Skip to main content
SMFM 2024 Pregnancy Meeting Main banner
Final Program


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 on social media
Faculty have requested this content not be shared on social media
CME Credit Offered
CME Credit Offered
Poster Icons
Foundation Awardee
Foundation Awardee
No Social Media
No Social Media
Back
Q&A

Ultrasound/Imaging

Poster Session 2

(553) Evaluating AI’s Prediction of Time Until Delivery Across Trimesters with Consideration of Influential Factors

Tuesday, February 13, 2024
3:30 PM – 5:00 PM  
Location: Prince George's Exhibit Hall CDE, Lower Atrium Level
Objective: To examine the accuracy of artificial intelligence (AI) in predicting delivery timing as gestational age progresses with consideration of risk factors.

Study Design:

A proprietary AI software was trained and evaluated on 555,948 de-identified ultrasound images from 19,920 unique exams in 5,714 patients who delivered at the University of Kentucky from 2017-2021. Deidentified ultrasound images from 79% of this cohort (4,505 patients, 15,614 unique studies, 435,857 images) were used to train the AI. Clinical characteristics were excluded to blind the AI towards confounding demographics.  



The remaining 21% (1,209 unique patients, 4,306 ultrasound studies, 120,091 total images) were used for derivation and validation of the AI’s performance. Delivery outcomes were blinded from the AI analysis by independent 3rd party monitoring. The predicted number of days from exam until delivery were made for each patient after each unique ultrasound exam. R2 values were generated to correlate the AI’s predictions with the actual number of days until delivery.  



The validation set was divided into subgroups based on GA at exam. The CI at each trimester were compared. The influence of potentially confounding variables on AI prediction within each trimester was examined by comparison to a reference subgroup. 


Results: 1,514 patient exams were evaluated to assess prediction, See Table. The Mean Absolute Error between the AI prediction vs actual time to delivery was 15.06 days (95% CI 12.13-17.98) in the 1st trimester, 14.25 days (95% CI 12.59-15.91) in the 2nd trimester, and 12.16 (95% CI 10.79-13.53) days in the third. The overlapping confidence intervals suggest the AI did not operate differently over time. Accuracy of prediction did not vary across most risk factors examined (Age, BMI, and Race) except those with a history of PTB in the mid-trimester.

Conclusion: AI can predict delivery timing from ultrasound images alone and the performance did not vary appreciably across trimesters nor was it influenced by most patient characteristics. Further improvement in prediction for patients with a prior preterm birth is ongoing.