Polarization-sensitive optical coherence tomography with a conical beam scan for the investigation of birefringence and collagen alignment in the human cervix

By measuring the phase retardance of a cervical extracellular matrix, our in-house polarization-sensitive optical coherence tomography (PS-OCT) was shown to be capable of (1) mapping the distribution of collagen fibers in the non-gravid cervix, (2) accurately determining birefringence, and (3) measuring the distinctive depolarization of the cervical tissue. A conical beam scan strategy was also employed to explore the 3D orientation of the collagen fibers in the cervix by interrogating the samples with an incident light at 45° and successive azimuthal rotations of 0-360°. Our results confirmed previous observations by X-ray diffraction, suggesting that in the non-gravid human cervix collagen fibers adjacent to the endocervical canal and in the outermost areas tend to arrange in a longitudinal fashion whereas in the middle area they are oriented circumferentially. PS-OCT can assess the microstructure of the human cervical collagen in vitro and holds the potential to help us better understand cervical remodeling prior to birth pending the development of an in vivo probe.


Introduction
Preterm birth (PTB), which is defined as birth before 37 weeks of gestation, is the leading cause of neonatal morbidity and mortality not attributable to congenital malformations worldwide. It accounts for more than 1 million deaths a year [1]. Across the world, more than 15 million births are preterm every year, with prevalence rates that range from 5% to 18% [2][3][4]. In the UK, around 8% of babies are born prematurely whereas in the US approximately 12% of live births occur before term [5,6]. Despite advances in perinatal health, the incidence of PTB has continued to increase. Given the multifactorial etiology of PTB, diagnosis and prevention have proven difficult. However regardless of what triggers PTB, there seems to be common gradual changes in the stroma of the cervix. The cervix, which plays an essential part in maintaining a pregnancy to term, has to remain closed throughout gestation so that the fetus can develop in utero [7]. However, for birth to occur, it has to shorten, soften and dilate. This crucial remodeling process is required for uterine contractions to lead to delivery [8,9]. Since PTB requires premature cervical remodeling, improved understanding of this process is essential for the development of more accurate screening tools for PTB [10]. Such tools may also facilitate better targeted clinical interventions. Cervical remodeling begins several weeks/months before parturition, but its exact timing and processes have not yet been fully characterized in humans, most evidence stemming from studies on rodents [11]. Experiments conducted on rat and human cervical biopsy tissues in the late 1980s using X-ray diffraction showed that collagen fibrils exhibited preferential orientation in the non-pregnant cervix: around the endocervical canal and in the outermost area, collagen fibers were mostly arranged longitudinally, whereas in the middle area, fibers were predominantly circumferential [12]. This orientation pattern is thought to be lost during pregnancy. Current evidence also suggests that cervical remodeling involves a change in the orientation, morphology and assembly rather than in collagen amount. However, current in vivo assessment of the remodeling of the cervix in women is confined to cervical length ultrasound measurement and digital examination approaches incapable of assessing the key molecular changes associated with extracellular matrix remodeling [9].
Several research imaging techniques have been employed to investigate cervical collagen microstructure, including X-ray diffraction, second harmonic generation (SHG) microscopy, magnetic resonance diffusion tensor imaging (MR DTI) and optical coherence tomography (OCT) [13][14][15][16][17]. However, none of these modalities has been successfully translated into the clinical setting due to inherent limitations to the technique. MR DTI, for example, is too slow for real-time processing; SHG holds limited imaging speed and does not perform well endoscopically, and OCT lacks accuracy to assess the collagen structure. An emerging technique, Full-field Mueller colposcopy, has also been developed for investigation of cervical microstructure [18][19][20][21]. This technique has regarded as a potential alternative to the current screening methods, e.g. histological diagnoses, due to the advantages of low cost, rapid imaging with wide field images and ready endoscope [19]. However, the technique of Full-field Mueller colposcopy is not in the mainstream clinic yet, and cannot provide depthresolved changes in tissue's phase retardance, birefringence and relative fast axis orientation nor the thickness of the overlying epithelium.
Polarization-sensitive OCT (PS-OCT) is a functional extension of OCT, which has the potential to be an appropriate tool for investigation of cervix or cervical remodeling in clinical studies. This is because PS-OCT not only shares the advantages of OCT, including high resolution (4-20 μm), high-speed 3-D imaging and easy integration with a catheter or a handheld probe, but it offers additional information such as the polarization state of backscattered optical light [22,23]. The polarization state can be used to measure tissue's depth-resolved phase retardance, birefringence and relative fast axis orientation, which allows PS-OCT to differentiate anisotropic tissues, such as collagen fiber, muscle and tendon, from other structures [24]. In 2008, Lee et al demonstrated that PS-OCT could detect cervical intraepithelial cancer (CIN) on human cervical biopsies with a sensitivity of 94.7% and a specificity of 71.2% when results were correlated with histology [25]. However, little is known about the ability of PS-OCT to assess changes in the orientation of cervical collagen [26].
In this study, we sought to assess whether PS-OCT was capable of detecting changes in the alignment of cervical collagen fibers in vitro. Cervical cross-sections obtained from uterine specimens of patients undergoing hysterectomy for benign gynaecological conditions were fully scanned with PS-OCT. Additionally, the three-dimension (3D) orientation of collagen in the samples was assessed using a conical beam scan protocol, originally developed for studying collagen alignment in articular cartilage [27].

Configuration of PS-OCT
Our in-house PS-OCT for this study was developed based on the method reported by Al-Qaisi et al [28]. This system and its characteristics have already been described in our previous paper [27]. Here, we provide concise summaries of the PS-OCT configuration and its principle. The schematic diagram of PS-OCT is shown in Fig. 1. The light source of the system was a wavelength-swept laser (HSL-2000-10-MDL, Santec, Japan) with a center wavelength of 1315 nm, a full width at half maximum of 128 nm, a wavelength scanning rate

PS-OCT image processing
Our image processing was carried out in MATLAB. The retardance ( ( )) where z is the physical depth into sample and 0 λ is the center wavelength of the light source. The true birefringence, defined as e o n n − , can be expressed as following relationship [30]: Here, e n is the extraordinary refractive index and C θ is angle between the direction of light propagation and the optic axis of the fiber, i.e. the c-axis in optical terminology. For obtaining more precise values of the apparent birefringence, the phase retardances of 10000 A-scans (arranged in an XY grid of size 100 by 100) at each depth within the sample were averaged, and plotted as a function of depth to get the retardance slope of the birefringent tissue, e.g.
. The slope was obtained by linear fitting method, and the birefringence values were calculated as illustrated in the Eq. (1). To visualize apparent birefringence, a 2D birefringence image was mapped out by the derivative of the retardance versus axial image depth after the retardance B-scan was smoothed using a 50 by 50 median filter.
The depolarization of tissue was quantified on the basis of the theory developed by Götzinger et al [31]. The degree of polarization uniformity (DOPU), which can be regarded as a spatially averaged degree of polarization (DOP), was quantified for evaluation of tissue depolarization. DOPU was processed as follows. Firstly, a thresholding procedure was applied to the intensity data ( ), I i.e.
where I, Q, U, and V are Stokes vector elements, and then the Strokes vector elements were averaged by a 2D mean filter (a size of 15 by 6 pixels). Finally, the DOPU was processed in term of 2

H&E histology
Three cervical specimens, previously scanned with PS-OCT were then fixed with 3.7% formaldehyde and stained with a modified H&E technique in order to better visualize collagen fibers. Histological slides were subsequently assessed with an optical microscope (10X, LEICA DM750).

Statistical analysis
Collagen birefringence in the center, middle and edge areas was compared using ANOVA with Bonferroni correction. Collagen birefringence was also correlated with age, parity, mode of delivery, menopausal status and indication for surgery using Pearson's correlation coefficient, ANOVA when the assumption of equality of variances was met and nonparametric Kruskall-Wallis when the Levene's test was significant.

Intensity, retardance and birefringence images
Whole in vitro cervical cross-sections were scanned with our in-house PS-OCT. We have included, as an example, intensity, retardance and birefringence images obtained from the middle region of one of the samples analyzed, and shown how they were computed into precise numerical values (Fig. 5). The intensity and retardance images were acquired using LABVIEW control software, shown in Figs. 5(a) and 5(b) respectively. In Fig. 5(a), the intensity image resulting from the difference of refractive index between various layers of tissue displays two discernible tissue layers. The superficial layer which presents comparatively lower intensity is assumed to be the cervical epithelium, and the deeper layer, the collagen content of the stroma. This assumption also enables to explain the retardance image, in which the cervical epithelium can be discerned as a blue band on the top due to its lack of birefringence. The thickness of the cervical epithelium can therefore be calculated immediately which could be of potential benefit in evaluating disorders such as cervical cancer [32] and the acquired human immunodeficiency syndrome [33]. Just underneath the epithelium layer, a significant increase of retardance is observed resulting from the birefringence of aligned collagen fibers. Compared with traditional modality, e.g. confocal fluorescence microscopy, which might cannot measure the epithelial thickness because the maximum imaging depth (<33 μm) is insufficient to cover all epithelial thickness [32], PS-OCT has much larger imaging depth (~800 μm) to readily measure the thickness of epithelium. In our experimental results, the retardance image has the advantage to differentiate the epithelial and collagen layers than intensity image, because a part of intensity images is featureless.
For generating a birefringence image, the gradient of retardance as a function of physical depth is computed after smoothing the retardance image using a median filter. The birefringence image is mapped out by the gradient of retardance, from which the collagen distribution can be inferred, shown in Fig. 5(c). The precise value of apparent birefringence of collagen is evaluated with a linear fitting method, shown in Fig. 5(d). In Fig. 5(d), the retardance at each particular depth within sample is laterally averaged to reduce speckle noise and plotted as a function of depth. The slope of collagen retardance, namely ( ) , S d z dz δ is calculated by linear regression. The precise value of birefringence can be directly calculated from Eq. (1). In our example, the slope of the regression equation in Fig. 5(d) is 2.48 rad/mm, and the value of collagen birefringence is 3 3 2.48 2 1.315 10 0.52 10 .
Since the retardance increases linearly with depth within sample, it is expected that the birefringence of collagen stays constant with depth. Background noise is gradually dominated at the deeper depth, masking the linear increase of retardance.

Orientati
The cervical which depend true birefring orientation, an apparent bire According to light propaga travels along finding the im within the hum circumferenti normal PS-OC surface, it is e the middle are been confirme one of the sam center, middle notably as a f the collagen b center and ed between the d center, middl OCT images i  Fig. 3 rection is perp ) n Δ will reach will be close to Fig. 6. A set of rdance and bire area, retardanc e center and ed and much mo ges of Fig. 6 Fig. 7(a) (2). This the center tween the g conical e roughly equal in theory. To explain this discrepancy, we hypothesized that the c-axis of the collagen fiber is actually oriented at a polar angle tilted away from the normal axis. The polar and azimuthal angles of collagen fiber can be estimated by comparing simulated results of retardance images and real results in polar format. The estimation process and results are shown in Figs. 7(b), 7(c), and 7(d). 360 a-scans of phase retardance as a function of azimuthal angle are displayed in the traditional OCT format in Fig. 7(b), where each A-scan is fetched from each B-scan one by one over the azimuthal angle from 1° to 360°. These A-scans are extracted by a semi-automatic program to ensure every extracted A-scan corresponds to the center point of the plate rotation or close to the center point. This program runs on the assumption that the only center point of rotation has a constant altitude of sample surface in each B-scan due to the curved cervical surface. Therefore, the program is designed to find the 360 A-scans which have small variation of the surface altitude in each B-scan. The 360 Ascans are then converted to polar format, shown in Fig. 7(c), where the circle center and radius are sample surface and depth respectively. A simulated patterning of phase retardance in polar format is generated by a layered model based on the extended Jones Matrix calculus (EJMC) previously developed by our group [36], shown in Fig. 7(d).
The general process of EJMC is introduced briefly here. In the EJMC model, the sample of biological tissue is treated as a multi-layered structure, and each layer is considered as a linear retarder with a constant fast axis orientation. In this case, the signal-pass Jones matrix of sample (P) is the product of Jones matrices of individual layer, which can be expressed as: Here, ( ) can be found in previous papers [37,38]. In brief, the Extended Jones Matrix Calculus of Gu and Yeh [38] is used to calculate ( ), where R T and ' R T are the Fresnel reflection coefficients at the interface between air and sample surface. The light beam of PS-OCT (e.g. circularly polarized light) passing through individual layers of sample and then reflected back onto the detector can be modelled as: Here, QWP denotes the Jones Matrix of the quarter wave plate in PS-OCT system. Consequently, the depth dependent retardance ( ( )) S z δ of sample detected by PS-OCT can be calculated as: The parameters of EJMC, including the ordinary refractive index, true birefringence and polar and azimuthal angles of collagen over the depth of the sample, can be set to find a simulated image which matches the pattern of the ization image scrambling or n, can also b S-OCT is able t arameter: DOP Section 2.4, the Fig. 9. At thi ntrast between in Fig. 5(a). n the range of indicates the c tissue prefers t ixed polar ang 3 10 .
− × We con mferential fashi a previous SHG arrangement i hree cervical s When the histo lel aligned in llagen bundles histological sl organized whe nly arranged in r 80°, have bee nfirms our conc des of three cervic acquired from the r loss of line e used in bio to measure the PU [31]. Follow e DOPU of hu is particular ti n epithelium a In Fig. 9(c), t 0-1, and displ correlation bet to preserve pol gle of 80° over cluded that th ion, which has G study [17] The polari and density o sample birefri [39][40][41][42]. Seve proportional t tissue normal Consistently, higher level preclinical ap and identify c results. DOPU natural pigme another poten

Benefits
In summary, P resolved chan depolarization information, s modalities fin measure the e depth-wise a high degree of alue beyond th yer can be seen on the intens n. It is shown t layer which h mage ( Fig. 9(b) Fig. 10(a), th n the center an ween the middl boundary of ould be vague . a) and b) are the heir corresponding isplay the physical by a median filter ed out due to poor the participant was seen only dge area the cor middle region, g. 11, which su th our previous ction 3. 1  No other statistically significant differences were found between birefringence and (1) parity, (2) previous mode of delivery, (3) BMI, (4) indication for surgery, or (5) type of surgery. However, as our study was exploratory in nature and not particularly powered for any of these outcomes, further research is needed before a firm conclusion can be reached between optical and physiological variation in the non-gravid human cervix.

Conclusion
To the best of our knowledge, this is the first study to ever report on the phase retardance, the birefringence, the orientation of c-axis and depolarization of collagen fibers in human nongravid cervix using PS-OCT. We have been able to show some of the unique advantages of using PS-OCT to study the cervix, including its ability to (1) easily identify the cervical epithelium and measure epithelial thickness, (2) rapidly image the distribution of cervical collagen, (3) accurately determine birefringence, (4) estimate the 3D alignment of collagen fibers and (5) measure the distinctive depolarization of the cervical tissue. After interrogating 20 cervical cross-sections from non-gravid women using PS-OCT, we found a significant higher birefringence in the middle area compared with the center and edge regions (p< 0.05). As previously seen in studies with SHG, we also identified a significant increase in the apparent birefringence of the middle area with age which could respond to a physiological remodelling in the cervical collagen fibers as the reproductive function of the cervix diminishes. All in all, we have shown that PS-OCT is capable of assessing the arrangement of cervical collagen objectively and accurately, thus holding promise as a potential tool to better understand cervical remodeling prior to birth. This in turn could lead to earlier identification, more timely prevention and better stratification of management of PTB pending the development of a hand-held probe.

Funding
Engineering and Physical Sciences Research Council (EPSRC) (EP/F020422); Sheffield NHS Teaching Hospitals (the Jessop Wing Small Grant Scheme); scholarship of University of Sheffield.