Skip to main content
SMFM 2025 Pregnancy Meeting Main banner
Final Program


Program Book
Icon Legend
Presentation Icons
Additional registration fee required
Additional registration fee required
CME Credit Offered
CME Credit Offered
Abstract Award
Abstract Award
Poster Icons
Additional registration fee required
Additional registration fee required
Faculty have requested this content not be shared on social media
Faculty have requested this content not be shared on social media
CME Credit Offered
CME Credit Offered
Abstract Award
Abstract Award
Back

Poster Session 3

(860) Leveraging deep neural networks of transvaginal cervical characteristics to predict delivery timing

Friday, January 31, 2025
10:30 AM – 12:30 PM  
Objective: To utilize deep neural networks of biomechanical characteristics obtained from transvaginal ultrasound videos to predict timing of delivery within 7 days at term.

Study Design: This is a prospective longitudinal cohort of patients who underwent fully quantitative cervical elastography (FQ-CES) starting at 37 weeks and weekly thereafter until delivery. FQ-CES quantifies cervical tissue stiffness utilizing a transvaginal ultrasound probe to quantify both pressure applied and tissue deformation to allow operator independent comparisons. Singleton pregnancies at 37 weeks without delivery planed before 39 weeks were enrolled. FQ-CES and cervical length (CL) were collected weekly until delivery. Patients remained enrolled regardless of their eventual timing or indication for delivery. Transvaginal ultrasound videos from FQ-CES exams were used to build a deep convolutional neural network with Xception architecture. The primary outcome is delivery within 7 days.

Results: 118 individuals with 241 FQ-CES ultrasound exams were included. 80% of data was randomly selected as training dataset with a 5-fold cross-validation split. Among 241 FQ-CES exams, 102 (42.3%) delivered  ≤7 days and 139 (57.7%) delivered >7 days. Frames from ultrasound videos were treated as individual image inputs. For multiple image predictions inside one FQ-CES exam, majority vote method was applied. The deep convolutional neural network showed high predictive performance including high 92% sensitivity, 90% specificity, and accurate prediction of delivery within 7 days in 91% of the population. In contrast, the network utilizing CL < 2cm showed lower prediction performance (Table)

Conclusion: Deep learning models based on FQ-CES videos have much higher predictive performance for delivery within 7 days. This may be due to the ability to capture characteristics related to anatomy, biomechanical properties, and cervical tissue interaction with surrounding tissues. This technique could be be leveraged to capture intricate cervical characteristics to predict delivery timing.