How AI can improve antenatal estimation of gestational age

How AI can improve antenatal estimation of gestational age | Image Credit: © putilov_denis - © putilov_denis - stock.adobe.com.

Novice users with no previous ultrasonography training can have similar accuracy for estimating gestational age (GA) between 14- and 27-weeks’ gestation as expert sonographers performing standard biometry when using a low-cost, point-of-care artificial intelligence (AI) tool, according to a recent study published in JAMA.¹

Obstetrical sonography is a useful tool for estimating GA, which is used to guide various aspects antenatal care. GA is also vital for clinical decision-making, including whether to provide corticosteroids in patients likely to experience preterm delivery.

According to the World Health Organization (WHO), pregnant patients should receive an ultrasonography evaluation at least once before 24 weeks’ gestation.² Recent expansions in AI technology may allow this practice to expand to low- and middle-income countries.¹

To evaluate the accuracy of a diagnostic algorithm for estimating GA developed in 2022, investigators conducted a prospective diagnostic accuracy study. Participants included pregnant patients with single, nonanomalous, first-trimester pregnancies.¹

A “ground truth” GA was determined in patients through a transvaginal crown-rump length measurement, and follow-up visits occurred at random dates from 14 weeks to 27 weeks and 6 days’ gestation. GA was estimated by novice sonography users during these visits using the AI-enabled test.

The study standard consisted of obstetrics-trained sonographers evaluating GA with fetal biometry using a high-specification ultrasound machine. The index test from novice users was performed before the study standard, and index test users could not consult with study sonographers while using the tool.¹

Participants were aged at least 18 years, had a viable pregnancy under 14 weeks’ gestation, planned to remain in their current geographical area of residence during the trial, provided written consent, and were willing to adhere to study procedures. Exclusion criteria included a body mass index over 40, twin or multiple gestation, known major fetal anomaly, and any condition that would impact participation.

Novice users completed a 1-day training session before the start of the study, covering patient positioning, software navigation, probe orientation and pressure, gel application, and blind sweep collection. During the index test, the software guided users with a series of 10-second blind sweep videos.¹

During the study standard, a trained sonographer collected 2 fetal head circumference measurements. Additional measurements included biparietal diameter, abdominal circumference, and femur length.

The estimation errors were compared between the index text and study standard as the primary outcome of the analysis. This was assessed at the difference in mean absolute error (MAE) between methods.

There were 400 participants included in the final analysis, aged a median 29 years and with a median 13 years of education. Of participants, 63% were parous and 25 were HIV-seropositive. Neither the index test nor reference standard were linked to any adverse events.¹

The index test failed to get an estimate for 1 case, leading to 399 individuals with available paired assessments. An MAE of 3.19 days was reported for the index test vs 3.03 days for the study standard, indicating a difference of 0.16 days.

A correct classification within 7 days of the ground truth GA was reported for 90.7% for the index test and 92.5% for the study standard. This was a difference of -1.8%, indicating similar success rates between both methods. Additionally, high accuracy for GA estimates within 14 days was reported for both tests, each at 99.8%.¹

No significant changes were observed for the difference in MAE based on study site. However, the difference between tests was 0.70 for patients with a first-visit body mass index of 30 or greater.

These results indicated similar accuracy between novice users estimating GA with a low-cost AI-enabled ultrasonography and experienced sonographers conducting a standard biometry on high-specification machines. Investigators concluded this data could be used to advance the WHO goal of estimating GA with ultrasonography for all pregnant patients.¹

References

1. Stringer JSA, Pokaprakarn T, Prieto JC, et al. Diagnostic accuracy of an integrated AI tool to Eetimate gestational age from blind ultrasound sweeps. JAMA. 2024. doi:10.1001/jama.2024.10770
2. World Health Organization. Maternal and fetal assessment update: imaging ultrasound before 24 weeks of pregnancy. March 28, 2022. Accessed August 13, 2024. https://www.who.int/publications/i/item/9789240046009