Visualization enhancement by PCA-based image fusion for skin burns assessment in polarization-sensitive OCT

Polarization-sensitive optical coherence tomography (PS-OCT) is a functional imaging tool for measuring tissue birefringence characteristics. It has been proposed as a potentially non-invasive technique for evaluating skin burns. However, the PS-OCT modality usually suffers from high system complexity and relatively low tissue-specific contrast, which makes assessing the extent of burns in skin tissue difficult. In this study, we employ an all-fiber-based PS-OCT system with single-state input, which is simple and efficient for skin burn assessment. Multiple parameters, such as phase retardation (PR), degree of polarization uniformity (DOPU), and optical axis orientation, are obtained to extract birefringent features, which are sensitive to subtle changes in structural arrangement and tissue composition. Experiments on ex vivo porcine skins burned at different temperatures were conducted for skin burn investigation. The burned depths estimated by PR and DOPU increase linearly with the burn temperature to a certain extent, which is helpful in classifying skin burn degrees. We also propose an algorithm of image fusion based on principal component analysis (PCA) to enhance tissue contrast for the multi-parameter data of PS-OCT imaging. The results show that the enhanced images generated by the PCA-based image fusion method have higher tissue contrast, compared to the en-face polarization images by traditional mean value projection. The proposed approaches in this study make it possible to assess skin burn severity and distinguish between burned and normal tissues.


Introduction
The skin, which is one of the largest organs in the human body, comprises several layers of ectodermal tissues, playing vital roles such as safeguarding internal organs, defending against harmful microorganisms, regulating body temperature, providing thermal insulation, enabling the perception of touch, and producing vitamins [1].Daily behaviors, such as alcohol consumption, staying up late, sun exposure, and an imbalanced diet, can harm the skin.Unlike manageable and reversible skin injuries, acute skin injuries resulting from accidents are often irreversible and can pose life-threatening risks.The most commonly encountered form of acute skin injury is thermal burns, usually caused by contact with flames, hot gases, scalding liquids (such as water or oil), and heated metal [2,3].Burn injury results in coagulative necrosis and protein denaturation of different layers of skin and the underlying tissues.The severity of skin burns is related to the temperature, the energy transmitted by the heat source and the duration of exposure.Burn injury can trigger pathological symptoms such as wound edema and inflammation in less severe cases.In more severe cases, burns may lead to complete disruption of the epidermis and dermis, resulting in open wounds, further increasing the vulnerability to infection and posing a considerable threat to life safety.Therefore, the extent of burn injuries must be accurately assessed, and timely medical intervention must be provided to reduce the associated morbidity and mortality [4].Depending on how deep and severely they penetrate the skin's surface, skin burns generally can be classified into three degrees of severity [5].First-degree burns (or superficial burns) involve the epidermis only, which appear to be pink-to-red without blistering and heal without scarring in a few days.Second-degree burns (or partial-thickness burns) extend through the epidermis and into the dermis, which can be further divided into second-degree superficial and second-degree deep.A second-degree superficial burn affects the superficial layer of the dermis, which takes 2 to 3 weeks for healing with minimal scarring.A second-degree deep burn involves the deeper reticular dermis, which takes longer time for healing, but scarring is unavoidable.Third-degree burns (or full-thickness burns) are most severe and affect all layers of the skin and into subcutaneous fat or deeper, usually taking longer than 8 weeks to heal and requiring surgical intervention.The visual assessment method is commonly used to diagnose burns, but its accuracy of second-degree burns is relatively low, even for expert burn wound clinicians [6].Furthermore, empirical investigations have shown that better outcomes in terms of scarring and aesthetic appearance can be achieved if burn wounds can be successfully healed within three weeks [7].Consequently, medical practitioners must promptly and accurately assess the extent of burns and develop an appropriate treatment plan to achieve complete wound healing within this critical period.Currently, various medical imaging techniques have been developed for burn diagnosis, including histopathological biopsy [8], fluorescence imaging (FI) [9], infrared thermography (IRT) [10,11], magnetic resonance imaging (MRI) [12,13], and ultrasound imaging (US) [14,15].However, these techniques have inherent limitations that impede their applicability to burn diagnosis and prevent the provision of an objective and comprehensive assessment of skin burn status.For instance, histopathological biopsy carries a risk of trauma [16], FI may lead to adverse reactions [17], IRT necessitates specific operational environments [18], ultrasound imaging exhibits limited resolution and mandates contact with the injured surface [19], and MRI remains susceptible to the influence of factors such as tissue edema and hemorrhage [12,13].
Optical coherence tomography (OCT) is a high-resolution, non-invasive, and label-free optical imaging technique that rapidly provides cross-sectional and volumetric images [20].In ophthalmology, OCT has become an essential imaging tool for diagnosing and evaluating many eye diseases [21][22][23].In cardiology, OCT has also achieved successful clinical translation for coronary atherosclerosis assessment and intervention [24][25][26].Besides, OCT is gaining significant attention in other areas, such as dermatology, brain science, and dentistry [27][28][29][30][31][32][33].As OCT continue to evolve, various functional OCT extensions including OCT angiography, polarization-sensitive OCT (PS-OCT) and optical coherence elastography (OCE) have been developed, which have significantly broadened the potential applications of OCT by providing new sources of information about blood flow, birefringence, or elasticity beyond pure structural features [32][33][34][35][36]. PS-OCT is a functional extension of OCT that offers the advantages of structural OCT, including non-invasiveness, high sensitivity, and rapid imaging capabilities, while also possessing the capacity to acquire the birefringence characteristics of tissues through the measurement of polarization properties of backscattered light from the samples [37][38][39].The dermis of the skin has abundant collagen molecules with unique molecular construction and ordered arrangement that exhibit birefringent properties.Exposure of skin tissue to high temperatures leads to the denaturation of these collagen molecules, causing a transition of collagen from a rod-like alpha helix structure to a more disordered random-coil conformation [40].The thermal disruption of collagen's intra-and inter-molecular bonds influences the skin tissue's structure, elasticity, and firmness [41].Furthermore, different degrees of burn injuries yield varying levels of degradation in the birefringence properties of the tissue.This phenomenon can be detected using PS-OCT technology.Consequently, using PS-OCT to detect alterations in tissue birefringence characteristics is an efficacious means of evaluating the severity of tissue burns [42][43][44][45].The technique of single-input state PS-OCT based on single-mode fiber optics has been reported in previous literature [46,47].This design of PS-OCT has the advantages of low system complexity, no ghost peaks, low cost, and ease of integration, which is more practical for skin burn investigation.Nevertheless, due to the optical characteristics of biological samples and the instability of the system, the contrast of PS-OCT image is not high enough which requires further image processing for enhancement.
This study employs a high-precision scalding apparatus to create controlled burn injuries on ex vivo porcine skin, resulting in varying degrees of severity.A home-built all-fiber-based PS-OCT system with single-state input is utilized to acquire the polarization state of optical signal backscattered from the skin tissue.A simple and effective PS-OCT data processing algorithm is used to extract multiple birefringent parameters, including phase retardation (PR), optical axis orientation, and degree of polarization uniformity (DOPU).These birefringence parameters can describe the characteristics of skin tissue in terms of composition, growth pattern, and morphological structure.However, due to the limited birefringent information, the single birefringence parameter has relatively poor image contrast, and the individual polarization metrics prevent a comprehensive characterization of the complex skin tissues.Thus, we adopted principal component analysis (PCA) to fuse the diverse polarimetric parameters synergistically, generating images with more information content, clarity, and tissue contrast.Furthermore, the method of PCA is also utilized to reduce the feature dimensionality for the multi-parameter three-dimensional polarization datasets, which produces enhanced, high-contrast en-face projections to improve the demarcation of burned skin tissue.

Experimental setup
We built a PS-OCT system based on swept laser and single-mode fiber optics with a single input polarization state.We employed a swept laser (HSL-20 from Santec OIS Corporation) with a 100 kHz swept rate, 1310 nm center wavelength, > 12 mm coherence length, and ∼ 90 nm wavelength scan range.The schematic diagram and photograph of the system are shown in Figs.1(a) and 1(b), respectively.A standard swept-source OCT setup consists of a swept laser source, a reference arm, a sample arm, and a balanced photodetection unit.On this basis, our PS-OCT system adds four polarization controllers (PCs) to adjust the polarization state and two fiber polarization beam splitters (FPBSs) to separate the OCT signals into horizontal and vertical polarization channels.The PS-OCT system needs to adjust the PCs to calibrate the polarization state.The calibration steps are as follows: (1) adjust PC1 so that the light incident on the sample is circularly polarized light; (2) place a mirror at the sample stage, and adjust PC3 and PC4 so that the optical power in H channel is maximum and the optical power in V channel is the minimum; and (3) adjust PC2 so that the light reflected from the reference arm has the same optical power in the H and V channels.This setup uses a swept laser to achieve high imaging speed, long imaging range and excellent roll-off performance.Single-mode fiber optics can avoid ghost artifacts and offer easy alignment in the PS-OCT system.Figure 2 shows the SNR roll-off curves and the axial resolutions measured at different depth positions by changing the position of reflector in the reference arm.The imaging range of the PS-OCT system is 5.14 mm.The sensitivity of the system is ∼ 103 dB near the zero-depth position and maintained at >95 dB within the imaging range, which is mainly contributed to the narrow instantaneous linewidth of the swept laser.As shown by the red stars and dashed line in Fig. 2, the axial resolution is also relatively stable, varying from 17 µm to 18 µm within the imaging range.It should be mentioned that the axial resolution is mainly determined by the spectral bandwidth, where our OCT system has a full width at half maximum bandwidth of ∼ 40 nm.The lateral resolution is 21 µm at the focus position.In addition, by placing a quarter-wave plate in the sample arm, the stability of the phase difference between the H and V channels was measured, which had a standard deviation of <42 mrad with over 10 dB SNR.We created an accurate and reliable burn injury model by utilizing a desktop burn device with precise temperature control (Fig. 1(c)).Three patches of porcine skin were utilized to establish the skin burn experiment, where each patch of the porcine skin was scalded by setting the burn device at temperatures of 70, 75, 80, 85, 90 and 95 °C, with a duration time of 30 seconds.As the scald temperature increases, the ex vivo porcine skins' burning degree gradually intensifies, which manifests as the color of the burn site gradually deepens.The main reason is that the porcine skin can lose moisture and undergo thermal decomposition, and skin pigment begins to deposit at high temperatures, which make the burn area darker.After completing the preparation of burn models with different burn levels, the samples need to cool down to room temperature to ensure that the burn model is stable.Afterward, the injured skin models were imaged by the PS-OCT system.The PS-OCT imaging was set to 10 mm × 10 mm field of view, with 600 B-scan frames and 600 A-lines per frame.In addition, cedar oil, which has a refractive index of ∼1.516, was applied to the skin surfaces for refractive index match to reduce strong surface reflection and improve optical penetration into the tissue.

Theory of PS-OCT
The phase retardation (PR) δ c (z) of the sample can be derived from the amplitudes (A) of the horizontal and vertical channels, as shown by the following equation [39]: where z refers to the axial depth position, and A represents the amplitude.The optical axis orientation of the sample, denoted as α c (z), can be determined by assessing the phase difference (∆ϕ) between channels H and V.This relationship is formulated as follows [46]: where ϕ 0 is a constant phase offset originated from the phase delay between the H and V portion of the reference arm.The depolarization property of the sample can be estimated by the parameter of DOPU, which can be defined as follows [48]: where [I Q U V] is a four-element vector that refers to the Stokes vector.
V , which describes the light in horizontal and vertical polarization states.U = 2A H A V cos∆ϕ, which describes the light in + 45°and − 45°p olarization states.V = 2A H A V sin∆ϕ, which represents the light intensity in left-handed and right-handed circular polarization states.The symbol m indicates the mean value within a window or kernel, where the kernel size is 5 × 7 pixels in this study.

Method of PCA-based image fusion
PCA is a highly effective multidimensional statistical analysis technique frequently employed for data dimensionality reduction and feature extraction.The fundamental concept of PCA is to transform the high-dimensional data into a novel coordinate system to obtain lower-dimensional data while preserving most of the data's variation [49,50].It calculates the weights following each principal component's contribution to the overall variance, thereby ascertaining the significance of each principal component.Typically, the components with lower weights are often associated with noise, and discarding these components can help reduce the noise in OCT images.The computational steps of PCA are described as follows: Step 1: Data standardization.As the datasets obtained from the PS-OCT system have different scales, the variables with larger ranges will dominate over those with smaller ranges, leading to biased results for PCA.Each attribute is ensured to have the same level of contribution; the datasets of PS-OCT require standardization, which can be done by subtracting the mean and dividing by the standard deviation.
Step 2: Covariance matrix computation.The covariance matrix serves as a representation of the linear associations among variables.The diagonal elements of the matrix represent the variances of individual variables, while the off-diagonal elements depict the covariances between the variables.
Step 3: Eigenvalues and eigenvectors computation.By calculating the eigenvalues and corresponding eigenvectors of the covariance matrix, we can find the principal components of the datasets.The eigenvectors present the directions of the axes that contain most of the uncorrelated information.The eigenvalues give the amount of variance carried in the directions of eigenvectors.
Step 4: Principal components selection.The selection of principal components is performed based on the cumulative contribution rate, defined as the summation of each principal component's contribution to the total variance.By ranking the eigenvectors in the order of the eigenvalues from highest to lowest, we can obtain the principal components in order of significance.
Step 5: Data transformation.After selecting principal components, the input dataset remains the same on the original axes.This step aims to reorient the data from the original axes to the ones represented by the principal components, which can be done by multiplying the transpose of the original dataset by the transpose of the feature vector obtained from the principal components.
Different polarization parameters of skin tissue exhibit distinct values and variations, which is complicated to assess the severity of skin burns.We applied PCA on multiple en-face polarization images to enhance the visualization of skin burns, which is referred to Scheme 1 of PCA-based image fusion.The workflow is shown in Fig. 3, using the PS-OCT data of 95 °C burn model.The processing workflow has four key steps.First, the 3D datasets of DOPU, PR, and axis orientation were normalized and averaged in an in-depth direction to construct en-face projections (shown in Fig. 3(a-c)).It should be noted that the PR and axis orientation parameters are ambiguous, which can degrade the image quality of en face projections by averaging.Second, PCA was applied on these en-face images, following the standard PCA routine described above.Third, new images were reconstructed by data transformation into the three components (i.e., C1, C2, and C3), achieving feature extraction from the original datasets.Fourth, the reconstructed images of C1, C2, and C3 were mapped into the RGB channels to reconstruct a pseudo-colored en-face fusion image (Fig. 3(g)).
In addition, we proposed another algorithm that utilized PCA for dimensionality reduction of the 3D polarization volume along the depth direction to generate high-contrast en-face images, which is referred to Scheme 2 of PCA-based image fusion.The processing workflow is illustrated in Fig. 4, using the 3D DOPU data of the 95 °C burn model as an example.The traditional way to squeeze the 3D volume data is to perform mean value projection to achieve an en-face image (Fig. 3(a)).The PCA algorithm is applied to the 3D polarization volume to derive the principal component projections.The top three principal components were retained, typically having over 90% cumulative contribution.These principal components can be fused in two ways.One way is to fuse gray-scale images by combining them with their fractional variance contributions weights, as shown in Fig. 4(b).The other way is to fuse them into a pseudo-color image (Fig. 4(c)), where red, green, and blue correspond to C1, C2, and C3, respectively.

Results
We used the PS-OCT system to obtain structural and polarization images of the ex vivo porcine skin tissues burned at different temperatures.The burn device can create a circular burn area with a diameter of 6 mm. Figure 5 shows the cross-sectional OCT images of structure, PR, and DOPU for these porcine skins.The PS-OCT system has a penetration depth of ∼1.5 mm in the porcine skin, covering the epidermal layer and part of dermis layer.The images in the first row are the baseline of the porcine skin without burns.In the structural images, the thermal damage slightly increases the OCT penetration, but the appearance of normal and burned skin is quite similar.Since skin collagen is a weakly birefringent material which can alter the polarization state of light, the value of PR gradually increases with depth in normal skin.The orientation of the collagen fibers in normal skin is relatively irregular, causing decrease of DOPU with depth.When contact with high temperature, the collagen begins to denature and fuses with their neighbors.The burned skin becomes more isotropic, losing its property of birefringence and depolarization.As shown in the cross-sectional PR image, the color of the burned area in the skin tends to be blue, i.e., low PR value.On the other hand, the color of the burned area in DOPU images tends to be red, i.e., high DOPU value.When the burn temperature becomes higher, the areas of low PR value and high DOPU value expand to deeper position since the burned depth increases.The PS-OCT images indicate that the values of PR and DOPU can be employed to estimate the burned depth for skin.As the burn temperature increased, there was a tendency for the PR value to decrease and the DOPU value to increase.A PR threshold of 0.4 rad and a DOPU threshold of 0.7 were employed to delineate the estimated burned depth profile for each cross-sectional image.The average and standard deviation of the burned depth profiles in the burned areas of three porcine skin patches were calculated.Figure 6 represents a bar chart with error bars of the estimated burned depth by PR and DOPU at different burn temperature.The burned depths estimated by PR and DOPU exhibit a strong correlation with one another, and they increase linearly with the burn temperature to a certain extent.Since the skin burn severity is associated with the burned depth, the images of PR and DOPU can be used to predict the skin burn degree.The porcine skin sample is composed of three layers of tissue, which are epidermis, dermis and subcutaneous layer from the outside to the inside.The epidermis, or outermost layer, has a thickness of roughly 0.1 to 0.3 mm.The dermis layer is intertwined with elastin and collagenous fibers to form the main support structure of the porcine skin, which has a thickness of approximately 2 mm.The subcutaneous tissue is the innermost layer, generally ranging in thickness from a few millimeters to several centimeters.Based on the relationship between estimated burn depth and temperature shown in Fig. 5 and 6, we attempted to classify the burn degree of the porcine skins.The porcine skins burned at 70 °C temperature are categorized as first-degree burns, with an estimated burn depth of around 200 µm, primarily affecting the skin's epidermis.The porcine skins burned between 75 and 85 °C are classified as second-degree superficial burns, with an estimated burn depth ranging from ∼400 to 800 µm, affecting mainly the superficial dermis of skin.The porcine skins burned between 90 and 95 °C temperature are categorized as second-degree deep burns, which have estimated burn depth of ∼1000 to 1500 µm, damaging the deep dermis of the porcine skins.The PS-OCT system's penetrating capabilities makes it challenging to assess third-degree burns.It should be noted that the approach used in this study to estimate burned depth and classify burned degrees still needs to be confirmed by other reliable methods such as histology.
The algorithm of PCA-based image fusion of multiple en-face polarization images was applied to the PS-OCT data of 70-95 °C burn models.Figures 7(a), 7(b), and 7(c) show the original en-face images of DOPU, PR, and axis orientation, respectively.Figure 7(d) is the corresponding fused image by PCA.The individual images of DOPU, PR, and axis orientation manifest different features of the burned skin tissue.The DOPU image highlights the burned region with bright intensity, while the PR image and axis orientation image have low contrast and demarcation between the burned and normal regions.The PR image and axis orientation image clearly visualize the skin's surface texture.By applying PCA on these polarization images, the fused pseudo-color images greatly enhance the visualization of skin tissue details, where the severely burned areas appear red, the normal skin areas are mainly green or purple, and the surface textures are represented by yellow.The images fused from multiple en-face images by PCA help improve the clarity, contrast, and discrimination between burned and normal skin regions.Instead of using multiple en-face polarization images for PCA-based image fusion, we also applied PCA on the 3D polarization image along the depth direction.Figure 8 shows the results of 3D DOPU data for burn models from 70 °C to 95 °C.Among them, Fig. 8  We use the contrast value and contrast-to-noise ratio (CNR) as evaluation indicators to evaluate the enhancement performance of the PCA-based weighted fusion approach.The contrast value is defined as follows: where w represents the image width, h represents the image height, I(x, y) is the pixel intensity in the X-Y plane, and µ I w×h is the mean pixel value across the entire image.Another important evaluation metric is CNR, defined as the ratio between the signal intensity and background noise levels.CNR is calculated as follows: where u i represents the mean pixel value of regions of interest (ROI), u b is the mean pixel value of background, σ i is the pixel standard deviation of ROI, and σ b is the pixel standard deviation of background.In this study, ROI is the skin burn region, while the background is the intact tissue region.A higher CNR value indicates greater divergence between the thermal damage and intact tissue regions.
Figure 9 provides a quantitative comparison, in terms of image contrast and CNR, between the mean value projections (Fig. 8(a)) and PCA-weighted fusion en-face images (Fig. 8(b)) of 3D DOPU image data.The mean value projections have contrast values ranging from 15 to 20, whereas the PCA-weighted fusion approach considerably enhances the contrast values, achieving 146%, 153%, 178%, 190%, 197%, and 196% improvement over the mean value projections for the burn temperature of 70, 75, 80, 85, 90 and 95 °C, respectively.The larger contrast value (C σ ) indicates more bright and dark levels in the PCA weighted fusion image, which has more pronounced and discernible details.Since the CNRs of these images have quite a wide range, we plot the CNRs in dB scale, which is 10log(CNR) in the y-axis.According to Fig. 9(b), the CNRs of PCA-weighted fusion image have 4.6, 4.5, 5.4, 4.8, 2.2, and 6.1 dB gains over the mean value projections for the burn temperature of 70, 75, 80, 85, 90, and 95 °C, respectively.The gain of CNR demonstrates the superior capability of the PCA-weighted fusion approach to accentuate the polarization divergence between thermally damaged and intact skin regions.We implemented PCA-based image fusion on the 3D PR volumes of the porcine skin burn models.Figure 10 presents the perceptual comparison of the en-face between the methods of mean value projection and the PCA-based image fusion.Figure 10 image by the PCA algorithm, the difference between intact skin areas and burned regions is strengthened, where intact parts mainly appear in red and the burned regions mainly appear in dark blue or dark Quantitative analysis in terms of image contrast value and CNR (shown in Fig. 11) validates the enhancement effect by the PCA-weighted image fusion approach (shown in Fig. 10(b)) compared to results by the mean value projection (shown in Fig. 10(a)).According to the quantitative analysis, the images in Fig. 10(b) achieve contrast improvements of 72%, 71%, 73%, 165%, 195%, and 175%, along with CNR gains of 5.7, 11.8, 10.2, 11.8, 12.1, and 12.3 dB, respectively.The PCA approach extracts salient architectural morphology from the 3D PR data sets along the depth direction, generating fused en-face images with superior acuity and contrast.

Discussion and conclusion
In this work, we have implemented an all-fiber-based PS-OCT system with simple configuration and high imaging performance.The swept source we used enables high imaging speed and long coherent imaging range.The optical design for polarization-sensitive function is mainly based on single-mode fiber devices, including polarization controllers and couplers, which can avoid ghost artifacts.By adjusting the polarization controllers properly, we can achieve single circular polarization input and orthogonal polarization channels for signal detection.Compared to other optical designs of PS-OCT, such as bulk polarization optics [51,52], polarization maintaining fiber optics [53,54], depth multiplexing [55], and electro-optic modulation [56], the design of PS-OCT in this study is compact and low cost, and has low system complexity and less ghost artifacts, suitable for many biological research and clinical applications.
Besides, we have investigated the schemes to apply PCA to improve the image quality for skin burn assessment.One scheme uses PCA to integrate the en-face images of DOPU, PR, and optical axis orientation, improving clarity, contrast, and discrimination between burned and normal skin regions.Another scheme is applying PCA on 3D polarization data volume along the depth direction to generate high-contrast en-face images.Image fusion can reconstruct the en-face image in two ways: one is to fuse gray-scale images by the weights of PCA, and the other is to fuse them into pseudo-color images.The fused gray-scale images comprise shades of gray, carrying only one channel of information, which make it easier to quantify the contrast and CNR for skin burn assessment.The fused pseudo-color images are useful for enhancing the visual perception, improving the interpretation of the image content and highlighting the features of skin burns.Both schemes exhibit higher clarity and tissue contrast for skin burn assessment than the mean value projection images.The scheme of 3D-data PCA fusion shows superior visualization enhancement compared to the scheme of en-face PCA fusion.This is because the averaging operation in en-face PCA-fusion scheme leads to too much information loss and lacks sensitivity to skin burn changes, while the approach of 3D-data PCA-fusion takes full advantage of the 3D data information and highlights the most important features and relationships along depth direction.It should be noted that the PCA-based image fusion is an unsupervised method which can be difficult to interpret the principal components.The experimental results and quantitative analysis validate the efficacy of the PCA-based image fusion method for augmented tissue demarcation, enabling unambiguous discrimination between thermal damaged and intact skin regions.
A simple and efficient PS-OCT system based on an all-fiber and single-state input was implemented for skin burn investigation.The PS-OCT system has the advantage of providing multiple parameters of birefringence information, enabling high-resolution and depth-resolved 3D imaging of skin tissue.The parameters of PR and DOPU are strongly associated with the birefringence and depolarization of skin, which can be employed to quantify the burned depth and predict the skin burn degrees.Besides, PCA is an effective feature extraction methodology, transforming multidimensional data into more compact and interpretable representations.This study demonstrates the validity and utility of PCA-based fusion of PS-OCT images to assess the porcine skin burn.Compared to conventional mean value projection of PS-OCT images, the proposed PCA-based algorithm improves burn tissue visualization, contrast, and information content.The techniques used in this work have the potential to provide more precise quantitative evaluation and enhanced visualization of skin burns, which is promising to push the PS-OCT technology for clinical burn diagnosis.

Fig. 2 .
Fig. 2. The signal-to-noise ratio (SNR) roll-off curves and the axial resolutions measured at different depth positions.

Fig. 3 .
Fig. 3. Diagram of PCA-based image fusion of multiple en-face polarization images.(a), (b), and (c) are the en-face images of DOPU, PR, and orientation, respectively.(d), (e), and (f) are the cross-sectional images of DOPU, PR, and orientation, respectively.(g) is the pseudo-colored image fused by PCA.

Fig. 4 .
Fig. 4. Diagram of PCA-based image fusion of 3D polarization images (DOPU data of the 95 °C burn model is used as an example) along the depth direction: (a) Traditional method of mean value projection.(b) Gray-scale weighted fusion by PCA.(c) Pseudo-colored fusion by PCA.

Fig. 5 .
Fig. 5. Cross-sectional images of structure (a), PR (b) and DOPU (c) for the ex vivo porcine skins burned at different temperatures with 30 s burn duration.

Fig. 7 .
Fig. 7. Comparison of original en-face polarization images and PCA-based image fusion of these en-face images.Column (a): original en-face images of DOPU.Column (b): original en-face images of PR.Column (c): original en-face images of axis orientation.Column (d): PCA-based image fusion of these en-face images.
(a) shows the mean value projection images of the 70 °C to 95 °C burn model, Fig. 8(b) shows the PCA weighted fusion en-face images of the 70 °C to 95 °C burn model, and Fig. 8(c) shows the PCA pseudo-colored fusion en-face image of the 70 °C to 95 °C burn model.Among these images, the PCA weighted fusion en-face images and the PCA pseudo-colored fusion en-face images exhibit higher clarity and tissue contrast than the mean value projection images.For example, in the slight burn models of 70 °C and 75 °C, the thermally damaged regions in Figs.8(b) and 8(c) exhibit well-demarcated boundaries between the burn area and the surrounding intact tissue.

Fig. 9 .
Fig. 9. Quantitative comparison of contrast value and CNR between the mean value projection and PCA weighted fusion en-face image of 3D DOPU image data.

Fig. 10 .
Fig. 10.Perceptual comparison of mean value projections (a), PCA-weighted fusion en-face image (b), and PCA pseudo-colored fusion en-face images (c) of 3D PR images data.

Fig. 11 .
Fig. 11.Contrast and CNR of principal component weighted fusion en-face and mean en-face of PR