1.

Introduction

Preterm birth (PTB) has many causes, but irrespective of its etiology¹, mechanical cervical failure or change in the cervix extracellular matrix (ECM) is a common endpoint. Fibrillar collagen is the major structural protein in the cervix that determines its load-bearing capabilities. With progression of pregnancy, the cervix undergoes changes in the collagen structure and corresponding mechanical strength. Structural defects in the cervix result in PTB as exemplified in women with cervical insufficiency. Several human and animal studies have suggested that atypical changes in the ECM of the cervix precede PTB.^2–6 Thus, the development of diagnostic modalities that could identify premature abnormal cervical remodeling holds great potential as a tool for early and accurate assessment of cervical disease. Numerous imaging technologies are under exploration though none has yet yielded established clinical tools.^7,8 Research groups have utilized second harmonic generation (SHG), a nonlinear microscopy modality, to obtain exquisite images of cervix collagen fibers (alignment, porosity, and fiber size) in the cervix of nonpregnant and pregnant women and mice, and have shown that alteration in collagen structure and assembly corresponds to the biomechanical properties of the cervix during pregnancy.^8–11 The use of SHG in the clinical setting is yet hampered by the instrumentation’s low field of view ($<500\mu \mathrm{m}$) and high susceptibility to movement artifacts. At this stage, SHG systems are also expensive because of the use of femtosecond Ti:sapphire lasers.

1.1.

Cervix

The cervix is an extension of the lower part of the uterus comprising a portion of the female reproductive system. It is cylinder shaped, $\sim 3$- to 4-cm long and 3 cm in diameter, and has a central canal through its entirety. A cartoon representation of the cervix is shown in Fig. 1. The cervical canal serves as the entrance, via the vagina, for sperm in reproduction and as an exit canal, via the uterus, for childbirth. The two ends of the canal are termed internal orifice (OS) connecting to the uterus and external OS at the vagina. The cervix is divided into two regions because of their cellular differences: the ectocervix and the endocervix. The ectocervix, the lower area, is visible through the vaginal opening and is comprised of a stratified squamous epithelium, having several cell layers of differing morphological characteristics.¹² The thickness of the epithelial layer is between 200 and $500\mu \mathrm{m}$.^13–15 The dramatic structural changes undergone by the cervix during pregnancy have been categorized by four main phases: cervical softening, ripening, dilation, and repair (postpartum).^6,16,17

Fig. 1

The cervix.

The physiological and mechanical properties of the cervix are altered by a cascade of microenvironmental events in each of these phases, which are not completely understood. Cervical softening begins within 1 month of the initiation of pregnancy and the cervix undergoes increased vascularity and edema.⁶ Softening is a longer phase than the others, progressing through the 33rd week of pregnancy. During this phase, the collagen of the stroma becomes less organized and the cervix becomes pliable and begins shortening.¹⁷ The ripening phase begins after softening, in this phase the cervix goes from a load-bearing function to a birthing canal.¹⁸ By the end of cervical ripening, when dilation begins, a 30% to 70% decrease of collagen^16,19,20 occurs since the onset of pregnancy. Others have shown an increased collagen solubility in weak acids rather than change in collagen content,^21,22 and an increase in water (5% to 10%)^18,23,24 with a shift from insoluble to soluble collagen^19,25 with as much as 90% soluble collagen by the third trimester.¹⁸ Dilation follows ripening and involves an influx of leukocytes,¹⁸ similar to an inflammatory response, which may serve to cause an increase in the matrix metalloproteinase, collagenase.^6,26,27 Collagenase causes the breakdown of collagen cross links and allows the cervix to weaken^21,28,29 and open, thereby radically changing shape by shortening and effacing in preparation for delivery of the baby.

1.2.

Collagen in the Cervix

An ECM lies under the epithelium of the cervix, separated by the thin basal lamina layer of collagen type IV fibers,¹³ and consists mainly of collagen, $\sim 10\%$ to 15% smooth muscle cells^18,30,31 and a ground mixture of biomolecules.

Two types of collagen comprise the cervical ECM; $\sim 70\%$ is collagen type I and 30% is collagen type III.

Numerous researchers have studied the collagen of the uterus to determine how this structure maintains its integrity during pregnancy.^{6,10,28,32–36} Aspden found the structure of collagen is oriented in three unique areas surrounding the cervical canal, the anisotropic alignment of the collagen differing within each area. The cervical fibrils are aligned both around and along the canal for increased strength as shown in Fig. 2.^21,37–39

Fig. 2

Alignment of collagen fibers in human cervix in three anisotropic zones.

In previous work, Gan et al.⁴⁰ utilized unpolarized spectral domain optical coherence tomography (OCT) to determine the distribution of collagen in the cervix. Their algorithm uses preprocessing of images to first promote edge detection of fibers in different depth planes of the OCT acquired three-dimensional stack before determining the orientation of subsections in the planes. This is done over selected regions in the sample. The orientation information across different regions can then be recomposed to create a complete orientation map of the entire region that was imaged.

We proposed the use of an optical modality, Mueller matrix polarimetry (MMP), to target the fibrous ultrastructure organization of the cervix. This paper describes how MMP can be used to ascertain collagen arrangement in the cervix and will validate its findings with OCT.

2.

Materials and Methods

A combined OCT–MMP system introduced in previous work⁴¹ was modified to a fiber-based OCT system shown in Fig. 3. The system is based on a Michelson interferometer. The laser light source is a broadband superluminescent diode (Bayspec, San Jose, California) with 840-nm central wavelength and 50-nm full width at the half-maximum bandwidth. A telecentric scan lens (LSM03-BB, Thorlabs, Newton, New Jersey) focuses the light onto the sample and collects the backscattered light. Galvanometer-mounted mirrors (GVS012, Thorlabs, Newton, New Jersey) on the sample arm enable transverse beam scanning on the sample. A custom-made spectrometer detects and measures the interference signal between reference arm and sample. The spectrometer consists of a collimating lens with a focal length of 75 mm (Edmund Optics, Barrington, New Jersey), an $1800\text{lines}/\mathrm{mm}$ volume holography transmission grating (Wasatch Photonics, Logan, Utah), an assembly of triple lenses with an effective focal length of 150 mm, and a line array CCD camera (spL4096-140k, Basler, Highland, Illinois). Data processing algorithm, control and display software are developed using MATLAB (Mathworks, Natick, Massachusetts). The images produced by the system have a theoretical resolution in depth of $5\mu \mathrm{m}/\text{pixel}$, and an axial resolution of $20\mu \mathrm{m}/\text{pixel}$.

Fig. 3

Combined OCT and Mueller matrix system schematic (red light, OCT; green dashed, MMP).

The coregistered Mueller matrix system shown in Fig. 3 is traced by the dashed green line and consists of a CCD camera (Lu175, Lumenera, Ottawa, Ontario) with a Computar MLH-10× 1/2 in 13 to 130 mm f5.6 10× macrozoom lens. The camera was secured above the sample objective to allow focusing on the height adjustable stage. A linear polarizer (Prinz, Northbrook, Illinois) and two liquid crystal retarders (LCR) (Meadowlark Optics, Frederick, Colorado) between the lens and sample objective form the polarization state analyzer of the polarimeter. Employing the same sample objective for the two imaging systems ensures they are imaging the same region of interest. The illumination port in Fig. 3 indicates the light source for the MMP. A 530 nm LED (M530L, Thorlabs, Newton, New Jersey) was oriented to illuminate the sample at an incident angle of 45 deg and collimated with a 30-mm diameter tube and a 75-mm focal length plano-convex lens (Newport, Irvine, California). Two LCRs were mounted after the light source and a linear polarizer using a cage system to form the polarization state generator. The calibration of the MMP system follows a standard methodology utilized by our group in several applications^42,43 and resulted in a condition number of 2. A Mueller matrix of air was also acquired and showed an error well below 1%.

2.2.

Optical Coherence Tomography Image Processing

We aim to mimic Gan et al.’s approach to measure the orientation of birefringent tissue using OCT, and to compare it with MMP, which should generate orientation information inherently summed across all depths in the sample for the entire field of view. Our approach preprocessed and averaged the en face data in a similar manner suggested by Gan et al. The determination of orientation, however, was calculated using our own method, which applies the Radon transform and calculation of the projection angle SNR as shown in Fig. 4. The Radon transform projects the sum of image intensities along the different orientation angles inputted into the transformation.⁵⁵

Fig. 4

OCT processing pipeline.

Fig. 5

Radon transform of section of silicon phantom.

Five $X–Y$ images (c-scans) were first averaged in the $Z$-direction to improve the orientation measurements across a depth of $40\mu \mathrm{m}$. The c-scans were then smoothed with a $3\times 3$ median filter to remove speckle noise. The resultant matrix was then convolved with a $3\times 3$ Sobel filter to generate an image with emphasized structural edges. This matrix was sectioned into $20\times 20\text{pixel}$ areas and Radon transformed. The SNR of all projection angles was calculated for the Radon transformed section using Eq. (3) to produce a graph, such as in Fig. 5. The SNR is defined here as the quotient of the standard deviation of a projected angle with the standard deviation of the entire image section. The peak projection angle SNR is denoted by a triangle and selected to be overlaid over the section of the enface image of the sample as a line. This peak corresponds to the orientation of the sample as it is calculated for sections throughout an enface image of the sample.⁵⁵

Eq. (3)

$$\mathrm{SNR}=\frac{{\sigma }_R(\theta )}{{\sigma }_1}.$$

The orientations measured by both imaging techniques were superimposed over the images they were calculated from as oriented lines. OCT was used as a standard for testing the effectiveness of MMP in this application.

3.

Results and Discussion

All images shown are of the $Y-X$ plane of the cervix denoted in Fig. 2. Orientation of 0 deg is parallel to the $X$ axis and a positive $\mathrm{\Delta }\theta$ is considered counterclockwise from horizontal. The silicon phantom was measured first. Its weak attenuation can be seen in the low depolarization value of the phantom seen in Fig. 6(b), the uniform value of retardation in Fig. 6(c). The dashed line box in Fig. 6(a) is the region where orientation was compared between OCT and MMP in Fig. 7.

Fig. 6

Silicon phantom Mueller matrix decomposition: (a) CCD image, (b) Mueller matrix decomposed depolarization, (c) Mueller matrix decomposed retardation, and (d) Mueller matrix decomposed orientation.

Fig. 7

Silicon phantom OCT–MMP orientation comparison: (a) OCT c-scan, (c) Mueller matrix decomposed orientation, and (b/d) orientation histogram circular statistics ($k=\text{kurtosis}$, $\mu =\text{mean angle}$).

The circular statistics shown in Figs. 7(b) and 7(d) was calculated from the regions encased by the dashed line squares in Figs. 7(a) and 7(c). It is important to note that unlike the MMP images that are taken at the surface of the sample, the OCT images are taken below the surface of the sample as they are c-scans averaged in depth. This is why striations can be seen by both imaging modalities in the silicon phantom which was completely uniform in structure throughout the sample but not in the tissue samples. Both imaging modalities had kurtosis values in the 90-percentile range indicating a tight distribution of angles around a mean angle of 68 deg and 65 deg for the OCT and MMP, respectively. After the orientation measurements were confirmed to properly change as the phantom was rotated on the sample stage, we moved to tendon as a biological sample which would attenuate light due to scattering and absorbing more than the silicon phantom but still have high anisotropy which would be seen in the orientation measurements.

The greater attenuation of the tendon is evidenced by the much greater depolarization value in Fig. 8(b) as compared to the silicon phantom. Despite its high anisotropy, tendon does not exhibit retardation as uniform as the silicon phantom. The changes in morphology and roughness of the tendon’s superficial layers compared to the silicon phantom can be seen as deviations in the orientations calculated by OCT and MMP in Figs. 9(a) and 9(c).

Fig. 8

Tendon Mueller matrix decomposition: (a) CCD image, (b) Mueller matrix decomposed depolarization, (c) Mueller matrix decomposed retardation, and (d) Mueller matrix decomposed orientation.

Fig. 9

Tendon OCT-MMP orientation comparison: (a) OCT c-scan, (c) Mueller matrix decomposed orientation, and (b/d) orientation histogram circular statistics ($k=\text{kurtosis}$, $\mu =\text{mean angle}$).

Fig. 10

Paraffin embedded cervix 1 Mueller matrix decomposition: (a) CCD image, (b) Mueller matrix decomposed depolarization, (c) Mueller matrix decomposed retardation, and (d) Mueller matrix decomposed orientation.

A breakdown in the Mueller matrix orientation is highlighted by a phase wrapping of orientation angle from $+70\mathrm{deg}$ to $-70\mathrm{deg}$. Regardless, the tendon produced high kurtosis values with both imaging modalities around a mean angle of $-74\mathrm{deg}$ in the selected area.

Once both imaging techniques had been tested on the anisotropic test samples, we imaged the paraffin embedded cervices to see whether the circumferential alignment of cervical collagen, which is a birefringent material, could be resolved by the techniques. The inner and outer regions of the cervix contain collagen oriented in the $z$-direction, which cannot be resolved with polarimetry of the sample’s surface. We are interested in the region between the OS and the outer edges of the cervix because it contains collagen circumferentially aligned around the OS and can be resolved noninvasively with polarimetry of the surface. Two different cervices are shown using OCT in Figs. 11(a) and 13(a).⁴⁰ Similarly, the orientations calculated by MMP also rotate around the OS of the cervices shown in Figs. 11(c) and 13(c). The depolarization and retardance images in Figs. 10 and 12 do not show any drastic changes as is expected in healthy tissue. Large changes in depolarization and retardance are known to occur in cancerous tissues.⁴⁴

Fig. 11

Paraffin embedded cervix 1 OCT-MMP orientation comparison: (a) OCT c-scan, (c) Mueller matrix decomposed orientation, and (b/d) orientation histogram circular statistics ($k=\text{kurtosis}$, $\mu =\text{mean angle}$).

The dashed line box in Fig. 12(a) designates the region that was used to measure the orientation shown in Fig. 13 for both imaging modalities. Circular statistics was applied to the OCT orientation and Mueller matrix decomposed orientation in the same subsections. The selected subsections of the cervices are denoted by the dashed line square boxes in Figs. 11 and 13.

Fig. 12

Paraffin embedded cervix 2 Mueller matrix decomposition: (a) CCD image, (b) Mueller matrix decomposed depolarization, (c) Mueller matrix decomposed retardation, and (d) Mueller matrix decomposed orientation.

Fig. 13

Paraffin embedded cervix 2 OCT-MMP orientation comparison: (a) OCT c-scan, (c) Mueller matrix decomposed orientation, and (b/d) orientation histogram circular statistics ($k=\text{kurtosis}$, $\mu =\text{mean angle}$).

The kurtosis values calculated from the selected cervix region are between 0.74 and 0.78 in comparison to the more ideal values of 0.92 and 0.99 in the nondepolarizing silicon phantom. This is expected due to the strong depolarization effect of biological tissue^41,56,57 because of its many constituents and the more complex arrangement of collagen in the cervix as opposed to tendon. Another cervix is shown in Fig. 12; as this sample was large, only a section was imaged with OCT, whose field of view is about $\sim 5\mathrm{mm}$ while the Mueller Matrix system can measure the full cervix. The dashed line boxes in Figs. 13(a) and 13(c) are the regions selected for the circular statistics calculated in Figs. 13(b) and 13(d), respectively.

A summary of the circumferential fiber alignment of collagen in the middle of the cervix samples is shown in Fig. 14. Eight regions around the OS of the cervix were selected from both the Mueller matrix and the OCT data. The mean orientation and percentage error between the two modalities are plotted for comparison. The mean orientation rotates fully around the cervix and flips sign at the vertical axis (north position). The mean errors between the modalities are 3.26% and 3.66%, respectively, for the two samples shown. Tissue fixation with paraformaldehyde and paraffin embedding has been shown to cause small increases in depolarization and retardation^58–60; however, in vivo imaging currently underway which is not shown here, shows very similar results of collagen orientation, hence, we believe that fixation does not alter the ultimate result of this study. We propose that in vivo measurements of collagen orientation could be used to identify these areas of disruption and lead to diagnosis of abnormal or early collagen disruption in the pregnant cervix.

Fig. 14

Summary statistics of cervix orientation between MMP and OCT.

4.

Conclusion

Orientation of the optical axis of birefringent material is important in the diagnosis of abnormal conditions in tissues with large amounts of ECM. Tissues that rely on collagen for mechanical strength are often highly anisotropic.⁶¹ The circumferential orientation of collagen in the cervix is important in maintaining the fetus in the uterus by not allowing passage through the OS and can be measured quickly with one set of images using MMP. To our knowledge, this is the first report that focuses uniquely on polarization-derived collagen orientation in the mammal cervix. Gan et al.⁴⁰ as well as Zhang et al.⁶² have shown that cervical collagen alignment can be measured using OCT and SHG microscopy, respectively. Yet, MMP is more adaptable to cervical imaging studies than those modalities due to its ability to capture the entire cervix at a distance and from outside the vagina. Current modalities for diagnosing preterm labor involve direct contact with the cervix during a pelvic ultrasound⁶³ or fibronectin discharge collection.⁶⁴

Depolarization and retardation were also decomposed from the Mueller matrices and showed little variation throughout the different cervical samples. The orientation parameter obtained with MMP was in good agreement with measurements made using OCT. Further studies of collagen orientation in cervices under different conditions are needed to understand if MMP can effectively measure the changes in collagen orientation that should occur when the normal ECM alignment is disrupted by pregnancy or disease. An in vivo study of human cervix is currently underway using a colposcope outfitted with an MM polarimeter of the same design as reported here and will be useful in establishing the diagnostic power of this technique in the determination of risk for preterm labor.