Machine Learning Models Show Promise for Predicting Placenta Accreta Spectrum Risks

Researchers in China are exploring the potential of machine learning to predict adverse outcomes in patients with placenta accreta spectrum (PAS), a dangerous perinatal complication that has become a leading cause of maternal mortality worldwide. Their findings suggest that artificial intelligence could help providers identify high-risk patients before delivery and guide individualized management strategies.

PAS occurs when the placenta fails to separate properly from the uterine wall, often leading to massive intraoperative hemorrhage during childbirth. The incidence of PAS has increased in recent decades, a trend linked to the rising number of cesarean deliveries. Although some patients experience severe complications such as hemorrhage or hysterectomy, not all women with PAS have poor outcomes, underscoring the importance of effective preoperative risk stratification.

“During delivery, patients with PAS are at high risk of experiencing placental retention, which can result in severe complications such as massive intraoperative hemorrhage and hysterectomy. However, not all pregnant women with PAS develop these adverse clinical outcomes,” explained lead author Hanlin Liu, from the Department of Radiology at the Third Affiliated Hospital of Shenzhen University, and colleagues. “Therefore, accurately predicting the risk of adverse outcomes in PAS patients before delivery is of critical importance for developing standardized and individualized clinical management strategies.”

To build a predictive tool, the team collected data from a large cohort of women with confirmed PAS. Using five MRI morphological indicators and six clinical features, they trained machine learning algorithms to identify patterns linked with adverse outcomes. Three classifiers—AdaBoost, TabPFN, and CatBoost—were evaluated through internal testing and external validation.

Of the three, the CatBoost model performed best, achieving AUROCs of 0.90 and 0.84 in internal and external validation, respectively. The model’s predictions closely matched the actual outcomes of patients. Variables most strongly associated with risk included prior cesarean deliveries, placental abnormal vasculature area, and parturition history. Meanwhile, longer cervical canal length and higher gestational age correlated negatively with high-risk predictions.

The researchers believe the model can support clinicians in tailoring treatment strategies before delivery. “Currently, there is no universally accepted optimal treatment strategy for PAS, as intraoperative blood loss, surgical complexity, and the need for additional hemostatic interventions can vary significantly depending on the depth and extent of placental invasion,” the authors wrote. “Therefore, accurate preoperative risk stratification for adverse outcomes is essential for developing individualized treatment plans.”

To make their tool more accessible, the team has also released it as a web-based platform, enabling providers to input patient data and receive risk predictions in real time.

If validated further, machine learning–based risk assessment could become a vital component of PAS management, helping physicians anticipate complications, prepare surgical teams, and improve maternal outcomes in one of obstetrics’ most challenging conditions.