Gestational Stage Prediction from Cervical Tissue Analysis Using Imaging Mueller Polarimetry Data

TL;DR

It is demonstrated that Mueller polarimetry combined with deep learning models can detect gestational collagen remodeling noninvasively, offering a potential pathway toward objective cervical assessment for preterm birth risk.

Abstract

Preterm birth is associated with premature cervical remodeling, yet current clinical assessments cannot detect the underlying microstructural changes in collagen organization. We apply imaging Mueller polarimetry to murine cervical tissue at three gestational stages (early, mid, late) and develop classification methods to predict gestational stage from polarimetric maps. Using Lu-Chipman decomposition, we extract orientation and azimuth local variability maps that capture collagen fiber alignment and disorder. We evaluate two approaches under 20-fold leave-one-out cross-validation: an analytical threshold classifier on mean azimuth local variability, and a lightweight CNN ensemble (approximately 76k parameters) operating on spatially resolved maps. The ensemble achieves 70..0% sample-level accuracy, outperforming the analytical baseline (55.0%), with strong performance on early (71.0%) and late (86.0%) gestation. Spatial prediction maps confirm that classification accuracy is highest in the stroma, where collagen remodeling is most prominent. These results demonstrate that Mueller polarimetry combined with deep learning models can detect gestational collagen remodeling noninvasively, offering a potential pathway toward objective cervical assessment for preterm birth risk.

Figures (6)

• Figure 1: Visualization of murine cervix samples across different gestation days: Day 6, Day 12, and Day 18. The tissue regions were filtered using algorithm described in chae2026intensity. The widening of the cervical canal is visible in later stages.
• Figure 2: Left: M11 intensity image (Day 12) with red contours indicating ROI selection. Right: Expanded polarimetric maps of the selected ROI.
• Figure 3: Overview of the classification pipeline. Murine cervical tissue samples were collected across gestation, imaged with a Mueller polarimeter, and the resulting polarimetric maps were augmented for training. Two parallel CNN branches operate on (A) the orientation channel and (B) the ALV channel, each with four convolutional blocks (blue) and max pooling (red), followed by global average pooling and a linear classifier (green). Branch outputs are combined by softmax averaging to produce the (C) ensemble prediction.
• Figure 4: Distribution of mean ALV per sample across gestation days, with optimized decision boundaries $T_1$ and $T_2$. D6 and D12 distributions overlap, while D18 is well separated from both groups.
• Figure 5: Confusion matrices under LOOCV for (a) the orientation model (A), (b) ALV model (B), (c) the ensemble (A+B), and (d) the analytical threshold classifier.