Martin G. Tolsgaard doesn't mince words. "We miss far more than half of risk pregnancies," says the maternal-fetal medicine specialist at Copenhagen's Rigshospitalet. It's a startling admission from a physician whose entire practice revolves around identifying expectant mothers and babies heading toward danger. But the numbers back him up: Across multiple large clinical cohorts, antenatal detection of fetal growth restriction hovers somewhere between 33% and 53%. Flip a coin, and you might have comparable odds.
That grim reality—frustrating obstetricians for decades—is now attracting a different kind of attention. A wave of deep-learning algorithms aimed at prenatal ultrasound is demonstrating performance gains that could, perhaps, finally shift those odds. Tolsgaard himself is now chief medical officer of Prenaital ApS, a Copenhagen-based startup claiming its AI detects roughly 30% more high-risk pregnancies than current clinical practice. A growing body of peer-reviewed evidence suggests that figure may even be conservative.
Prenaital recently secured $1.57 million in funding and is preparing its fetal growth assessment module for U.S. and European market entry in 2026. The company—spun out from the Capital Region of Denmark, the University of Copenhagen's Pioneer Centre for AI, and the Technical University of Denmark—represents one thread in a broader transformation. AI-assisted prenatal screening is moving from academic validation to hospital deployment at a pace that has surprised even optimistic observers. Whether it can deliver on its promise is another question entirely.
A System That Routinely Misses Danger Signals
The backdrop makes the urgency clear. The United States recorded a maternal mortality rate of 18.6 per 100,000 live births in 2023. Persistent racial disparities remain, and outcomes trail most wealthy nations. More than 35% of U.S. counties qualify as "maternity care deserts"—lacking obstetric facilities or clinicians. Over 5.5 million women live in counties with no or severely limited prenatal services. Between 2021 and 2024, more than 100 obstetric units closed. One in three counties lacks obstetric clinicians altogether.
Against that landscape, ultrasound—the primary tool for assessing fetal growth, detecting structural anomalies, and gauging pregnancy risk—depends heavily on operator skill, time, and equipment quality. Sonographer burnout is endemic. Musculoskeletal injury rates reach 75% to 90% in some surveys. The result? A system that routinely misses danger signals. Obesity lowers detection rates further. Many preterm births aren't predicted early enough for preventive therapy.
Tanja Danner, Prenaital's CEO, describes current screening as hampered by "subjectivity and tech limitations." The company's deep-learning models—trained on more than 10,000 ultrasound images from Danish hospitals—aim to identify patterns "not visible to the human eye," according to co-founder Aasa Feragen, a professor at the Technical University of Denmark. A study published in May 2025 in npj Digital Medicine, co-authored by several Prenaital founders, showed AI-based fetal growth assessment detected 70% of small-for-gestational-age cases versus 58% for the standard Hadlock biometry method. The AI also appeared to reduce demographic and technical bias, though real-world validation remains ongoing.
Regulatory Green Lights Arrive in Clusters
The past eighteen months have witnessed a cluster of FDA clearances signaling the sector's maturation. In June 2024, Clarius received 510(k) clearance for automated fetal biometrics on handheld ultrasound devices—targeting midwives and rural clinics. France's Sonio secured FDA clearance for its real-time view and structure detection software (Detect V1) in 2023, updated it (V2) in 2024, and later cleared a "Suspect" module for anomaly detection. BrightHeart, another French-U.S. player, has collected multiple FDA clearances, including one for its B-Right Views module in May 2025. The company also won Predetermined Change Control Plan (PCCP) approval—enabling iterative model updates without filing new 510(k)s each time.
In January 2026, BioticsAI announced FDA clearance for AI supporting fetal ultrasound quality assessment, anatomical completeness, and automated reporting. These aren't incremental workflow tweaks. They represent fundamental shifts in how ultrasound exams are acquired, interpreted, and documented.
Major equipment manufacturers are bundling AI into their platforms. GE HealthCare's Voluson line now integrates the "SonoLyst" AI suite—featuring view recognition and measurement automation. The company acquired Intelligent Ultrasound's clinical AI business in 2024. Samsung Medison launched its Z20 system with real-time AI workflow tools in the U.S. in January 2025, showcasing a partnership with Sonio at the Society for Maternal-Fetal Medicine conference. The message from original equipment manufacturers: AI is no longer experimental add-on software but a core feature hospitals expect in new purchases.
Mount Sinai became the first health system in New York City to deploy BrightHeart's FDA-cleared AI for congenital heart defect screening in December 2025. Published studies of the technology report greater than 97% identification of serious CHD in controlled settings, an 18% reduction in reading time, and a 15.3 percentage-point boost in sensitivity for CHD findings when used as a clinical assist tool. Whether those gains hold up outside controlled environments remains to be seen—but the fact that Mount Sinai, an academic medical center with rigorous quality standards, is willing to deploy the technology suggests something has shifted.
The Ecosystem Expands Beyond Imaging
Ultrasound AI is not advancing in isolation. Mirvie's cell-free RNA test, validated in a Nature Communications paper published in April 2025, identified 91% of preterm preeclampsia cases in women over 35—months before symptoms appeared. The company launched its Encompass test and pilot partnerships during 2025 after earning FDA Breakthrough Device designation. A separate Nature Medicine study in February 2025 highlighted cell-free DNA approaches for preeclampsia risk stratification. These molecular diagnostics may ultimately combine with imaging AI to create multimodal risk profiles.
Remote monitoring is scaling in parallel. Nuvo Group's INVU device—FDA-cleared for remote non-stress tests and maternal/fetal heart rate tracking—went public via SPAC in 2024. The platform is reimbursable under CPT code 59025 for clinician interpretation. Medicaid paid approximately $517 million for that code between 2018 and 2024, signaling a sizable reimbursement channel. Expanding 12-month postpartum Medicaid coverage in most states by 2025 has also improved continuity for high-risk pregnancies that need extended monitoring.
University of Utah researchers analyzed roughly 9,558 pregnancies using AI and uncovered previously unknown compound risk patterns for stillbirth and complications. Sera Prognostics continues to build payer traction for its PreTRM blood test, though commercialization remains early. The ecosystem is beginning to resemble oncology's convergence of imaging, genomics, and liquid biopsy—different modalities triangulating on the same clinical problem. Whether prenatal care will follow that path is still an open question.
Regulatory Complexity Deepens
The FDA finalized guidance on PCCPs for AI and machine-learning software as a medical device in January 2025, allowing companies to pre-authorize specific model updates without repeated clearances. BrightHeart's PCCP approval illustrates how that mechanism works in practice: the company can iterate its algorithms within predefined performance envelopes, accelerating clinical refinement without drowning in paperwork.
The European Union's AI Act entered into force in August 2024 with staggered timelines. Prohibitions took effect in February 2025. General-purpose AI requirements hit in August 2025. Most high-risk AI obligations arrive in August 2026. For AI embedded in regulated medical devices—such as prenatal ultrasound software—full compliance isn't required until August 2027. Those obligations include documentation, quality management system integration, post-market surveillance, transparency artifacts, and human oversight protocols. The compliance burden is real, but European vendors may differentiate themselves by building robust real-world evidence frameworks ahead of enforcement.
The International Society of Ultrasound in Obstetrics and Gynecology released a positioning statement in November 2025 supporting "responsible, equitable, evidence-based adoption" of AI, with explicit attention to bias, oversight, and transparency. Professional society endorsement matters; it shapes hospital purchasing committees and training curricula.
Reimbursement remains murky. Prenatal ultrasound AI is typically bundled into imaging fees with no separate CPT code—unlike the emerging Category 1 CPT code for coronary CTA AI that Medicare is evaluating. Hospitals adopt the technology for quality, efficiency, and workforce retention rather than direct fee-for-service revenue. That may change. Radiology AI precedents suggest eventual pathways for tools with demonstrated outcome impact. But for now, the business case hinges on operational value and risk mitigation.
The Path Forward—and the Obstacles
The fetal monitoring market overall is projected to grow from roughly $4.5 billion in 2024 to approximately $7 billion by 2030, driven by aging equipment replacement, workforce constraints, and payer pressure on outcomes. AI-specific ultrasound imaging is a smaller but faster-growing segment, with estimates ranging from $1.12 billion in 2025 to $2.52 billion by 2035. Remote fetal monitoring could reach $3.5 billion by 2035, up from $1.16 billion in 2024.
Prenaital targets U.S. and European market entry in 2026 with its fetal growth assessment product, followed by preterm risk and quality assurance modules. The company's claims—25% to 50% improvement over routine practice, narrowed to about 30% in most communications—align with published Danish validation studies showing meaningful detection gains versus standard biometry methods. Whether those gains translate into reduced adverse outcomes in prospective interventional trials is the critical question. Early evidence from Sera Prognostics' AVERT trial, which demonstrated an 18% reduction in severe neonatal morbidity and mortality when preterm risk testing guided management, suggests the answer may be yes. But one trial does not a revolution make.
Handheld ultrasound with AI biometrics—exemplified by Clarius—could democratize basic fetal assessment in underserved areas where skilled sonographers are scarce. GE's integration of AI into the Voluson line, claiming time savings up to 40%, addresses ergonomics and retention in hospitals struggling to staff ultrasound departments. Samsung's Z20 partnership with Sonio shows how vendors can embed third-party AI rather than building everything in-house.
The obstacles are familiar to any clinical AI deployment: integration burden, workflow redesign, training overhead, and the need for prospective outcomes data that moves beyond sensitivity and specificity to show lives saved or complications prevented. Bias and generalizability concerns remain legitimate, particularly when training data skews toward specific populations or scanning equipment. A recent JMIR scoping review of machine learning for maternal morbidity and mortality noted "many models; few deployed clinically."
Still, the trajectory is unmistakable. Equipment giants like GE and Samsung are bundling AI into flagship products because customer expectations have shifted. A Danish startup founded by hospital clinicians frustrated with current tools has raised capital to commercialize algorithms that detect 30% more high-risk pregnancies than standard practice because the market believes something better is not only possible but overdue.
The question is no longer whether AI will play a role in prenatal care. It's how quickly hospitals adopt, how regulators adapt, and whether the gains measured in controlled studies hold up when the technology meets the chaotic realities of overburdened labor and delivery units across diverse populations. The research base is growing. The clearances are accumulating. The clinical pain points are acute.
What remains to be seen is whether this convergence of technology, policy, and clinical need can finally close the detection gap that has frustrated obstetricians for decades—or whether it becomes another promising technology that looks better on paper than in practice. Tolsgaard and his colleagues are betting on the former. Given the stakes, the rest of us should hope they're right.
