Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Womens Health

Women's Health - Poster Session A

Poster L19 - Accurate segmentation and measurement of collagen layers in volumetric second harmonic generation images of human fetal membranes

Thursday, October 24, 2024
10:00 AM - 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Fetal membranes are the tissues that compose the amniotic sac surrounding the fetus. The membranes play a crucial role in providing protection and structural support during pregnancy. They consist of two layers: the inner amnion and the outer chorion. The amnion is largely responsible for membrane mechanical strength due to its high concentration of collagen fibrils.
Fetal membranes naturally rupture during labor. However, weakening of the membranes can result in premature rupture of membranes (PROM). PROM occurs in approximately 8% of pregnancies and increases the risk of complications such as infection.
Past studies have indicated that structural changes to the amnion collagen network create a “zone of altered morphology” where membrane rupture begins. Thus, other studies have explored the relationship between the structural and mechanical properties of the membranes.
Second harmonic generation (SHG) microscopy is an imaging technique capable of visualizing collagen structures in tissues. It has been used to analyze structural components of fetal membranes, including thickness. Prior studies have not found a correlation between total membrane thickness and mechanical strength (Samimi et al., 2023; Schober et al., 1994), but there is limited information on the relationship between sublayer thickness and mechanical strength.
Here, we developed a “human-in-the-loop” machine learning technique to segment the amnion basement collagen, compact, and fibroblast layers from SHG images, calculate their combined thickness, and compare the thickness measurements to prior data on membrane rupture pressure. This segmentation technique required manual processing of only 2% of the images, enabling efficient analysis of a large dataset.

Materials and

Methods:  Fetal membranes were sourced from normal-term pregnancies, both c-section births (n=32) and vaginally delivered (n=33). Multiple samples of each membrane were taken from the near-cervical (n=24) and near-placental regions (n=18). One sample of each location was tested for rupture pressure (inflation to failure). A location-matched sample was imaged using multiphoton microscopy for collagen thickness analysis.
To segment the microscopy images, the SHG channel z-stack of each sample (Fig. 1A) was imported into Ilastik. To train the model, basement collagen and fibroblast layers were manually annotated at 13 slices in the XZ and YZ planes. Segmentation output was exported as a probability map (Fig. 1B).
Probability map stacks were imported into MATLAB for post-processing. Images were binarized and the largest continuous volume was isolated. The median Z value for each X,Y coordinate pair in this volume was calculated, resulting in a median surface. The LOWESS (Locally Weighted Scatterplot Smoothing) algorithm was used to remove outliers and create a smooth median surface.
Using the Delaunay triangulation algorithm, this surface was partitioned into triangles (Fig. 1C). Triangle areas were calculated and summed to obtain the total surface area. Average amnion collagen thickness was estimated by dividing the volume of the binarized segmented collagen layer by the total surface area.
The relationship between layer thickness and rupture pressure for each combination of birthing method and anatomical sample location was evaluated using Pearson’s correlation coefficient. Significance of location-dependent differences in thickness and rupture pressure were evaluated using paired two-tailed t-tests.

Results, Conclusions, and Discussions: Paired t-tests did not show a significant difference in rupture pressure (Fig. 2A) or in amnion collagen thickness (Fig. 2B) between near-cervical and near-placental anatomical sampling locations in either labored (n=10) or C-section (n=6) birth groups. This is likely due to the high variability of these tissues and the small number of paired samples. These location-dependent differences may reach significance as we continue to analyze more data sets.
The Pearson correlation analysis (Fig. 2C), however, reveals a high degree of correlation between amnion collagen thickness and rupture pressure in all birth groups and sampling locations, suggesting that a thicker amnion collagen layer is indicative of increased membrane mechanical strength.
Prior studies (not using SHG microscopy) [1,2] that measured total thickness of the combined amniochorion did not find a correlation with strength. This is expected, given that most of the amniochorion thickness is comprised of chorion and decidua layers that are variable in thickness and are not load bearing.
Collagen is the structural component of the amniochorion, and multiphoton imaging uses the specific SHG signal to visualize collagen layers and their organization in intact 3D samples that have not been altered (e.g., histology preparation). Despite the excellent contrast, analysis and quantification of 3D image volumes poses a challenge. The large number of image voxels (512×512×500, X×Y×Z in this case) prohibits manual segmentation of structural layers. We developed a human-in-the-loop analysis pipeline using an open-source convolutional neural network (CNN) segmentation and classification software (ilastik by EMBL) followed by post processing in Matlab to accurately segment the structural collagen layers and measure their thickness over the entire field of view. This analysis pipeline allows efficient quantification of volumetric microscopy datasets and enables further measurements beyond collagen layer thickness, such as integrated intensity of the collagen SHG signal as a measure of collagen content in future studies. Such image-based measurements may eventually serve as indicators that identify at-risk pregnancies for PROM.

Acknowledgements (Optional):