue. Here, we present an algorithm that automatically removes areas of vasculature from the calculations, allowing more accurate quantification of the attenuation coefficient of scarred and normal skin tissue. We present results from six clinically assessed scars that demonstrate a systematically lower attenuation coefficient of scarred tissue as compared with normal skin tissue and propose a mechanism explaining this observation.

2.

Materials and Methods

2.2.

Vasculature Masking

The flow chart for vasculature masking is shown in the left column of Fig. 1. We denote by $i^2$ the sum of the squared OCT signals from the two detection channels of the PS-OCT system. This is equivalent to the square of a conventional OCT signal and is sometimes also referred to as the OCT signal power. In the remainder of this article, the term OCT data refers to $i^2$. Blood vessels were automatically segmented from this OCT data using speckle decorrelation.³⁹ In brief, the correlation between OCT data within a small moving window was calculated between pairs of adjacent B-scans, $A$ and $B$, using the standard normalized cross-correlation equation:

Eq. (1)

$$\mathrm{corrMap}(x,z)=\frac{\sum _{p=1}^M\sum _{q=1}^N\{[i_A^2(x+p,z+q)-\overline{i_A^2(x,z)}][i_B^2(x+p,z+q)-\overline{i_B^2(x,z)}]\}}{\sqrt{\sum _{p=1}^M\sum _{q=1}^N{[i_A^2(x+p,z+q)-\overline{i_A^2(x,z)}]}^2}\sqrt{\sum _{p=1}^M\sum _{q=1}^N{[i_B^2(x+p,z+q)-\overline{i_B^2(x,z)}]}^2}},$$

where the window size is $M\times N$, and $i_A^2$ and $i_B^2$ are OCT data from a pair of neighboring subimages in adjacent B-scans in an OCT data volume that are defined by the window. $\overline{i_A^2}$ and $\overline{i_B^2}$ are, respectively, the mean values of $i_A^2$ and $i_B^2$.

Fig. 1

Data processing flow for the calculation of the attenuation coefficient of vasculature-masked tissue in scarred and normal skin (see text for details). MIP: maximum intensity projection.

Regions of low cross-correlation in the resulting correlation volume image were identified as blood vessels. Motion artifacts caused by patient movement during scanning were then removed from this correlation volume by the use of a registration algorithm utilizing the metal fiducial marker. Details of the motion correction are described elsewhere.³⁷ By identifying the voxel of the lowest correlation (i.e., the highest decorrelation) along each A-scan ($z$-direction) at depths in the skin with adequately high OCT signal-to-noise ratio, a two-dimensional (2-D) en face maximum intensity projection (MIP) image of the vasculature was produced, showing the $(x,y)$ locations of blood vessels. Further details of the algorithm were given previously.²⁶ This MIP image was then thresholded to form a binary mask of blood vessels, indicating $(x,y)$ locations to be removed from subsequent calculation of the attenuation coefficient.

3.

Results

3.1.

Vasculature-Induced Artifacts

Figure 2 shows the calculated attenuation coefficient and related data for normal skin on the lower right leg of a 19-year-old Caucasian male patient. The OCT scanning area ($4\times 1.5\mathrm{mm}$) was in the center of the region marked with a square outline ($10\times 10\mathrm{mm}$) on the skin, as shown in Fig. 2(a). Figure 2(b) shows the 2-D en face vasculature MIP of the scanning region. The blue square ($1\times 1\mathrm{mm}$) delineates the region over which the attenuation coefficient map in Fig. 2(c) was extracted using the corrected, registered, and averaged structural OCT data. The attenuation coefficient determined for all (vascular and nonvascular) tissue regions is shown. The corresponding goodness-of-fit ($0\to \text{poor fitting}$, $1\to \text{good fitting}$) is presented in Fig. 2(d). By comparing Figs. 2(b) and 2(c), it is apparent that the attenuation coefficient values in the circled vascular regions are much lower than in the surrounding tissues, even giving rise to nonphysical negative values. Figure 2(e) compares the normalized attenuation coefficient distribution between the vascular (i.e., blood vessels) and nonvascular (i.e., all tissue excepting blood vessels) regions and highlights this problem. The attenuation coefficient distribution in the vascular region is artificially broadened and extended to low values.

Fig. 2

Attenuation coefficient of normal skin showing the corrupting effect of the vasculature: (a) Photograph and (b) vasculature MIP of normal skin. (c, d) En face attenuation coefficient map and corresponding goodness-of-fit of the vascular and nonvascular tissues in the region outlined in blue in (b) (each image $1\times 1\mathrm{mm}$). Examples of vascular regions are circled by the dotted lines in (b), (c), and (d). (e) Comparison of normalized attenuation coefficient distribution between vascular and nonvascular tissue regions. (f, g) Fitting examples, respectively, from nonvascular and vascular regions marked by the small purple squares and indicated by arrows in (c).

Figures 2(f) and 2(g) show two representative fitting examples, respectively, for nonvascular and vascular tissues, marked by the two small purple squares in Fig. 2(c). The region of OCT signal used for the fitting (200 μm in length) is shown in blue, while the remainder of the A-scan is rendered partially transparent. The attenuation coefficient is extracted from the line of best fit (red) for each A-scan. Figure 2(f) demonstrates the applicability of the single-scattering model to the nonvascular dermal tissue with a high goodness-of-fit value of 0.91. In Fig. 2(g), the presence of two blood vessels at different depths gives rise to the sharp drops in the observed signal. The subsequent increase in the measured OCT signal below the first blood vessel gives rise to an apparent negative attenuation coefficient value with a poor goodness-of-fit value of 0.06. It is also evident from Fig. 2(d) that the low attenuation in the outlined vascular regions is accompanied by poor goodness-of-fit to the single-scattering, log-linear model. Since scarred skin has been shown to have larger blood vessel diameters and higher vasculature density than normal skin,²⁶ this vascular artifact is expected to be still more prominent in scar tissue. Consequently, we incorporated a vasculature-masking step in the determination of the attenuation coefficient in order to minimize this corrupting effect.

3.2.

Case Study of a Hypertrophic Scar

Figure 3 shows a 12-month-old hypertrophic scar originating from a hot water burn on the medial left inner forearm of a 30-year-old Caucasian female patient. Figures 3(a) and 3(b) are photographs of the contralateral normal skin and scar tissue, respectively. OCT scans were performed in the central $4\times 1.5\mathrm{mm}$ of the outlined regions. Figures 3(c) and 3(d) show the vasculature MIPs for the normal skin and scar, respectively, with the central blue squares ($1\times 1\mathrm{mm}$) identifying the regions for attenuation coefficient calculation. As noted by Liew et al.,²⁶ the vasculature is observed to be much more prolific in the scar tissue. Figures 3(e) and 3(g) depict the masks generated (black: vascular region; white: nonvascular region) after thresholding the vasculature MIPs. Figures 3(f) and 3(h) present the 2-D en face attenuation coefficient maps, respectively, for normal skin and scar. The attenuation coefficient is visibly lower in the hypertrophic scar than in the contralateral normal skin. From the distribution of attenuation coefficient values for each scan plotted in Fig. 3(i), we observe that the attenuation coefficients in scar tissue have a distribution of similar width to normal skin tissue but are on average much lower, with the mean attenuation coefficients of scar and normal skin being 3.8 and $6.4{\mathrm{mm}}^{-1}$, respectively.

Fig. 3

Case study of the attenuation coefficient of a hypertrophic scar: (a, b) Photographs of the contralateral normal skin and scar. (c, d) Vasculature MIPs of the normal skin and scar in (a) and (b), respectively. (e, g) Vasculature masks for normal skin and scar ($1\times 1\mathrm{mm}$), respectively. (f, h) En face attenuation coefficient (${\mathrm{mm}}^{-1}$) maps of, respectively, normal skin and scar ($1\times 1\mathrm{mm}$). (i) Histogram of attenuation coefficients for normal skin and scar.

3.3.

Case Study of a Normotrophic Scar

The results of a case study on a 12-month-old normotrophic scar are shown in Fig. 4. The scar, outlined in red in Fig. 4(a), was caused by a wood fire burn on the left lateral forearm of a 58-year-old Caucasian female. Figures 4(b) and 4(c) show MIPs of the vasculature of the contralateral normal skin (photograph not shown) and the scar, respectively. After masking of the vasculature using Figs. 4(d) and 4(f), the calculated attenuation coefficient maps are shown in Figs. 4(e) and 4(g). Figure 4(h) plots the normalized attenuation coefficient distribution of the scar and normal skin, which have mean values of 5.4 and $\text{6.7\hspace{0.17em}\hspace{0.17em}}{\mathrm{mm}}^{-1}$, respectively. Note the increased degree of overlap of the distributions between this (nonpathological) normotrophic scar (red) and the contralateral normal skin (green), in comparison to the (pathological) hypertrophic scar in Fig. 3(i).

Fig. 4

Case study of the attenuation coefficient of a normotrophic scar. (a) Photograph of the scar (which is outlined in dotted red). (b, c) Vasculature MIPs of the normal skin and scar, respectively. (d, f) Vasculature masks for normal skin and scar ($1\times 1\mathrm{mm}$), respectively. (e, g) En face attenuation coefficient (${\mathrm{mm}}^{-1}$) maps of, respectively, normal skin and scar ($1\times 1\mathrm{mm}$). (h) Histogram of attenuation coefficients for normal skin and scar.

3.4.

Results from All Subjects

The mean ($\pm \text{standard error}$) attenuation coefficient of the vasculature-masked dermal tissue in hypertrophic scars ($n=3$), normotrophic scars ($n=3$), and normal skin ($n=6$), calculated from data on all subjects, is summarized in Fig. 5. The hypertrophic scars give rise to the lowest attenuation coefficient ($3.8\pm 0.4{\mathrm{mm}}^{-1}$); the normotrophic scars values are greater ($4.2\pm 0.9{\mathrm{mm}}^{-1}$); and the contralateral or adjacent normal skin has the largest values ($6.3\pm 0.5{\mathrm{mm}}^{-1}$). Student’s t-test shows that the difference between the attenuation coefficients of the scar and normal skin is statistically significant ($p<0.001$).

Fig. 5

Mean ($\pm \text{standard error}$) attenuation coefficient of hypertrophic scars ($n=3$), normotrophic scars ($n=3$), and normal skin ($n=6$) of all subjects.

4.

Discussion

In this study, we specifically focused on characterization of the dermis, since it undergoes very significant changes in content and structure during the wound healing process² and occupies most of the OCT imaging depth range. As the epidermis is only about 100-μm thick and is largely restored with new epidermal tissue during the wound healing process, we chose to exclude it. Our results show on average 36% lower dermal attenuation in scarred skin compared with normal skin. This strong effect can potentially be used as the basis for an objective method of scar characterization. As pathological scarring commonly leads to excessive production of both collagen fibers and blood vessels, the present method complements the scar assessment method solely based on vasculature that we have recently reported.²⁶ The integration of both methods could potentially be used to track the efficacy of various treatment strategies targeting both collagen fibers and blood vessels.

To explain the contrast that we observe in the vasculature-masked tissue, we have employed a model proposed by Jacques which has been shown to accurately describe the attenuation of light in the dermis.⁴³ This model was verified against measurements of the reduced scattering coefficient, $(1-g){\mu }_s$, where ${\mu }_s$ is the scattering coefficient and $g$ is the anisotropy factor, of ex vivo samples of dermis.⁴⁴ Jacques’ model treats scattering in the dermis as being equivalent to single scattering by a collection of infinite cylinders, each of which represents a collagen fiber. Jacques also considered the contribution of Rayleigh scattering by subwavelength scatterers, but it has been shown that Rayleigh scattering is insignificant compared with scattering by the collagen fibers at the wavelength of interest.⁴⁴ The key parameters in the model are, thus, the refractive index of the collagen fibers and the background material, the distribution of fiber diameters, and the volume fraction of collagen fibers. The refractive index of the collagen fibers was calculated by assuming that the fibers are composed of a combination of water and collagen. Mie theory, describing the scattering of light by infinite cylinders, was used to calculate ${\mu }_s$ and $g$.

In relating Jacques’ model to the mathematical form of the OCT signal in Eq. (2), we note that ${\mu }_t$ is proportional to ${\mu }_{s}a(g)$, where $a(g)$ represents the directional nature of scattering.⁴⁵ We, however, found that $g$ varies insignificantly compared with ${\mu }_s$ when contrasting normal skin and scarred tissue, enabling $a(g)$ to be neglected when assessing the relative magnitudes of ${\mu }_{\mathrm{t}}$ for the normal and scar cases. We did not have access to data on the distribution of fiber diameters for the particular cases presented in this article. We, thus, approximate the two ensembles of fibers, i.e., scar and normal, by their mean diameter and spacing from the literature.^7,46 In particular, the mean spacing of collagen fibers has been found to be 10.37 μm and 8.53 μm for normal and scar tissues, respectively, while the mean diameter was found to be 5.39 and 5.77 μm, respectively. Thus, if all other parameters remained equal, the higher density of fibers would result in ${\mu }_s$ for scar tissue being larger than that for normal tissue. It has, however, been noted that the percentage of water content is greater in hypertrophic scar tissue than in normal tissue (approximately 64% in normal skin).^9,10 The water content of hypertrophic scar tissue has been found to vary with scar age and, in one study, varied monotonically between 85.1% for a 0.5-year-old scar to 63.8% for a 2-year-old scar.⁹ Another study noted that the water content of hypertrophic scar was 82%, although the age of the scar was not noted.¹⁰ We cannot suppose that this water content is identical to that likely to be found in the collagen fibers. We note, however, that Jacques⁴³ assumed a water content of 65% in his model, which is in agreement with the measured figure for normal skin tissue.

Our approach was, thus, to calculate the ratio ${\mu }_s^{\text{scar}}/{\mu }_s^{\text{normal}}$, where ${\mu }_s^{\text{scar}}$ and ${\mu }_s^{\text{normal}}$ are the scattering coefficients for scar and normal skin tissues, respectively, for different percentages of water content in scar collagen fibers, while the water content percentage of collagen fibers in normal skin was fixed at 64% consistent with the literature. We used refractive indices for water and collagen of 1.3241 and 1.4367 at a wavelength of 1325 nm, respectively.⁴⁷ The scattering coefficients were calculated using Mie theory for scattering by cylinders, at normal incidence, using the mathematical formulation presented by Bohren and Huffman.⁴⁰ Figure 6 shows the simulated ratio of scattering coefficient between scar and normal skin, ${\mu }_s^{\text{scar}}/{\mu }_s^{\text{normal}}$, as a function of water content in the scar collagen fibers. The green triangle and red circle highlight simulated scattering coefficient ratios between scar and normal skin of particular interest: respectively, for the water content percentage of 64% (2-year-old hypertrophic scars) and 85% (0.5-year-old hypertrophic scars).⁹ The purple square corresponds to the ratio derived from our calculated attenuation coefficients of the scars and normal skin, and the resulting inferred water content ($\sim 76\%$) lies between that of the 0.5- and 2-year-old hypertrophic scars. This level is consistent with the 1.2-year-old mean age of the scars in this study but not with the lower reported water content of 1-year-old hypertrophic scars (66.5%).⁹ Although requiring further study, our model and previous reports suggest that the contrast in attenuation coefficient may come from higher water content in scars. Thus, results of the type plotted in Fig. 6 could be used to study the water content in scar tissue noninvasively from measurement of the OCT attenuation coefficient ratio.

Fig. 6

Simulated ${\mu }_s^{\text{scar}}/{\mu }_s^{\text{normal}}$ ratios versus water content in scar collagen fibers. The triangle and circle represent water content values reported in the literature,⁹ and the square represents the value calculated from the data in this study.

The masking of blood vessels removed a significant artifact from the calculation of the attenuation coefficient caused by the strong scattering by blood vessels. To the best of our knowledge, this is the first reported incorporation of vasculature masking into the calculation of attenuation coefficients from OCT data. Blood comprises plasma and other constituents, among which erythrocytes dominate the optical properties. Erythrocytes have a biconcave, disk-like shape with a size considerably larger than our OCT wavelength, which leads to strong forward scattering. This, together with the high-scattering cross-section and strong absorption of light,³⁶ generates a strong drop in the OCT signal, causing blood to exhibit a much higher attenuation coefficient than the nonvascular tissue in the skin. A previous study on fresh porcine blood, using 1300 nm OCT light, reported an attenuation coefficient of $12.15{\mathrm{mm}}^{-1}$,⁴⁸ which is almost twice the mean value of the vasculature-masked skin tissue measured in this study. This is particularly important in the evaluation of hypertrophic scars, which are known to have enhanced vasculature.²⁶ Analysis of the attenuation in these masked regions of blood flow may potentially provide additional information on metabolism such as the noninvasive monitoring of blood glucose concentration.⁴⁹

Skin comprises a collection of microstructures with varying scattering and absorption properties. Quantification of scar and normal skin attenuation coefficients is complicated by tissue heterogeneity. Even after masking out the vasculature, heterogeneity in depth can invalidate the single-scattering exponential decay model and results in modified attenuation coefficient values, even negative values when the fitting window initiates in a low-signal region and extends into a high-signal region. We empirically found a fitting range of 200 μm in the dermis to be an appropriate compromise between including sufficient data in the estimate of the attenuation coefficient and minimizing the effects of tissue heterogeneity. The average attenuation coefficient of normal skin we determined using this fitting length is $6.3\pm 0.5{\mathrm{mm}}^{-1}$, which lies between the results reported by Schmitt et al.²⁷ (4.6 to $4.7{\mathrm{mm}}^{-1}$) and Kholodnykh et al.³⁴ (10 to $13{\mathrm{mm}}^{-1}$) for forearm at 1300-nm wavelength. As discussed by Kholodnykh et al.,³⁴ the variation in attenuation coefficient could be caused by factors such as differences in experimental protocol, the use of a clearing agent, and variations in the skin tissue among individuals. One different feature of our experimental protocol from that of others is that we removed the contribution of blood vessels to target the nonvascular tissue.

In vivo OCT skin scanning can be affected by the motion between the scanner head and the skin. Such motion introduces image artifacts that distort the structural OCT images.³⁷ We used a fiducial marker fixed to the skin to guide a feature-based registration algorithm to remove such artifacts. However, motion artifacts also artificially increase speckle decorrelation, resulting in errors in blood vessel segmentation. An example of such an artifact appears as a pattern of horizontal lines in the vasculature MIP of Fig. 2(b). Alternative solutions have been proposed, such as rigidly affixing the scan head to the patient’s skin,^50,51 although this can present additional difficulties when scanning acute burn wounds. Another feasible method is to reduce the 3-D data acquisition time, although this may reduce the sensitivity of vessel detection in areas of slow blood flow due to reduced speckle decorrelation between rapidly acquired B-scans.

5.

Conclusions

In this article, we have presented a method for in vivo assessment of near-infrared attenuation coefficients of the dermis using OCT. We corrected for artifact caused by the presence of blood vessels using an automated method to segment and mask vasculature. We demonstrated the method of assessment on several in vivo human burn scars. Our results indicate that scars of mean age 1.2 years have on average 36% lower attenuation coefficients than contralateral or adjacent normal skin, and our modeling suggests that the lower attenuation coefficients arise from the previously reported increased water content in scars. This method could potentially be integrated with an automated vascularity quantification method in an objective scar assessment protocol to characterize both the vasculature and the connective tissue in scars. Future work will involve the investigation of this objective assessment protocol for the longitudinal study of human burn scars.