A Unified Deep Learning Pipeline for Placental Morphometric and Phenotypic Mapping in Gestational Diabetes

This study investigated morphometric differences in placentas from gestational diabetes mellitus (GDM) pregnancies by integrating deep learning with quantitative statistical analysis. Whole slide images (WSIs) of placentas from four pathological groups: small-for-gestational-age (SGA), appropriate-for-gestational-age (AGA), large-for-gestational-age (LGA), and non-GDM pregnancies were analyzed to evaluate cell type distributions, morphology, and connectivity. A state-of-the-art benchmark model was used for comparison. This pipeline produced plausible cell maps and revealed significant cell-type-specific differences across groups. SGA placentas showed reduced vascular myocyte connectivity, and AGA placentas maintained robust cellular interactions. LGA and non-GDM placentas showed intermediate or compensatory remodeling patterns. These results highlight the potential of morphometric analysis as a biomarker discovery tool for placental dysfunction and suggest that deep learning-based cellular mapping can provide clinically relevant insights into pregnancy outcomes.