Identification of Preeclamptic Placenta in Whole Slide Images Using Artificial Intelligence Placenta Analysis

Area of Science:

• Digital pathology
• Computational imaging
• Reproductive medicine

Background:

• Preeclampsia (PE) is a hypertensive pregnancy disorder associated with placental dysfunction and characteristic pathological lesions.
• Current pathological diagnosis lacks a gold standard, hindering differentiation between PE and non-PE pregnancies.
• Computational pathology offers automated analysis for improved diagnostic capabilities.

Purpose of the Study:

• To assess the efficacy of computational pathology in identifying preeclamptic placentas.
• To develop and validate a computational model for PE detection using placental whole-slide images (WSIs).

Main Methods:

• Utilized 168 placental WSIs (84 PE, 84 controls) for model development and internal validation.
• Employed unsupervised learning (Auto Encoder, K-means clustering) and supervised learning (U-Net segmentation) for quantitative villi assessment.
• Developed a prediction model using ensemble methods, validated externally with 76 additional placental slides.

Main Results:

• The ensemble computational pathology model achieved an AUPRC of 0.771 (95% CI, 0.752-0.790) with 77.3% sensitivity and 71.1% specificity.
• External validation demonstrated good discrimination with an AUPRC of 0.725 (95% CI, 0.720-0.730).
• The computational model outperformed a clinical feature model (AUPRC 0.713).

Conclusions:

• Computational pathology shows significant potential for improving the identification of preeclamptic placentas.
• The developed model effectively distinguishes PE placentas, offering a valuable tool for clinical diagnosis.