Use of Mueller matrix colposcopy in the characterization of cervical collagen anisotropy

Abstract

Annually, about 15 million preterm infants are born in the world. Of these, due to complications resulting from their premature birth, about 1 million would die before the age of five. Since the high incidence of preterm birth (PTB) is partially due to the lack of effective diagnostic modalities, methodologies are needed to determine risk of PTB. We propose a noninvasive tool based on polarized light imaging aimed at measuring the organization of collagen in the cervix. Cervical collagen has been shown to remodel with the approach of parturition. We used a full-field Mueller matrix polarimetric colposcope to assess and compare cervical collagen content and structure in nonpregnant and pregnant women in vivo. Local collagen directional azimuth was used and a total of eight cervices were imaged.

Keywords: Mueller matrix; anisotropy; birefringence; cervix; collagen; colposcopy.

Fig. 1

Preterm imaging system. VLCR, variable liquid crystal retarder; P, polarizer; L, lens; PBS, polarizing beam splitter; EP, eye piece; and LS, light source.

Fig. 2

The MMP image processing pipeline.

Fig. 3

In-vivo (a) nonpregnant human and (b) pregnant cervices raw image and MMP decomposed orientation. Circular color bar refers to the retarder orientation calculated from the Mueller matrix of the cervix.

Fig. 4

In-vivo nonpregnant and pregnant human cervices: (a) B/W CCD image with orientation lines: blue subsections > kurtosis = 0.6 > red subsections, (b) kurtosis, and (c) Mueller matrix decomposed orientation. KI = % of kurtosis values > 0.6 across the entire sample. Circular color bar refers to the retarder orientation calculated from the Mueller matrix of the cervix.

Fig. 5

In-vivo nonpregnant and pregnant human cervices. The arrows in the grayscale image indicate the subsections on the histogram. Blue subsections > kurtosis = 0.6 > red subsections. $X$-axis histograms use the subsections going from the left to the right. $Y$-axis histograms use the subsections going from the top to the bottom. Kurtosis of nonpregnant (KNP) sample; kurtosis of pregnant (KP) sample. KI = % of kurtosis values > 0.6 across the entire sample. There is a poorer collagen alignment in the pregnant cervix as compared to the nonpregnant cervix, as shown by a lower kurtosis and a broader distribution of angles.

Fig. 6

In-vivo nonpregnant and pregnant human cervices: (a) B/W CCD image with orientation lines: blue subsections > kurtosis = 0.6 > red subsections, (b) kurtosis, and (c) Mueller matrix decomposed orientation. KI = % of kurtosis values > 0.6 across the entire sample. Circular color bar refers to the retarder orientation calculated from the Mueller matrix of the cervix.

Fig. 7

In-vivo nonpregnant and pregnant human cervices. The arrows in the grayscale image indicate the subsections on the histogram. Blue subsections > kurtosis = 0.6 > red subsections. $X$-axis histograms use the subsections going from left to right. $Y$-axis histograms use the subsections going from top to bottom. KNP sample; KP sample. KI = % of kurtosis values > 0.6 across the entire sample. There is a poorer collagen alignment in the pregnant cervix compared to the nonpregnant cervix, as shown by a lower kurtosis and a broader distribution of angles.

Fig. 8

Kurtosis mean and standard deviation of nonpregnant (NP) and pregnant (P) cervices. One sided $T$-test between both groups showed than mean KNP cervices was significantly higher than that of pregnant cervices, with significance level of 95%.