Alexandra Hammerquist, MD, explains AI-based ultrasound screening to identify early placenta accreta spectrum

At the Society for Maternal-Fetal Medicine 2026 Pregnancy Meeting in Las Vegas, Nevada, a study abstract demonstrated that a novel artificial intelligence (AI) model was able to “robustly” achieve accurate and sensitive Placenta Accreta Spectrum (PAS) prediction prior to delivery, which, according to the study authors, warrants future prospective trials.

In the video above, abstract presenter Alexandra Hammerquist, MD, a Maternal-Fetal Medicine clinical fellow at Baylor College of Medicine in Houston, Texas, explains the background of the study and its initial and feasibility results.

Hammerquist and colleagues noted that PAS is a leading cause of maternal moridity and mortality, with an increasing incidence. Still, only 30% of cases are diagnosed antenatally.

Investigators analyzed 756 placental ultrasound DICOM files, from which 38,907 grayscale PNG frames were extracted, representing 113 patients at risk for PAS between 2018 and 2025. Mean gestational age at ultrasound was 30.89 ± 3.67 weeks. Patients were stratified into 79/17/17 train/validation/test groups, and images were classified by final pathologic FIGO grades.

An ImageNet pretrained EfficientNetB0 convolutional neural network (CNN), with global average pooling and a regularized sigmoid head, generated frame-level probabilities that were averaged into patient-level consensus scores. These scores, along with the number of prior Cesarean sections and previa status, were incorporated into a logistic regression, random forest, and gradient boosting classifier, which were ensembled for final PAS prediction.

The CNN ensembled model predicted the presence or absence of PAS with 88% accuracy, 100% sensitivity, and 75% specificity. Positive predictive value was 81.8% and negative predictive value was 100%, with no false negatives in the testing cohort.

“So we tested this ensemble model on 17 of our patients. It’s only 15% of our total cohort,” said Hammerquist. “This is definitely a feasibility study. It’s more meant to give us hypotheses for the future and find out if this is really something we can pursue.”

The AUC-ROC was 0.972. Variable importance scores localized to the placental interface, supporting biological plausibility. The authors conclude that this AI model achieved accurate, sensitive PAS prediction before delivery and may serve as a screening tool, warranting future prospective trials.

Overall, Hammerquist said these initial results are exciting, though more work and research are needed prior to any clinical implementations.

“So this is really encouraging,” she said. “I think, the main takeaway from our study: that AI really has a lot of potential to help identify placenta accreta spectrum antenatally. But we need to validate this model that we’ve built before it’s ready for clinical practice. As long as this is able to be validated in a multicenter fashion, I think that this sort of model really has the potential to reduce maternal morbidity and possibly even mortality, especially in those resource-limited areas, because those patients could be sent for consultation or referred and transferred to tertiary care centers that could care for their delivery in a safe way.”

Reference:

Hammerquist AL, Sarada S, Mendez YH, et al. AI Based Ultrasound Screening for Early, Accurate Identification of Placenta Accreta Spectrum. Oral Plenary Session. Presented at: SMFM 2026. February 8-13, 2026. Las Vegas, Nevada.