Machine learning-based model predicts risk of severe pregnancy complication using MRI data
Experts are hopeful machine learning techniques could improve patient outcomes related to placenta accreta spectrum (PAS), a serious and sometimes deadly perinatal complication.
A leading cause of maternal mortality, PAS can trigger the failure of placental separation during delivery. Instances of the condition have risen in recent decades, potentially due to the increased utilization of cesarean deliveries. If PAS is not managed properly, it can lead to massive intraoperative hemorrhage. As such, there is an unmet need to stratify patients’ risk prior to delivery to ensure providers can proactively address the potential for complications.
“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,” Hanlin Liu, with the department of radiology at the Third Affiliated Hospital of Shenzhen University in China, and colleagues explained. “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.”
In the case of PAS, machine learning-based predictions could arm providers with valuable information relative to patients’ risks prior to delivery. To develop an algorithm capable of doing so, researchers trained two models using five MRI morphological indicators and six clinical features from a large group of patients with confirmed PAS. Three machine learning classifiers (AdaBoost, TabPFN and CatBoost) were trained and evaluated via internal testing and external validation.
The CatBoost model recorded the highest performance, yielding AUROCs of 0.90 and 0.84, respectively. The team determined the model’s risk predictions were in line with their actual reported outcomes. Cervical canal length and gestational age negatively correlated with high-risk predictions, while prior C-sections number, placental abnormal vasculature area and parturition were positively associated.
The team suggested their model has strong potential as a preoperative risk assessment tool.
“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. Therefore, accurate preoperative risk stratification for adverse outcomes is essential for developing individualized treatment plans,” the group wrote.
To assist other providers, the team has made their tool available on a web-based platform for ease of use. It can be found here.
Learn more about the study’s findings here.
In addition to her background in journalism, Hannah also has patient-facing experience in clinical settings, having spent more than 12 years working as a registered rad tech. She began covering the medical imaging industry for Innovate Healthcare in 2021.