Ƶ

AI-Powered Device Estimated Gestational Age as Accurately as Sonographers

— Tool could be useful in resource-limited settings without expensive ultrasound equipment

Ƶ MedicalToday
A photo of a person holding the Butterfly IQ+ handheld ultrasonography device.

An artificial intelligence (AI)-powered handheld ultrasonography device operated by novice users estimated gestational age as accurately as credentialed sonographers using ultrasound equipment, a prospective diagnostic accuracy study showed.

In estimating gestational age among 400 pregnant individuals in Lusaka, Zambia, and Chapel Hill, North Carolina, the AI-enabled device had a mean absolute error of 3.2 days compared with 3.0 days with the study standard of fetal biometry with a high-specification machine, which met criteria for equivalence.

The percentage of assessments within 7 days of the ground truth gestational age -- assessed via transvaginal crown-rump length measurement -- was also comparable, at 90.7% with the AI tool and 92.5% with the study standard, reported researchers led by Jeffrey Stringer, MD, of the University of North Carolina School of Medicine in Chapel Hill, in .

"What the study did was compare the gestational age tools to the expert, and it found that throughout pregnancy, up until 37 weeks of gestation -- which is term pregnancy -- that the AI tool works as well as an expert," Stringer told Ƶ.

He noted that when he was a physician practicing in Zambia, it became clear to him that it was hard to deliver high-quality obstetric care without ultrasounds because they "anchor care all the way from the very beginning through delivery."

Handheld devices are far more affordable than traditional ultrasound equipment, and AI algorithms that can interpret medical imaging have "offered this opportunity to us to try to change the game," he said. Plus, the device only takes 3 minutes and requires less trained labor.

In a , Stringer tested this AI model's ability to estimate gestational age based on thousands of blind ultrasound sweeps in low-resource settings, reporting that it estimated gestational age with similar accuracy to trained sonographers conducting standard fetal biometry.

The new study is "validation that the thing works," he said.

Stringer pointed out that the tool does not use the same kind of AI as ChatGPT, but rather is smaller, simpler, and built for very specific tasks.

"What we've learned is that the AI does exactly what you teach it to do and nothing more and nothing less," he explained. The team is working on algorithms that would allow the same device to make five diagnoses.

"I fully anticipate that over the next several years, we'll have a suite of these kinds of tools that can make it so that a patient will not have to drive 2 and a half hours to the specialty referral center to get an ultrasound, but that it can be done at the very lowest-level primary care facility in a maternity health desert in the United States," Stringer said.

Authors of noted that resource-limited settings -- especially low- and middle-income countries -- may not have the expensive equipment or skilled sonographers to offer this crucial service.

If this study's results are replicated, these tools could "significantly aid in accurately dating pregnancies, optimizing the timing of [gestational age]-dependent interventions, and enhancing overall pregnancy care" in such areas, wrote Methodius Tuuli, MD, MPH, MBA, of Women & Infants Hospital of Rhode Island and the Alpert Medical School of Brown University in Providence, and co-authors.

They noted that the blind sweep technique is easier to learn and, previously, "other more complex AI-assisted algorithms have required precise plane acquisition and caliper placement that are unlikely to have broad application for novice users."

Still, they cautioned "that although blind sweep ultrasonography dating for [gestational age] using a low-cost, easy-to-learn technique is a major step forward, it is far from a silver bullet."

This study included 400 adults with viable, single, non-anomalous, first-trimester pregnancies. People with a BMI greater than 40 were excluded, as were those who were pregnant with twins or multiples, and those with known fetal anomalies.

Median maternal age was 29, and median gestational age at first ultrasound was 11.7 weeks. Eight percent were HIV-seropositive (all from Zambia), and 63% were parous.

After ground truth gestational age was established by credentialed sonographers via transvaginal crown-rump length measurement, participants returned for two randomly assigned follow-up visits during which a novice user with no previous sonography training used blind sweeps of the maternal abdomen with the AI-enabled Butterfly IQ+ handheld ultrasonography device to assess gestational age (they underwent a 1-day training session on how to use the device). Obstetrics-trained sonographers used a high-specification ultrasound machine to determine gestational age with fetal biometry.

Mean absolute error was the primary outcome and was calculated by "comparing each method's estimate to the previously established [gestational age] and considered equivalent if the difference fell within a prespecified margin of ±2 days," the authors wrote.

Stringer and team noted that the study wasn't designed to assess high-risk patients, such as those with hypertension, diabetes, or class III obesity, which was a limitation. In addition, testing more geographic locations would increase the generalizability of the findings. Since pregnant women with known fetal anomalies were excluded, it is unknown if the tool is effective in those cases.

The editorialists also pointed out that the study had a modest sample size and limited information on how the individuals doing the scans were trained.

  • author['full_name']

    Rachael Robertson is a writer on the Ƶ enterprise and investigative team, also covering OB/GYN news. Her print, data, and audio stories have appeared in Everyday Health, Gizmodo, the Bronx Times, and multiple podcasts.

Disclosures

The study was funded by the Bill and Melinda Gates Foundation. Butterfly Systems donated ultrasonography probes for the study and worked with the team to implement an encrypted version of the software.

Study authors reported no conflicts of interest.

Tuuli reported receiving grants from the NIH and the Gates Foundation, as well as nonfinancial support from Organon and Pharmacosmos.

Other editorialists had no conflicts of interest.

Primary Source

JAMA

Stringer JSA, et al "Diagnostic accuracy of an integrated AI tool to estimate gestational age from blind ultrasound sweeps" JAMA 2024; DOI: 10.1001/jama.2024.10770.

Secondary Source

JAMA

Gimovsky AC, et al "Enhancing obstetric ultrasonography with artificial intelligence in resource-limited settings" JAMA 2024; DOI: 10.1001/jama.2024.14794.