1. Introduction

A number of optical imaging techniques noninvasively assess articular cartilage surface features and zonal architecture. These microscale features are important for proper articular cartilage function and are often altered in joint pathologies [1–3]. For example, polarization-sensitive optical coherence tomography (PS-OCT) detects depth-resolved birefringence from aligned macromolecules in cartilage, particularly collagen fibrils. Many studies using PS-OCT showed banding patterns and loss of birefringence beneath the articular cartilage surface is associated with early degeneration [4–7]. Studies using polarization-sensitive tomographic approaches are complicated by the dependence of signal on specimen microstructural orientation with respect to input polarization [8,9], and by the need to acquire multiple adjacent tomograms to build an image of a larger volume of tissue. Optical polarization tractography (OPT) creates three-dimensional maps of microstructure in tissue from Jones matrix polarization parameters, including three-dimensional articular cartilage zonal architecture and surface roughness [10–13], skeletal muscle fibers, blood vessels, and heart tissue [12,14–16]. In contrast to tomographic and computational approaches, widefield reflectance microscopy rapidly images large areas of the cartilage surface and, with added polarizers, noninvasively assesses subsurface birefringent features, which makes it potentially useful in the evaluation of cartilage explants and of multiple sites within intact joints.

Cartilage zonal architecture consists of four layers with distinct collagen network microstructure and chondrocyte morphology: the superficial, middle, and deep zones; and calcified cartilage. Together these zones form an arcade-like structure with gradients of mechanical properties that bears a broad range of stresses during normal joint motion [17,18]. An intact superficial zone (SZ) confers normal mechanical and biochemical properties to articular cartilage. The SZ accounts for 10-20% of total cartilage thickness, and is comprised of collagen fibrils arranged parallel to the surface [19,20]. Chondrocytes in this layer appear flattened and express superficial zone-specific genes, contributing to mechanical functioning [21–23] [24]. The alignment of the collagen network within the superficial zone has been mapped using split lines, and contributes to the superficial zone orthotropic tensile modulus [25]. Computational modeling and optical analyses of the superficial zone following mechanical loading and in pathology and aging suggest superficial zone collagen microstructural alignment is functionally important [13,26–29].

Damage to articular cartilage alters the articular surface and superficial zone in both animal models of osteoarthritis and in humans. Articular surface roughening may begin with blisters in the superficial zone depleted of proteoglycans and with degraded/disorganized collagen fibrils [30–32]. Wear of the articular surface produces wear lines parallel to the direction of joint articulation, and also microscale clefts that undercut the superficial zone [25,33]. Electron microscopic analyses display twisted fibril bundles on the articular surface consistent with shear-induced damage to the collagen network [30,34]. Frank fibrillation and advanced stages of wear in osteoarthritis eventually lead to an apparent total loss of the superficial zone [28,35,36]. Thus, articular surface roughness and superficial zone collagen microstructural disruption occur together in cartilage pathologies, and are difficult to assess independently. The extent to which these two hallmarks of articular cartilage damage occur in specimens at different stages of pathological progression is of significant pre-clinical and clinical interest as a potential imaging biomarker of early pathology.

Polarized light microscopy scoring of articular cartilage recently stemmed from more qualitative histological scores lacking microstructural sensitivity. Fibrillation and disruption of the superficial zone is a component of several histology scoring systems [9,30,37]. One of the pioneering scoring systems, the Mankin Score, evaluates large surface irregularities as part of its four categories [38]. The OARSI score considers superficial and focal fibrillation, as well as extent and depth of cartilage lesions, and thus differentiates alterations occurring in early OA as well as more severe forms [39]. Other scores evaluate in vivo repaired cartilage or in vitro engineered cartilage, and count similar features of surface disruption [40]. Polarized light microscopy (PLM) assesses optical birefringence primarily of the collagen network, which is sensitive to local fibril concentration and alignment. The PLM technique generates a map of zonal collagen alignment from unstained histology sections, which can be assessed as a quantitative score from parameter maps, or semi-quantitatively using a grading system [41–43]. These analyses are invasive, requiring processing of cartilage into thin sections to resolve microscale features.

Polarized reflectance microscopy is a rapid, label-free way to assess surface and subsurface features of aligned scattering media [44]. In skin, a collagen-rich multilayered tissue, polarized reflectance signal distinguishes features of carcinomas, burn scars and venous abnormalities [45]. In cartilage, polarized reflectance reveals a lattice texture of collagen birefringence from normal, intact cartilage that was lost in surface-scraped cartilage [44]. These observations led to the hypothesis that polarized reflectance texture originates in the superficial zone, while loss of texture implies more depolarization or loss of the superficial zone. The goal of this study is to relate polarized reflectance image texture to zonal microstructure, surface roughness, and chondrocyte organization. To accomplish this, we first compared polarized reflectance texture alignment to split lines from the same location. We then analyzed the dependence of polarized reflectance signals on superficial zone thickness and total cartilage thickness using multiple linear regression. Finally, we co-registered polarized reflectance maps with fluorescence microscopy of chondrocyte nuclei, quantified polarized reflectance map texture using several quantitative parameters, and performed factor analysis to identify latent factors affecting the polarized reflectance maps. Results indicate that the superficial zone collagen network contributes primarily to polarized reflectance map texture, while surface roughness and total cartilage thickness contribute to map brightness, with implications for use of the technique to non-invasively evaluate multiple microscale features of articular cartilage.

2. Materials and methods

2.1 Cartilage harvesting

Adult bovine knees (n = 6) were obtained from a local abattoir (JW Treuth and Sons, Catonsville, MD) on the same day of slaughter. Explants and cores were subjected to different treatments and endpoints (Fig. 1). Articular cartilage explants were harvested with scalpel and 3 mm diameter dermal punches from the patellofemoral groove without the underlying bone tissue and stored in ice-cold phosphate buffered saline (PBS). On the femoral condyle, osteochondral (OC) cores of cartilage with subchondral bone attached were harvested using an electric drill with a cylindrical core-bit of diameter 15 mm. OC cores were kept cool on ice in PBS until usage. Two types of cartilage specimens, 3mm-diameter explants and 15mm-diameter OC cores, were harvested for different sets of experiments.

 

Fig. 1. Schematic of experimental flow.

Download Full Size | PDF

2.2 Split-line creation

A total of n = 30 OC cores were subjected to split-line generation. A fine-tip circular awl (IMT-8800) dipped in commercial grade India ink (Chartpak Inc., Leeds, MA) was used to prick into the articular surface at the center of OC cores causing the tissue to split with a characteristic, small popping sound. After that, OC cores were rinsed with PBS to wash off any residual ink leaving the surface pattern of clearly visible split-lines. Images of split-lines were captured under a brightfield microscope and co-registered with polarized reflectance maps at the center of the OC cores taken before split-line generation (Fig. 2). Co-registration was performed manually in ImageJ 1.50b (NIH, Bethesda, MD).

 

Fig. 2. Split-line and co-registered Pol texture orientation. (A) Representative co-localized split line (A, first row) and Pol images (A, second row), corresponding to locations 1-5 in (B). All OC cores were taken from the indicated regions of lateral (L) and medial (M) femoral condyles with a cylindrical core drillbit, and imaged in the center. (C) Comparison of split line and birefringence texture orientation, n = 30 OC cores, with cores 11-30 from L and M sites, but not in order as cores 1-10.

Download Full Size | PDF

2.3 Explant treatments

A total of 97 cartilage explants were used in four different experiments (Fig. 1). Specifically, n = 60, 18, 14, and 5 explants were used in experiments 1, 2, 3, and 4, respectively. In the first experiment (exp.1), cartilage explants were divided into three groups (n = 20/group) of control, scraped and collagenase. Scraped and collagenase treatments were similar to previously described [44]. Briefly, for the scraped group, scraping was performed by passing 600-grit ultrafine waterproof sandpaper with a 200 g weight over the top of the explant surface 10 times. For the collagenase group, explants were incubated in collagenase solution (Sigma Aldrich, St. Louis, MO) concentration of 500 units/ml activated in deionized water mixed with 0.36 nM CaCl₂ for 3 h then rinsed twice with PBS. All explants were imaged with polarized reflectance microscopy system as described in Methods Section 2.4.

In the second experiment (exp.2), a small group of explants were scraped with 1, 3, 5, 10 and 20 passes (n = 3 per scrape number). These explants were then imaged with polarized reflectance in overlapping fields of view to make a strip across the explant. After that, low profile razor blades set 200 µm apart and parallel to each other were used to cut transversely into the explants. Transverse sections of 200 µm thickness were laid flat on the slide in PBS for imaging of the maximum intensity polarization signal in transmission mode (details in section 2.4) [46].

In the third experiment (exp.3), explants (n = 14) were also cut longitudinally in half using a scalpel and high-resolution (12 MP) digital photographs of explants were taken to measure total explant thickness. In the fourth experiment (exp. 4), n = 5 explants were also imaged with polarized reflectance, fixed overnight in neutral buffered formalin, stained for 1 hour with 3 µM DAPI, and imaged using an epifluorescence microscope (BX60, Olympus, Japan) with a 10x objective for the purpose of co-registration of polarized reflectance signals and the distribution of chondrocytes.

2.4 Image acquisition and analysis

The polarized reflectance imaging setup was the same as described in Huynh et al [44]. Briefly, a polarized light microscope (MT 9930, Meiji Techno, Ltd., Japan) was operated in reflectance mode with green interference filter (λ=546 nm) and linear polarizer placed in the incident illumination path, and the analyzer (a second linear polarizer in the camera path) in two configurations, with transmission axis parallel (Par) and perpendicular (Per) to that of the polarizer. Signal in the Per configuration was orientation-dependent, and largest with the apparent optical axis of birefringent material aligned at 45° with respect to polarizer and analyzer transmission axes. Cartilage specimens were aligned in this way for each field-of-view, leading to the brightest possible signal across the entire field-of-view in the Per channel. Images were captured with either 10X/0.25 or 4X/0.1 magnification/numerical aperture, strain-free objectives. The polarization contrast parameter, Pol, was calculated from images as Pol = (Par − Per)/(Par + Per), to form spatial parameter maps.

All polarized reflectance maps were filtered with a Gaussian filter (σ = 2 pixels) to smooth noise. Captured images were 1296 × 968 pixels, each pixel 0.59 µm × 0.59 µm on a side, smaller than the lateral resolution of the objective at 1.09 µm. This ensured that the Gaussian filter lowered shot noise in the image without obscuring meaningful features. In split-lines experiments, the split-lines long axis direction on OC core surfaces and the Pol map texture direction were measured using the angle tool in ImageJ 1.50b (NIH, Bethesda, MD). The orientation of Pol map texture was largely the same across a single field-of-view. The angle tool was used to take measurements in a five-dot domino pattern within each Pol map. The first line of the angle tool was drawn parallel to the perceived texture direction, over about 50 pixels, then the angle defined by creating an arm parallel to the image horizontal axis. Similarly, the angle tool was used to measure the orientation of split lines, with respect to the horizontal axis. For several maps, the Pol map texture direction was compared to orientation derived from the two-dimensional Fourier space ellipse and was found to yield very similar values (data not shown).

Maximum polarization signal intensity was obtained from the transverse sections using the MT9930 in transmission mode. We rationalized that obtaining a maximum, orientation-independent signal from transverse sections in transmission mode would reveal the superficial zone with less bias than if one linear polarization image was used. The maximum signal intensity was obtained from a pixel-wise least-squares fitting method using seven co-registered images of the section between crossed polarizers, rotated in 15° increments, as described by Eq. (3) and associated text in ref. [46]. Least squares fitting of the sinusoid function was performed in MATLAB. A typical average residual was 5.2, 7% of a typical average fitted signal of 78 on an 8-bit depth scale. The theoretical maximum signal from each pixel was placed in the maximum signal intensity map. In practice, this produced a uniformly intense, thick superficial zone with clearly delineated dark middle zone. It should be noted that the maximum intensity signal is robust and orientation-independent, but is not maximum birefringence, a parameter which requires calibrated measurements as described previously. Here, calibration was not performed as the goal was to estimate superficial zone thickness from a simple binary threshold of an intensity map. The enhanced features of the Leitz Ortholux II POL microscope base described in ref. [46], not present in our set-up (monochromators and extra optical filters) were not expected to significantly alter uncalibrated maximum signal intensity measurements.

Various texture parameters were extracted from Pol maps of unscraped and scraped explants. All texture parameters were calculated from polarized reflectance maps converted to 8-bit depth. A custom-made MATLAB R2018a (MathWorks, MA) script was developed to convert images into the two-dimensional spatial frequency Fourier domain, threshold the central region, and fit the binary image to an ellipse with the same second order image moment. A covariance matrix was calculated from the second order moments of the binary image, from which ellipse axes, area, and eccentricity were defined. The eccentricity of the fitted ellipse was used as a measure of texture feature alignment within the map. The gray-level co-occurrence matrix (GLCM) ImageJ plug-in, created by Julio E. Cabrera, was used to collect texture parameters considering distance between the pixel pairs from 1 to 160 pixels and average all four angles of 0°, 45°, 90°, 135° [47]. The GLCM algorithm calculates texture parameters such as angular second moment (asm), contrast, correlation, inverse difference moment (idm) and entropy as explained elsewhere [48,49]. Texture parameters were pooled over all explants and grouped using factor analysis in MATLAB.

For progressive scrape groups, co-registered regions of interest 200 µm long (and 200 µm wide for en face reflectance images) were created from the transverse and en face Pol images, for 12-13 regions per explant. The SZ thickness was measured at each pixel after a global threshold (>64 on an 8-bit scale) was applied to the maximum intensity polarization image. The total thickness was measured from high resolution digital photographs of the explant profile. The Pol parameter was correlated with SZ thickness for each explant.

2.5 Statistical analysis

Data were presented as mean ± SD. The correlation between split lines orientation and Pol texture were evaluated by Student’s t-test with 95% confidence intervals. The relationship between SZ thickness and Pol was assessed by linear correlation with significance set at p <0.05. The effect of total thickness on Pol was also determined by Student’s t-test with p < 0.05. Single and multiple linear regression of Pol on SZ thickness, total thickness and # of scrapes was performed in Systat 10 (Systat Software, San Jose, CA). Data and residuals were examined for distribution and normality before regression and associated t-tests.

Factor analysis was performed in MATLAB. Factor analysis models variation in correlated parameters as ${Z_{ij}} = \mathop \sum_{k = 1}^2 ({L_{jk}}{F_{ki}}) + {\epsilon _{ij}}$, where ${Z_{ij}}$ is the standardized value of the j^th parameter (here, of 9) of the i^th subject (here, of 40 explants, 20 scraped and 20 collagenase-treated), and k is the number of latent factors (here, 2). ${L_{jk}}$ are factor loadings, ${F_{ki}}$ are independent latent factors determined by the model, and ${\epsilon _{ij}}$ represents unobserved stochastic error. Unlike the input parameters, the factors are not correlated. The 9 input parameters were Pol mean, standard deviation, kurtosis, skew, contrast, inverse difference moment, entropy, correlation, and angular second moment differences from pre-treatment values from the same explants, calculated as post-treatment minus pre-treatment. The same location was imaged pre- and post-treatment by lining up the explant with dog-eared cut mark along the bottom edge of the eyepiece viewfinder using a 4x magnification objective, then switching to a 10x objective without moving the explant. The MATLAB factoran() function determines factor loadings using maximum likelihood estimation, returning a 9 × 2 matrix of factor loadings from a 40 × 9 matrix of image and texture parameters from Pol, standardized using the z-score. Another output was a 40 × 2 matrix of prediction scores, determined by weighted least-squares estimation. A final output is a 2 × 2 rotation matrix, R, used to transform the loadings, L, and predictions, P, as LR⁻¹ and PR′, multiplying by the inverse and transpose of the rotation matrix, respectively. The rotation allows a coordinate transformation that lends itself to simpler interpretation of the measured parameters with respect to the latent factors. For this work, the ‘promax’ oblique rotation method was used. Detailed discussions of the estimation of prediction scores and differences between factor and principal component analysis are beyond the scope of this work but exist in the literature [50–52]. Finally, prediction scores and loadings were normalized to present in the same biplot in Fig. 9B by dividing each value by the absolute value of the maximum score/loading in factors 1 and 2.

3. Results

3.1 Split lines orient with lattice-like polarized reflectance texture

Split lines correlated with Pol texture (Fig. 2). The angle of the longest split tended to be in the same general direction, oblique to the posterior-anterior and lateral-medial axes, as the dark lines in Pol maps, derived from birefringence (Fig. 2(A)). The direction coordinates in Fig. 2(A) were determined in reference to the osteochondral block location in situ (Fig. 2(B)). The direction of the birefringence texture strongly correlated with the major, long axis of split lines (Fig. 2(C), Δθ=9.2 ± 7.8°, mean ± SD, p = 0.86, Student's t-test).

3.2 The polarized reflectance parameter correlates with superficial zone thickness

Transverse sections revealed articular cartilage zonal architecture, including a 50-150 µm thick superficial zone (Fig. 3(A)) underlying explant surfaces with Pol texture in en face images (Fig. 3(B)). For example, in the unscraped control (Fig. 3(A)-(E)), SZ thickness varied from 57.0 ± 4.2 µm to 104.2 ± 3.2 µm for 200 µm regions, while corresponding Pol values ranged from 0.21 ± 0.004 to 0.44 ± 0.03 (mean ± SD). Pol mean values correlated inversely with SZ thickness (Fig. 3(E)) only in explants with no or minimal surface scrape. For explants with 0, 1, and 3 scrapes, the correlation of Pol with SZ thickness produced R² values of 0.66, 0.55 and 0.63, respectively (Table 1). For explants with 5 (Fig. 3(F)-(J)), 10, or 20 scrapes, Pol did not correlate significantly with SZ thickness, producing R² values of 0.26, 0.01, and 0.02, respectively (Table 1). With more scrapes, more cartilage was lost and less SZ remained, typically on the left and more prominent portions of the explant surfaces, most exposed to the scrape.

 

Fig. 3. SZ thickness and Pol correlation regarding to scrape level. Maximum intensity polarization maps derived from explant transverse sections were co-registered with their en face polarized reflectance at various number of scrapes. Correlation plot of SZ thickness and Pol average every 200 µm across explants were generated. (A,B) Representative intact explant’s transverse section and Pol map. (C,D) Larger magnification revealed smooth superficial surface and (E) high correlation between Pol and SZ thickness. (F,G) Representative 5-scraped explant transverse section and Pol map. Magnified sections displayed (H) ruptured or (I) lost superficial zone and (J) the relationship of Pol vs SZ thickness was weakened.

Download Full Size | PDF

Table 1. Pearson’s correlation and slope of Pol vs SZ thickness varied with scrapes

View Table

3.3 Polarized reflectance depends upon total explant thickness and is related to chondrocyte organization

Removal of the bottom half of explants produced higher Pol values, visible in two representative images of an explant before (Fig. 4(A)) and after (Fig. 4(B)) longitudinal bisection. The Pol mean values were significantly higher (Fig. 4(C, D), p < 0.001, Student’s t-test), by 36% (0.25 ± 0.06 for intact to 0.34 ± 0.06 after cut), as explant thickness was lower by 42% (1.53 ± 0.05 mm for intact to 0.88 ± 0.17 mm after cut).

 

Fig. 4. Effects of explant thickness on Pol values. Representative explant Pol map (A) before and (B) after bisection. (C) Mean Pol for n = 14 explants, and (D) bar graph of group averages (Student’s paired t-test, p < 0.001).

Download Full Size | PDF

Chondrocyte nuclei in the superficial zone organized in strings and clusters stained with DAPI and in some cases aligned with polarized reflectance texture in co-registered maps (Fig. 5). Some locations had more chondrocyte strings, which tended to co-localize with regions of higher Pol (Fig. 5(A)-(C)), whereas chondrocytes in other locations appeared more in clusters or unaligned groups, and did not appear to associate with Pol texture alignment, while still co-localizing with regions of higher Pol (Fig. 5(D-F)).

 

Fig. 5. Chondrocyte organization in the superficial zone. (A,D) Representative Pol and (B,E) co-registered DAPI epifluorescence micrographs; (C,F) overlay of DAPI on Pol contrast map. Chondrocytes were distributed either in strings (dashed yellow box) or pairs and groups (dashed red circle).

Download Full Size | PDF

For explants with 0-3 scrapes, the linear regression of Pol on SZ thickness and total thickness separately produced adjusted R² values of 0.31 and 0.15, respectively. Multiple linear regression of Pol on number of scrapes from 0-3 (K), SZ thickness (SZT) and total thickness, TT (Fig. 6(A)) produced an adjusted R² of 0.52, with slope coefficients of −0.025, −1.09 and −0.35/mm, respectively (Student’s t-test, p < 0.001). A residual plot (Fig. 6(B)) displayed a random distribution of estimated Pol values from multiple regression. A three-dimensional plot shows Pol dependence on SZ thickness and total thickness (Fig. 6(C)). Regression coefficients ± standard errors were: constant, 0.749 ± 0.08; K, −0.025 ± 0.008; SZT, −1.09 ± 0.17; and TT, −0.35 ± 0.07.

 

Fig. 6. Multiple linear regression of Pol on number of scrapes at 0, 1, and 3 (K); SZ thickness (SZT); and total thickness (TT). (A) Comparison between estimated Pol produced by multiple linear regression versus actual Pol, (B) residual plot estimated from regression, and (C,D) relationship of Pol with SZ thickness and total thickness, including the regression equation (red box).

Download Full Size | PDF

3.4 Scrape and collagenase treatments systematically alter Pol texture

Intact controls, collagenase-treated, and 10-scrape explants possessed distinct textures quantified by Pol contrast maps. The lattice-like textures with a single, leading orientation appearing in control explants (Fig. 7(A)) were partially lost in collagenase-treated explants (Fig. 7(E)) and completely lost in scraped explants (Fig. 7(I)), resulting in lower eccentricity of the ellipse bounding the central threshold region, from 0.84 ± 0.07 for intact control explants (Fig. 7(A-D), 0.65 ± 0.09 for collagenase-treated (Fig. 7E-H), and 0.55 ± 0.12 for scrape-treated explants (Fig. 7(I-L)).

 

Fig. 7. Image texture analysis using Fourier transforms (FTs). Representative analyses of (A-D) untreated, (E-H) collagenase-treated, and (I-L) scraped explants. (B,F,J) Thresholded FTs were used to calculate (C,G,K) ellipses with identical second-order image moments. (D,H,L) Distributions of eccentricity for control, collagenase-treated and scraped explants (n = 20 explants/group).

Download Full Size | PDF

Texture parameters from the GLCM were most different between untreated and surface-scraped explants at an optimal step size of 13 pixels for asm (10% difference), 32 pixels for contrast (18% difference), 25 pixels for correlation (24% difference), 14 pixels for idm (13% difference) and 18 pixels for entropy (8% difference) (Fig. 8(A)). Intact and scraped explants were well-separated in some parameters with optimized step size, such as contrast (Fig. 8(B)) while other parameters, like entropy (Fig. 8(C)), poorly separated the two groups. Texture parameters from collagenase-treated Pol maps had similar trends versus intact controls as from scrape-treated Pol maps (data not shown).

 

Fig. 8. GLCM texture parameters for control and scraped explants optimization. (A) representative of normalized difference between control & scraped explants (upper right) across all GLCM parameters and (B,C) each individual parameter.

Download Full Size | PDF

Predicted explant scores derived from GLCM parameters and pixel histogram statistics of mean, standard deviation, skew, kurtosis were projected into two latent factors (factor 1 and factor 2) that explained 72% of total variation and well-separated scrape and collagenase-treated explants with few exceptions (Fig. 9(A)). More latent factors did not explain enough additional variation to justify the more complicated model. It was noted that Pol mean, contrast and correlation covaried and were influenced by the second latent factor (y-axis) while SD, asm, idm, and entropy covaried and were affected by the first latent factor (x-axis). Skew and kurtosis were off-axis (Fig. 9(B)).

 

Fig. 9. (A) Predicted scores from factor analysis for scraped (gray circles) vs. collagenase-treated (black diamonds) explants based on 9 parameters derived from Pol maps. (B) The nine parameters (blue circles connected to the origin) and scores were normalized to plot them on the same axes, showing their relative contributions to factors 1 and 2. Normalized prediction scores from all explants are plotted as gray circles.

Download Full Size | PDF

4. Discussion

Polarized reflectance imaging provides information about articular surface damage and subsurface superficial zone collagen network in a single, en face image parameter Pol map. This statement is supported by alignment of Pol image texture with split lines, and correlation of mean Pol with SZ thickness from explants with mild (0-5 sandpaper passes) but not severe (>5) scrape damage. The Pol value depended to a ∼3-fold lesser extent on total explant thickness than SZ thickness, as determined from multiple linear regression analysis. Texture parameters extracted from the spatial Fourier transform and gray level co-occurrence matrix were sensitive to birefringence features and surface damage from scraped and collagenase-treated explants. Factor analysis grouped nine Pol map parameters into two latent factors that explain most variation in Pol map texture and average value. Taken together, these data demonstrate the origins of polarized reflectance map features as birefringence from the superficial collagen network, producing image map texture, and surface/subsurface bulk scattering, contributing a constant offset to Pol maps.

Polarized reflectance texture, assessed noninvasively, follows the direction of split lines, a common but destructive method used to visualize the major collagen fiber orientation of the cartilage surface. Split lines are thus impractical for clinical use as they could potentially weaken the articular cartilage [12]. Previous studies have shown split lines coincided with the maximum applied tension [25,33,53] and superficial collagen sustained largest tensile stress when the deforming forces run parallel to the direction of these fibers [54–56]. Therefore, polarized reflectance could serve as a convenient way to assess subsurface collagen network orientation when the surface mechanical anisotropy is important, such as during osteochondral graft placement. It is particularly relevant to note the high correlation of polarized reflectance texture with split lines did not extend to a clear relationship to the organization of chondrocytes in the superficial zone, suggesting that polarized reflectance depends more upon collagen microstructure than scattering from cells. Potentially, improved graft placement with proper alignment to adjacent host cartilage would reduce mismatched mechanical properties across the graft-host interface, though the benefits of this alignment need to be clearly demonstrated clinically. Osteochondral graft alignment led to increased transplant longevity in both animal and human studies [57–59], supporting a link between higher mechanical continuity and graft clinical outcomes. Polarized reflectance microscopy could provide a direct way to test whether alignment of the superficial zone collagen network with maximum shear stress reduces wear at the articular surface, since the technique is sensitive to both subsurface collagen microstructure and articular surface roughness. In vivo assessment of polarized reflectance may also prove useful to assess superficial microstructural alignment in tissue-engineered cartilage implants [60], or for screening healthy articular cartilage within joints at higher risk for post-traumatic osteoarthritis, during reconstructive surgeries, for example [61].

The dependence of the polarized reflectance parameter, Pol, on surface features and sub-surface birefringence indicates potential utility in detecting early alterations to articular cartilage in osteoarthritis. Such early alterations include articular surface roughening and a disorganized superficial collagen network [30–32]. Data from this study suggest those alterations should be detectable as higher Pol map values, with more severe damage leading to a loss of polarized reflectance texture. The average Pol map values in both intact and mild-scraped explants were highly correlated with SZ thickness, demonstrating the great sensitivity of the technique for SZ features. Photons from this layer are most likely to be singly-scattered, retaining birefringence information [62]. The clinical potential of polarized reflectance signals and the Pol parameter in orthopedics is high despite well-developed tomographic and three-dimensional imaging techniques, including OCT, laser scanning microscopy, x-ray computed tomography (CT), and magnetic resonance imaging (MRI) [63–67]. The non-invasive medical imaging techniques of CT and MRI have poorer resolution and more difficulty in resolving microscale articular surface and subsurface features that optical approaches, which require arthroscopy or excision of grafts and explants [65,68,69]. Polarization-sensitive OCT has excellent depth resolution in cartilage and the ability to resolve cartilage “brushing direction” [70], which may yield unique insight into cartilage layer mechanics. Still, widefield detection of polarized reflection signals may compete in convenience, utility, and adaptability to use with current rigid rod-lens arthroscopes. Polarized reflectance imaging offers a distinct combination of convenience, sensitivity to surface features and the superficial zone collagen network through the Pol parameter, and potential adaptability for use during arthroscopic procedures. The convenience could possibly be enhanced by using incident circular polarization (instead of linear), avoiding the necessity to align birefringent microstructure diagonal to the polarizer transmission axes, as has been demonstrated in transmission [46].

The determination of Pol map parameter dependence on two latent factors aids in the physical interpretation of polarized reflectance signal from articular cartilage. The Pol mean, contrast, and correlation parameters depended mainly on latent factor 2 from factor analysis and relate to a constant offset term in the Pol map. The Pol standard deviation, entropy, angular second moment, and inverse difference moment depended mainly on latent factor 1 from factor analysis and relate to texture in the Pol map. The Pol parameter relates superficial reflection, ${R_S}$ from reflection of deeply-penetrating light (in dermis, > 300 µm), ${R_D}$, estimated previously as $Pol = \; {R_S}/({{R_S} + {R_D}} )= \; ({Par - Per} )/({Par + Per} )\; $[45], and maps to the anisotropy factor g, which depends on nanoscale tissue structure [71]. Table 1 and Figs. 4 and 6 corroborate this model of Pol, developed in several previous studies [45,72,73]. A lower Pol with thicker SZ derives from a greater proportion of birefringence in ${R_S}$ (higher relative signal in Per), while scraping produces a lower proportion of birefringence and/or more depolarization in ${R_S}$, leading to higher Pol (and lower relative signal in Per) (Table 1). Leaving the articular surface intact but removing the deeper half of the cartilage explants raises Pol (by lowering ${R_D}$) (Fig. 4). Conversely, thicker explants contribute to lower Pol (Fig. 6). Hence, latent factor 2 may be linked to a constant offset term derived from altering ${R_S}$, $\; {R_D}$, or signal in the channels Par and Per. Latent factor 1, which covaries with most of the texture parameters, may be linked to the proportion of single scattering events which preserve polarization information and contribute to Pol mainly though ${R_S}$. Thus, the Pol parameter depends on scattering in both superficial and deeper layers of articular cartilage, but superficial light-tissue interactions predominate.

In conclusion, polarized reflectance microscopy is capable of producing a microstructural assessment of articular cartilage without labeling, with utility for in vitro studies using cartilage explants, evaluation of osteochondral grafts and tissue-engineered implants, and potential for use during arthroscopy. Polarized reflectance texture derives from collagen network microstructure in the superficial zone, while Pol map offset has several dependencies, including deep scattering. Polarized reflectance sensitivity to articular surface damage, total explant thickness, and superficial zone thickness indicates the potential to evaluate multiple microstructural features of cartilage relevant to osteoarthritis and cartilage repair. Future work should test the relationships determined in this study using adult bovine articular cartilage in immature bovine articular cartilage and cartilage from human donors. Biomedical applications of polarized reflectance microscopy would be extended if the relationship of optical indices derived from polarized reflectance to histological cartilage scores were determined, and the technique was adapted for arthroscopy. The relationships determined in this study aid in the physical interpretation of polarized reflectance signal from articular cartilage, in support of future clinical translation.