2.1.

Biomechanical Response of Skin, Physiology, and Age

Alterations in systemic physiological conditions could be reflected in the biomechanical response of skin. For instance, the imbalance of fluid pressure, decreased drainage capacity, and increased permeability of the vasculature may lead to fluid accumulation within the extracellular space.¹³ By utilizing palpation or localized compression along with visual inspection, the physician can clinically evaluate the mechanical response of skin tissue, which can facilitate the detection of fluid retention or edema (an accumulation of interstitial fluid at the millimeter scale that is common in the lower extremities) and infer both local-regional and systemic diseases, such as chronic venous insufficiency, deep venous thrombosis, lymphedema, heart failure, and diabetes.^14,15 Additionally, it is widely recognized that aging can also influence the morphology and functions of the skin (such as those that occur in collagen and elastin) and, hence, could also affect the overall biomechanical response of the skin. Structurally, aged skin exhibits reduced dermal thickness, collagen content, ground substances (mainly composed of water and glycosaminoglycans), and vascularity (particularly observed in the papillary dermis).^16–18 Functionally, decreased rates of dermal clearance of fluids, vascular responsiveness, and moisture content of the stratum corneum were also reported.^16–18 Adult skin and infant skin have demonstrated different properties. Compared to adult skin, the collagen in the upper reticular dermis of infants is thinner and less dense than in adults. Moreover, the level of skin hydration of older infants (3 to 12 and 8 to 24 months) was observed to be significantly higher than that of adults.¹⁹ The above-mentioned age-dependent changes in skin could contribute to an alteration in both the intrinsic stiffness and the time-dependent fluid redistribution behavior.^11,16 To date, the diminished resilience and elastic recovery after deformation of aged skin have been widely recognized.^11,16,18 In particular, it has been observed that the recovery time of mechanically depressed skin of younger adults (minutes) is greatly different from that of older adults ($>24\mathrm{h}$).²⁰

Applying a millimeter-scale deformation on human skin tissue affects not only the skin surface geometry but also the optical properties of skin tissue due to the temporary change in architectural morphology and interstitial fluid resulting from the applied indentation. Mechanically induced changes in optical properties have been revealed previously using OCT. Studies have shown an elevation of the backscattered intensity amplitude within the skin specimen due to the compression-induced local fluid transport.²¹ In fact, when a pressure gradient is introduced to the tissue, the unbound water or interstitial fluids are expelled aside, which decreases the absorption coefficient of the skin, especially when the light source is close to 1450 nm (where water has high absorption peak).²¹ In addition, mechanical compression can lead to an increase in optical tissue homogeneity, closer packing of tissue components, and a decrease of tissue thickness at the indented site.²² As a result, the index mismatch in refractive index is reduced, and the optical attenuation diminishes; hence, a greater backscattered intensity and penetration depth are observed.^22,23 Mechanical compression has also been applied with optical clearing, and the prolonged clearing effect on in vivo human tissue even after the release of a localized stress has been reported.^21,22 In addition, compression-induced alterations of OCT intensity contrast within tissues has been shown to be effective for distinguishing between ex vivo inflamed and tumor-bearing rectal tissues.²⁴

2.2.

Mechanical Compression and Tissue Morphology

Upon indentation, morphological alterations within the skin tissue, which can affect the size, quantity, and distribution of the scatterers, could also occur and, hence, alter the OCT speckle patterns. A number of researchers have investigated methods to differentiate or classify pathological tissues based on the speckle pattern or statistics. Based on OCT signal intensity, Fourier or autocorrelation analysis detects the periodic pattern of the tissue structures.²⁵ Basic statistics, such as the statistical moments and Gamma fitting of the intensity histogram, have also been applied for tissue characterization.^26–28 Commonly performed on OCT images, texture analysis often takes advantage of other statistical features that are both intensity sensitive and directionally sensitive, such as the co-occurrence matrices²⁹ and the gray-scale run-length matrices.³⁰ In this paper, fractal analysis was utilized to characterize the intensity distribution within the tissue. The central concept of fractal analysis is to reveal the statistical self-similarity and geometrical complexity of an object based on the box-counting dimension. Previously, fractal analysis has been carried out in OCT images and has demonstrated its capability of distinguishing various types of tissue components within ex vivo arterial tissues and cancerous tissues.^31,32

3.1.

Human Subjects

Two groups of subjects with distinct age differences were recruited to participate in this study: eight Caucasian female adults (20 to 33 years of age, two pregnant and six nonpregnant) and eight Caucasian infants (9 to 18 months of age, four females and four males). All the adult participants and the legal guardians of the infant participants provided written, informed consent before participating in the study. The study protocol was reviewed and approved by a Procter and Gamble (P&G) clinical research review panel and was determined to be of minimal risk. The study was conducted with oversight from P&G’s clinical personnel in a manner consistent with Good Clinical Practice principles and procedures. The pregnant adults were excluded from the study due to well-known physiological changes in skin, such as the presence of edema, exhibited during pregnancy.³³ For the adult group, the indentation sites were selected on the volar forearm (a quarter of the wrist-crease length measured from the elbow crease), while, for the infants, the indentation was induced on the lateral thigh (approximately halfway between the knee and the buttock). The two body sites were selected due to their comparable scale of circumferences, which would result in a similar scale of displacements applied to the skin, as the strain exertion was fixed in this study. For the adults, the body mass index (BMI) and body fat percentage (BF %) were also obtained; the former was calculated based on the height and weight of the subject and the latter was measured by a body fat analyzer (Model HBF-306, Omron Healthcare, Inc., Illinois). Demographics of the subjects are included in Tables 1 and 2.

Table 1

Demographics of human adult subjects.

Table 2

Demographics of human infant subjects.

3.2.

Indentation Procedures and Devices

For all subjects, the targeted indentation locations were marked prior to the application of the localized loading to ensure the image locations were kept the same at all times. Before indentation, baseline OCT data were acquired. Subsequently, indentation was applied on in vivo human subjects by a customized polymer-based indenter [Fig. 1(a)]. Plastic indenters with a dimension shown in Fig. 1(b) were machined, and an elastic band (Seraket^® Automatic Tourniquet 48300-770, Propper manufacturing Co., Inc., New York) was used to hold the indenter in place for 3 min [Fig. 1(c)]. The strain applied was fixed at $\sim 16.7\%$, so the length of the elastic band was adjusted according to the circumferences of the body sites of the subjects. Due to the large strain applied ($>10\%$), a red marking and indentation could be visually observed on the skin upon the release of the indenter [Fig. 1(d)]. After indentation, the OCT probe was immediately placed at the indented skin site for image acquisition at the following time-points: immediately ($<10\mathrm{s}$, for simplicity, this is denoted as “0-min”) and 1, 2, 3, 4, and 5 min after the indentation (postindentation). The observation period of the skin recovery was chosen at the minute-scale based on the finding of a previous study.²⁰ During the imaging session, the subjects’ forearms were rested on a metal “hand stand” covered with a soft fabric to keep the skin surface flat and stable at all times [Fig. 1(e)].

Fig. 1

Indentation device and setup. (a) The photograph of the custom-made polymer indenter, along with (b) its geometry and dimensions. The measured dimensions are illustrated in millimeters (mm). (c) The elastic strap used to hold the indenter in place. (d) A representative surface image of the in vivo human skin tissue, immediately ($<10\mathrm{s}$) after the release of the indenter. The scale bars in both the vertical and horizontal directions both represent 1 mm. The vertical red line indicates the scanning location for OCT images. (e) An adult subject undergoing the 3-min indentation procedure.

3.3.

Optical Coherence Tomography Instrumentation and Data Collection

A commercial spectral domain OCT system (center wavelength $\sim 1300\mathrm{nm}$, Telesto SD-OCT, Thorlabs, Inc., Germany) with an axial resolution of $\sim 6.5\mu \mathrm{m}$ in air and a lateral resolution of $\sim 15\mu \mathrm{m}$ was utilized. A 36-mm focal length objective (LSM03, Thorlabs, Inc., Germany), which provided a Rayleigh range of $\sim 106\mu \mathrm{m}$, was used for imaging. The B-scans acquired from the OCT system had dimensions of 512 pixels ($\sim 2.5\mathrm{mm}$) along the depth and 1024 pixels ($\sim 6\mathrm{mm}$) along the lateral direction. During imaging, an axial line scan rate of $\sim 5.5\mathrm{kHz}$ was utilized and a sensitivity of $\sim 111\mathrm{dB}$ was achieved. The signal-to-noise ratio was 77 dB, and the signal fall-off measured between 10% and 90% of the imaging depth was 3.5 dB. The $1/e$ penetration depth of light into the nonindented skin tissues was characterized to be $\sim 0.40\pm 0.07\mathrm{mm}$ ($\text{average}\pm \text{standard deviation}$) in optical distance.

A customized polymer-based attachment was mounted at the end of the rigid OCT scanner to enable handheld imaging and user-friendly assessment of the body sites of interest (i.e., volar forearm for adults and lateral thigh for infants), as well as to keep the working distance fixed between the tissue and the lens [Figs. 2(a) and 2(b)]. To minimize the possibility of introducing vascular occlusion or other sources of perturbation near the target sites of interest, the physical dimensions of the probe attachment [Fig. 2(a)] were designed so that the diameter of the aperture ($\sim 2.7\mathrm{cm}$) was significantly larger than the width of the indenter tip ($\sim 0.75\mathrm{mm}$). In addition, the ring-shaped edge around the aperture was designed to have a larger area ($\sim 24.5{\mathrm{cm}}^2$) so that the amount of pressure applied at the edge would be reduced.

Fig. 2

The handheld probe and the imaging process. (a) The design and the dimensions of the handheld attachment. The photographs were taken from (left) the side, (top right) the side facing the skin, and (bottom right) the side carrying the rigid scanner head. The dimensions of the aperture are illustrated in centimeters (cm). (b) By mounting the rigid scanner to the attachment, (c) handheld OCT imaging was enabled. The vertical and horizontal scale bars represent 4 and 3 cm, respectively.

Standard OCT data processing on the raw data was performed using the commercial software available from Thorlabs (OCT 3.0.7, Thorlabs, Inc., Germany). The OCT data were saved as 8-bit gray scale images. These intensity-based images were then used as the input for extracting the different OCT features used in this study, where all the quantitative processing was performed on MATLAB^® (R2016a, MathWorks, Massachusetts). Multiple frames were acquired on each patient until at least three stable frames of the images were obtained. Eventually, three frames were utilized as image inputs per time-point (i.e., pre- and postindention at 0, 1, 2, 3, 4, and 5 min) per patient for quantitative analysis. As a result, the analyses included three replicate measures ($n=3$) for each patient and $N=8$ (number of subjects) for each group, leading to a total number of frame $N_{total}=24$ per scenario. Representative OCT images of the adult and the infant skin are shown in Fig. 3.

Fig. 3

Representative OCT images of in vivo human skin before and after indentation. (a) Representative adult forearm skin and (b) representative infant thigh skin. Both the horizontal and vertical bars indicate $500\mu \mathrm{m}$.

3.4.

Image Preprocessing

To correct for imaging artifacts due to the presence of surface skin hair and misalignments induced by inevitable subject and operator motion, the OCT images were preprocessed. To assist artifact removal and image alignment, a skin surface boundary was first identified by determining the axial location at which the cumulative intensity along the depth was greater than the empirically determined threshold value. Subsequently, artifacts were removed from the images [Figs. 4(a)–4(g)]. The surface hair artifacts (above the surface) were identified by finding the lateral location where the averaged intensity within the “air ribbon” region was above a threshold [Figs. 4(c) and 4(d)], and the shadow artifacts (beneath the surface) were identified by searching for the lateral position where the intensity slope beneath the skin surface was below a threshold [Figs. 4(e) and 4(f)]. Finally, image alignment in both lateral and axial directions was implemented. Laterally, the “indented site,” as opposed to the nonindented skin regions surrounding the indented site, was determined by the local minimum (valley) of the surface boundary [Fig. 4(i)]. To ensure that the in-tissue OCT features could be extracted from the dermis, the data were axially aligned to the DEJ, which was identified based on the maximum intensity found in each moving-average filtered A-scan [Fig. 4(k)]. Detailed preprocessing procedures are described in the Appendix.

Fig. 4

Data preprocessing for (a–g) artifact removal and (h–l) image alignment. (a) Original image with (red triangle) surface hair and (red squared box) shadow artifacts, with a zoomed-in image of the latter shown on the right. (b) Moving average filtered $I_{\text{sum}}$ map. (c) The initial estimation of the surface including the surface hair, $S_i$, (green trace), and excluding the surface hair, $S_f$, (red dashed trace). The upper boundary of the “air ribbon” region is also identified (cyan trace). (d) The corresponding ${\overline{I}}_{\text{air}}(X)$ at each column (blue plot) and the threshold (red line). (e) Aligned and smoothed (b) $I_{\text{sum}}$ map with a depth of 150 pixels. (f) The corresponding $Sl(X)$ at each column (blue plot) and the threshold (red line). (g) A combined result after removing the surface reflection (b–d) and shadow artifacts (b, e, f). Image alignment was performed on the artifact-removed unaligned image (h). (i) The center indentation site (yellow dashed line), determined by valley detection at the neighboring region (blue lines). (j) The $D_{residua\mathrm{l}}$ (solid yellow line) determined near the center indented site as the distance between the skin surface (blue lines) and the interpolated line (red dashed) between the two neighboring peaks (filled red dots). (k) Determined DEJ (cyan trace), which was used for (l) axial alignment. The defined center ${\mathrm{ROI}}_{\text{indented}}$ and side ${\mathrm{ROI}}_{\text{nonindented}}$ (red boxes). The center indented site ($x_{\text{indented}}$) (yellow dashed line) determined in (i). The vertical and horizontal scale bars correspond to 200 and $500\mu \mathrm{m}$, respectively. Detailed preprocessing procedures are described in Appendix.

3.5.

Image Analysis and Quantitative Metrics

To characterize the deformed tissue, quantitative OCT metrics were developed based on three aspects: the change in skin surface geometry, optical property, and dermis morphology. The latter two metrics are “in-tissue” parameters.

3.5.3.

Dermis-morphology metrics

Dermal morphology was quantified based on the fractal dimensions (FDs) computed at the regions defined in Sec. 3.5.2. The fractal dimension ratio (FD ratio) was defined as $\frac{{\mathrm{FD}}_{\text{indented}}}{{\mathrm{FD}}_{\text{nonindented}}}$, where ${\mathrm{FD}}_{\text{indented}}$ and ${\mathrm{FD}}_{\text{nonindented}}$ denote an average of the one-dimensional FD (1-D FD) obtained within ${\mathrm{ROI}}_{\text{indented}}$ and ${\mathrm{ROI}}_{\text{nonindented}}$, respectively [Fig. 4(l)]. For example, if the width of the ${\mathrm{ROI}}_{\text{indented}}$ spanned across 68 columns, the 1-D FD would first be calculated from each column, and an average of the 68 1-D FD values would be used to represent ${\mathrm{FD}}_{\text{indented}}$. In each ROI, the 1-D FD of each column was obtained based on a 1-D box-counting method. The total length-of-interest of each column ($l_0$) was determined by the highest power-of-two value less than or equal to the axial range of the original ROI. For instance, if the original ROI had an axial range of 35 pixels, $l_0$ would be $2^5=32\text{pixels}$.

Within each ROI, each column was segmented into several intervals, each with a length of $l_i$, where $l_i=\frac{l_0}{2^i}$ was the length of the interval at the $i$’th iteration. The number of iteration was determined by $l_0$ so that a minimum length of 1 pixel was achieved for $l_i$ eventually. For instance, a $l_0$ of $2^5\text{pixels}$ would result in $i=\mathrm{1,2},\dots ,5$ and $l_5=1$. The number of intervals that included any pixel with an intensity greater than the threshold value ($I_{\text{threshold}}$) was denoted as $N_i$. Note that $I_{\text{threshold}}$ was adaptively determined so that only the predominated signal, instead of the noise floor, was picked up. Initially, the FD as a function of various threshold intensity values was obtained. Eventually, the threshold value that yielded an FD of 75% of the maximum achievable FD [typically 1, as shown in Fig. 5(b)] was determined as $I_{\text{threshold}}$. The 1-D FD of each column was determined as the slope of log ($N_i$) over $\log (\frac{1}{l_i})$, fitted via linear least square regression [Fig. 5(a)]. The averaged $R$-squared values for the fitting at all ROIs for all data are $\sim 0.93$. The FD infers the self-similarity in the sense that each part of the curve reflects the entire curve except for a scaling factor difference.³² A more disordered or complex pattern is expected to give rise to a higher FD value.³¹

Fig. 5

Determination of the 1-D FD and the threshold intensity ($I_{\text{threshold}}$). (a) Representative log–log plot of number of boxes ($N_i$) versus the box size ($l_i$) for calculation of FD for one column. The slope obtained by linear least square fitting is the FD. (b) The $I_{\text{threshold}}$ was defined as the intensity value that resulted in an FD equivalent of 75% of the maximum FD.

3.6.

Statistical Analysis

Given the relatively small sample size and nonnormal distribution of the data, nonparametric tests were used. A two-sided sign test was performed on the nonindented and the 0-min postindented skin from the same group of subjects, and a two-sided Wilcoxon rank sum test was performed on the infant and the adult skin at the postindentation scenarios. For both tests, the significance level was set to 5%.

4.1.

Indentation-Induced Alterations

As shown in Fig. 6, before the application of any localized compression, the surface geometrical feature, quantified by RD %, exhibited a value close to zero (a flat skin surface) for both datasets. Immediately after the removal of the indenter (0-min postindentation), visible deformation was seen in the OCT images, while positive RD % values were detected for both the adult and the infant datasets ($p\text{-}\text{value}<{10}^{-5}$ and ${10}^{-6}$, respectively). This indicated a delayed recovery of the skin, which was similar to the observation reported in previous studies.^5,20 Likewise, the $I$ ratio and the FD ratio were close to unity before the indentation, suggesting little variation in the optical and morphological features across lateral locations of the dermal tissue. However, upon the release of the indentation (0-min postindentation), a significant increase was observed for both the $I$ ratio ($p\text{-}\text{values}<{10}^{-5}$ for adults and ${10}^{-2}$ for infants) and FD ratio ($p\text{-}\text{value}<{10}^{-3}$ for both adults and infants) metrics. This indicated that the average intensity, along with the averaged 1-D FD, have a greater value at the indented site as compared to those of the nonindented nearby regions. The observed increase of the backscattered intensity was in agreement with previous studies, showing the compression-induced intensity change within soft tissues.^21,22 In addition, the increased FD value revealed after indentation suggested an increased complexity of the distribution of intensity in OCT images, possibly due to the denser distribution of the tissue scatterers caused by the indentation.

Fig. 6

Results from the adults and the infants, quantified based on the developed OCT features (a–f) and the postindentation recovery coefficients $\gamma$ (g–i). Box plots of the adult skin response at the preindentation and 0–5 min postindentation time-points for the (a) RD %, (b) $I$ ratio, and (c) FD ratio parameters. Box plots of the infant skin responses at the preindentation and 0–5 min postindentation time-points for the (d) RD%, (e) $I$ ratio, and (f) FD ratio parameters. Shown in the third column are the box plots of (g) ${\gamma }_{\mathrm{RD}\%}$, (h) ${\gamma }_{I\text{ratio}}$, and (i) ${\gamma }_{\mathrm{FD}\text{ratio}}$ for both the adults (red) and the infants (blue). ${\mathrm{N}}_{total,\text{adult}}=18$. ${\mathrm{N}}_{total,\text{infant}}=24$.

4.2.

Postindentation Recovery

After the release of the indenter, the three OCT parameters were tracked over time to monitor the rebound of the skin for both the adult forearm and the infant thigh tissues. Illustrated in Figs. 6(a)–6(f), both the adult and the infant datasets exhibited a decrease in the metric values at the postindentation time-points, indicating that the deformed skin was gradually recovering toward its initial preindented state. This could be a combined result of displacement recovery from the elastic tissue components (e.g., collagen, elastin, etc.) and the redistribution of the unbound fluids of the ground substances over time, reflecting the viscous property of skin. To quantify the recovery trend over time ($t$), a weighted least squares fitting based on an exponential function ($e^{-\gamma t}$) was performed on the three OCT metrics to obtain the recovery coefficient $\gamma$. A higher, positive value of $\gamma$ represents a faster rate of recovery toward the preindented state. The $\gamma$ values were calculated for each individual dataset collected from each human subject, and the pool of $\gamma$ values was subsequently utilized for statistical analysis.

Box plots of the recovery coefficients for both the adult and the infant datasets are shown in Figs. 6(g)–6(i), where ${\gamma }_{\mathrm{RD}\%}$, ${\gamma }_{I\text{ratio}}$, and ${\gamma }_{\mathrm{FD}\text{ratio}}$ represent the recovery coefficients of the RD %, $I$ ratio, and FD ratio, respectively. Comparing the results from the adult skin to that of the infant skin, significant differences were observed for both in-tissue parameters, ${\gamma }_{I\text{ratio}}$ and ${\gamma }_{\mathrm{FD}\text{ratio}}$ ($p\text{-}\text{values}<0.01$), but not for ${\gamma }_{\mathrm{RD}\%}$ ($p\text{-}\text{value}=0.6$). Both dermal recovery coefficients from the infant skin were positive (the medians of ${\gamma }_{I\text{ratio}}$ and ${\gamma }_{\mathrm{FD}\text{ratio}}$ were $1.1\times {10}^{-2}/\min$ and $1.4\times {10}^{-2}/\min$, respectively), indicating a tendency of the dermal parameters to return toward the nondeformed states. However, the adult datasets exhibited recovery coefficients that were an order-of-magnitude smaller (the medians of ${\gamma }_{I\text{ratio}}$ and ${\gamma }_{\mathrm{FD}\text{ratio}}$ were both at a scale of ${10}^{-3}/\min$), implying negligible temporal variation within 5 min after the release of the localized loading. This suggests that the rebound of the fluids and the deformable tissue components within the 5-min postindentation period was less appreciable than in the infant group, which agrees with the findings of other studies that aged skin has a decreased rate of dermal clearance of fluids and vascular responsiveness.^16,17 The fact that significant differences were observed in the adult and the infant groups based on the distribution of ${\gamma }_{I\text{ratio}}$ and ${\gamma }_{\mathrm{FD}\text{ratio}}$ suggests that the in-tissue parameters (based on the optical intensity and morphology of the dermis) could be more effective than the surface geometrical metrics in distinguishing the age-induced variations of the biomechanical response within the first 5-min postindentation period. This demonstrates the potential merits of using OCT imaging for quantitative characterization of the biomechanical response of skin tissue.

While these differentiating results were acquired within the 5-min postindention measurement period, it is interesting to note that none of the OCT metrics (both surface geometry and in-tissue-based) completely returned to their initial values within the experimental measurement time period. Additionally, all OCT features at the 5-min postindentation were different from the preindentation features ($p\text{-}\text{value}\le 0.001$ for the adults and 0.06 for the infants), suggesting a high likelihood that mechanical equilibrium had not yet been fully reached after 5-min postindentation. Future studies will extend this temporal characterization to longer time points, which will further improve fitting of the recovery kinetics and provide additional data on potential differences between adult and infant skin conditions.