Machine Learning Assessment of Gestational Age in Accelerated Maturation, Delayed Maturation, Villous Edema, Chorangiosis, and Intrauterine Fetal Demise.

Context.—: Assessment of placental villous maturation is among the most common tasks in perinatal pathology. However, the significance of abnormalities in morphology is unclear and interobserver variability is significant.

Objective.—: To develop a machine learning model of placental maturation across the second and third trimesters and quantify the impact of different pathologist-diagnosed abnormalities of villous morphology.

Design.—: Digitize placental villous slides from more than 2500 placentas at 12.0 to 42.6 weeks. Build whole slide learning models to estimate obstetrician-determined gestational age for cases with appropriate maturation and normal morphology. Define the model output as "placental age" and compare it to the chronologic gestational age.

Results.—: Our model showed an r2 of 0.864 and mean absolute error of 1.62 weeks for placentas with appropriate maturation in the test set. Pathologist diagnosis of accelerated maturation was associated with a 2.56-week increase in placental age (±2.91 weeks, P < .001), while delayed maturation was associated with a 0.92-week decrease in placental age (±1.82 weeks, P < .001). Intrauterine fetal demise causes diverse changes to placental age, driven by the nature of the demise. We tested the impact of training a model, using all live births. The resulting r2 was 0.874 and mean absolute error was 1.73 weeks. Furthermore, by including cases with abnormal maturation in the training data, the effect size of accelerated maturation was blunted to only 0.56 ± 2.35 weeks (P < .001).

Conclusions.—: We show that various abnormalities of villous maturation and morphology correlate with abnormalities in placental age. This "no pathologist" model could be useful in situations where pathologists are unavailable or the need for consistency outweighs the utility of expertise.