A Denver startup says its artificial intelligence software can look at routine prenatal ultrasounds, estimate when a baby is likely to be born, and flag which pregnancies may be at higher risk for delivering early. The company says the tool works off standard scans, including ones taken very early in pregnancy, and it has been rolling out study results while chasing regulators ahead of a planned launch in U.S. clinics.
On Feb. 11, the U.S. Food and Drug Administration granted a De Novo classification for the company’s software, Delivery Date AI, according to the FDA. The De Novo process is for new, moderate-risk devices and creates a defined pathway to market along with special controls and post-marketing requirements.
Peer-reviewed validation and the study
The company leans heavily on a peer-reviewed PAIR study that trained and validated its model on thousands of exams and roughly two million ultrasound images. The paper reports that performance improved as the model was retrained over time, according to The Journal of Maternal‑Fetal & Neonatal Medicine. Investigators at the University of Kentucky and co-authors from Ultrasound AI are listed, and the study describes strong correlation metrics for term births, while also acknowledging that earlier model versions had lower sensitivity when it came to spotting spontaneous preterm deliveries.
In a company press release, Ultrasound AI founder Robert Bunn called the FDA decision a “major milestone” and argued that the software could widen early risk detection in clinics that do not have specialist resources on hand, according to PR Newswire. The announcement also pointed to previous international regulatory moves as part of a broader commercial strategy.
How the technology works
Delivery Date AI is built to plug into existing ultrasound setups rather than replace them. It analyzes standard DICOM images and returns a personalized estimate of “days until delivery,” according to Ultrasound AI. The company says the tool requires no extra hardware and produces results in seconds, which it argues should make it relatively easy to fold into routine prenatal workflows.
What the numbers actually say
The PAIR study paints a mixed but promising picture. Earlier versions of the model had modest sensitivity for spontaneous preterm birth, around 39 percent, while specificity stayed high. The authors report better performance after additional rounds of retraining, per The Journal of Maternal‑Fetal & Neonatal Medicine. In practical terms, that suggests Delivery Date AI may help sort pregnancies into higher- and lower-risk groups for closer monitoring, but it is not a crystal ball that can guarantee whether a baby will or will not arrive early.
Why now: Ultrasound AI’s peer-reviewed data landed in 2025, and the company moved in parallel on non-U.S. approvals and distribution partnerships while it sought a green light at home. The Feb. 11 De Novo decision tracks that timeline and opens a clearer path to U.S. commercialization, the company told partners and media, according to PR Newswire. Ultrasound AI has said it plans to phase access as it scales up and supports clinical adoption.
Local impact and next steps
Headquartered in the Denver metro area, Ultrasound AI is positioning the FDA classification as a green light for local clinics and hospital systems to eventually add the software to routine scans once commercial roll-out, training, and payer talks are in place, according to Ultrasound AI. Expect a slower, pilot-style opening act. Early deployments are likely to focus on testing the tool in real clinical workflows and gathering post-market data before anything like broad adoption.
Local television coverage homed in on the company’s claim that the tool can deliver predictions from early ultrasounds, sometimes as soon as eight weeks into pregnancy, and highlighted the regulatory milestone for Denver viewers, according to 9News. Independent reporting and the peer-reviewed study both urge a note of caution. Experts say AI tools look promising but should ride alongside, not replace, clinical judgment and established screening until there is more real-world, post-market evidence, as noted by industry coverage.
