AI Accurately Detects Placenta Accreta Spectrum in Ultrasound Images

A new artificial intelligence (AI) model is demonstrating promising accuracy in detecting placenta accreta spectrum (PAS) – a dangerous and increasingly common pregnancy complication – before delivery. The findings, presented at The Pregnancy Meeting on February 13, 2026, suggest the model could significantly improve early diagnosis and potentially reduce maternal morbidity and mortality associated with this life-threatening condition.

Understanding Placenta Accreta Spectrum

Placenta accreta spectrum (PAS) occurs when the placenta abnormally attaches to the uterine wall. This attachment can range in severity, but in all cases, it poses significant risks, particularly during delivery. The condition is often linked to prior uterine surgical procedures, such as cesarean delivery, and its incidence is rising in the United States. Currently, diagnosis relies on identifying risk factors and ultrasound findings, but these methods can be inconclusive or lead to misdiagnosis.

According to the American College of Obstetricians and Gynecologists (ACOG), the prevalence of PAS has increased substantially in recent decades. A study from 2016 indicated a rate of one in 272 for women with a birth-related hospital discharge diagnosis, a significant increase from one in 533 between 1982 and 2002. This rise underscores the urgent need for improved diagnostic tools.

The AI Model and its Performance

Researchers at Baylor College of Medicine developed the AI model by retrospectively analyzing 2D obstetric ultrasound images from 113 patients at risk for PAS who gave birth at Texas Children’s Hospital between 2018 and 2025. A total of 38,907 grayscale PNG frames were extracted from the 756 images for analysis. The mean gestational age at the time of the ultrasound was 30.89 ± 3.67 weeks.

The model demonstrated an overall accuracy of 88% in predicting the presence or absence of PAS. Importantly, it achieved 100% sensitivity, meaning it correctly identified all cases of PAS without any false negatives. Specificity was 75%, with a positive predictive value of 81.8% and a negative predictive value of 100%. The area under the receiver operating characteristic curve (AUC-ROC) was 0.972, indicating excellent discriminatory ability.

“The findings in this feasibility study are very promising,” said Alexandra L. Hammerquist, MD, a maternal-fetal medicine fellow at Baylor College of Medicine. “What is especially encouraging about our results is that our AI model was able to capture all cases of PAS, without any false negatives.”

Why Early Detection Matters

PAS is a leading cause of maternal morbidity and mortality in the U.S., yet, according to researchers, only approximately 30% of cases are diagnosed before delivery. An underdiagnosed condition can lead to massive maternal hemorrhage, multisystem organ failure, and even death during or after childbirth. Early detection allows for careful planning and preparation for delivery, often involving a multidisciplinary team and a planned cesarean hysterectomy to minimize risks.

The researchers noted that the model’s variable importance scores concentrated at the placental interface, suggesting a biologically plausible basis for its diagnostic capabilities.

Looking Ahead: Validation and Clinical Implementation

While the initial results are encouraging, Dr. Hammerquist emphasized that the model requires further validation before it can be implemented in clinical practice. “The main takeaway from our study is that AI models may be able to significantly assist in prenatal diagnosis of placenta accreta spectrum, but that our model must be validated before it is ready for clinical use,” she stated.

She clarified that the AI model is not intended to replace the expertise of trained sonographers or physicians. Instead, it is designed to serve as a supportive tool, helping clinicians synthesize patient data and arrive at a more confident diagnosis. “Ideally, models such as ours would allow a physician to better synthesize a patient’s clinical diagnosis and establish a more sure diagnosis,” she explained.

This research represents a significant step forward in the effort to improve the diagnosis and management of placenta accreta spectrum, offering hope for reducing the risks associated with this dangerous pregnancy complication.