Unlocking the Potential of Automated Placental Screening to Predict Cardiovascular Risk After Pregnancy Complications

Abstract

Introduction: Placenta-mediated diseases (PMDs) are critical sex-specific indicators of future maternal cardiovascular disease (CVD) risk. Placental maternal vascular malperfusion (MVM) may provide insight into postpartum CVD risk, but standard pathology assessment is time-consuming, resource-intensive, and dependent on specialized expertise.

Objective: This study aimed to investigate whether placental MVM, assessed either by manual pathology review or automated image analysis, could help identify women at elevated lifetime CVD risk at six months postpartum following a pregnancy complicated by a PMD.

Methods: In a multi-site cohort of women with PMDs (n=280), placental whole-slide images were reviewed by a perinatal pathologist, and lifetime CVD risk was calculated using postpartum clinical and biochemical cardiovascular risk factors. An attention-based deep learning model was applied to a subset of placental images (n=208) for automated MVM detection.

Results: Manually assessed MVM and algorithm-derived MVM were not significantly associated with high lifetime CVD risk (AUC=0.52) and provided no incremental predictive value beyond clinical factors. The automated model demonstrated weak-to-modest performance for MVM detection due to lesion complexity and gestational context.

Conclusion: Neither manual nor automated MVM assessment currently provides sufficient accuracy for use as a standalone CVD triage tool. Future progress requires integrated, multimodal approaches combining placental pathology with clinical and biochemical measures to improve postpartum risk stratification.