Skip to main content
SMFM 2026 Pregnancy Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.
Presentation Icons
Additional registration fee required
Additional registration fee required
Faculty have requested this content not be shared outside of the session
Faculty have requested this content not be shared outside of the session
CME Credit Offered
CME Credit Offered
Abstract Award
Abstract Award
Recording available 2/16-5/2
Recording available 2/16-5/2
AIUM Credit
AIUM Credit
Foundation Awardee
Foundation Awardee
Poster Icons
Abstract Award
Abstract Award
Foundation Awardee
Foundation Awardee

Oral Plenary Session 2 - Fellows Plenary

Oral Plenary Sessions

Oral Plenary Session 2 - Fellows Plenary

(39) AI Based Ultrasound Screening for Early, Accurate Identification of Placenta Accreta Spectrum

Thursday, February 12, 2026
8:30 AM - 8:45 AM  
Location: Summit 201 - 204
Objective:
Placenta Accreta Spectrum (PAS) is a leading cause of maternal morbidity and mortality and is increasing in incidence, yet only 30% of cases are diagnosed antenatally. While ultrasonographic findings may inform risk of PAS, multiple factors may lead to an inconclusive or misdiagnosis. We hypothesized that a novel artificial intelligence (AI) model could be used as a screening tool to accurately predict PAS by interpreting 2D ultrasonographic placental images.

Study Design:
This single-center retrospective study included 756 placental ultrasound DICOM files from which 38,907 grayscale PNG frames were extracted from 113 patients at risk for PAS from 2018 to 2025. Mean gestational age at ultrasound was 30.89 ± 3.67 weeks. Patients were stratified to produce 79/17/17 train/validation/test groups. Images were classified by final pathologic FIGO grades. We used an ImageNet pretrained EfficientNetB0 backbone with global average pooling and regularized sigmoid head. Frame level probabilities from the convolutional neural network (CNN) output were averaged to patient level consensus scores (Fig 1A). These scores, together with number of prior Cesarean sections and previa status were input as variables into training a logistic regression, random forest, and gradient boosting classifier which were ensembled for final prediction of PAS incorporating patient identifiable risk factors (Fig 1B).

Results:
The CNN ensembled model predicted the presence or absence of PAS accurately in 88% of cases with a sensitivity of 100% and specificity of 75%. The positive predictive value was 81.8% and the negative predictive value was 100%. There were no false negatives in the testing cohort (Fig 1C). The AUC-ROC was 0.972 (Fig 1D). Model variable importance scores concentrated at the placental interface, highlighting biological plausibility (Fig 2).

Conclusion:
This novel AI model robustly achieved accurate, sensitive PAS prediction before delivery. The results of the model support its use as a screening tool to reduce missed diagnoses in PAS screening, warranting future prospective trials.