[Paper Review] Gestational Stage Prediction from Cervical Tissue Analysis Using Imaging Mueller Polarimetry Data

Sooyong Chae, Ajmal Ajmal+6 more|arXiv (Cornell University)|Feb 27, 2026

Preterm Birth and Chorioamnionitis0 citations

TL;DR

This study uses imaging Mueller polarimetry on murine cervical tissue across three gestational stages and compares an analytical ALV-based classifier with lightweight CNN ensembles to predict gestational stage, achieving 70% sample-level accuracy.

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.

Motivation & Objective

• Detect microstructural collagen remodeling in the cervix across gestational stages using imaging Mueller polarimetry.
• Extract polarimetric features via Lu–Chipman decomposition to characterize collagen organization.
• Compare analytical threshold-based classification with lightweight CNN ensembles for gestational stage prediction.
• Assess sample- and ROI-level performance and identify spatial regions driving accuracy (stroma).
• Discuss potential clinical translation for non-invasive preterm birth risk assessment.

Proposed method

• Acquire wide-field imaging Mueller polarimetry maps of murine cervix tissue at three gestational days (D6, D12, D18).
• Apply Lu–Chipman decomposition to obtain diagonal Mueller components and derive orientation maps and ALV (Azimuth’s Local Variability) as a measure of collagen disorder.
• Compute mean ALV per ROI and evaluate a two-threshold analytical classifier for three gestational classes.
• Train two lightweight CNNs (Model A on orientation psi map; Model B on ALV map) and form an ensemble by softmax averaging for final prediction.
• Evaluate all models under 20-fold leave-one-out cross-validation, ensuring ROIs from the same specimen are held out together.

Experimental results

Research questions

• RQ1 Can Mueller polarimetry-derived features distinguish early, mid, and late gestational stages based on cervical collagen remodeling?
• RQ2 Do orientation and ALV channels provide complementary information for gestational stage prediction?
• RQ3 Is a lightweight CNN-based approach capable of real-time, non-destructive gestational staging from polarimetric maps?
• RQ4 What is the relative performance of an analytical ALV-threshold classifier versus deep learning models in LOOCV?

Key findings

• CNNs on orientation (psi) or ALV maps each achieve 70.0% sample-level accuracy under LOOCV.
• Ensemble of orientation and ALV models achieves 70.0% sample-level accuracy and 65.2% ROI-level accuracy, the highest among methods.
• Analytical ALV threshold classifier achieves 55.0% sample-level and 57.6% ROI-level accuracy.
• Per-class results show 71% (D6), 50% (D12), and 86% (D18) accuracy for CNNs; the analytical method perfects D18 but struggles with D6 and D12.
• Spatial maps indicate the highest prediction performance in the stroma, where collagen remodeling is prominent.
• Mean ALV increases across gestation (D6 < D12 < D18) with D18 well-separated statistically from the others.