There's a peculiar sort of mental gymnastics that happens when you're trying to thread a needle into a vein you can't see. One hand grips the ultrasound probe, the other holds the catheter. Your eyes are locked on a grayscale slice of tissue—a flat, two-dimensional cross-section—while your brain scrambles to build a three-dimensional map of where that blood vessel actually runs beneath the skin.
Every emergency physician and intensivist knows this dance. Ultrasound guidance has improved the odds considerably; studies have shown first-attempt success rates climbing from under 50% to north of 85% for certain techniques. But the procedure itself remains stubbornly, frustratingly hard to learn. In emergency departments and ICUs around the world, that learning curve translates into missed attempts, patient discomfort, and sometimes, when time matters most, delayed care.
The bottleneck isn't really the technology's fault. Ultrasound works. But asking clinicians to mentally translate flat images into spatial reality while simultaneously advancing a needle through living tissue imposes a cognitive burden that simulation training and repetition only partially solve. It's a question the medical device industry has circled for years, often warily: what if the ultrasound could simply show you the anatomy in three dimensions—the way you actually need to see it?
A Market in Motion
Point-of-care ultrasound has vaulted from niche specialty to clinical mainstream with a speed that surprised even its early advocates. The POCUS market was valued at $4.2 billion in 2025 and is projected to reach $8.4 billion by 2035, expanding at 7.2% annually. A national survey found that 74% of U.S. medical schools now integrate handheld ultrasound devices into their curricula. Pediatric hospital medicine divisions, emergency departments, even mountain rescue teams have woven bedside imaging into standard practice.
For vascular access specifically, the evidence base has moved beyond debate. A randomized controlled trial published in 2025 demonstrated ultrasound guidance achieving first-attempt success rates of 85.7% for peripheral IV placement in pediatrics versus around 32% with traditional landmark techniques—a difference too large to ignore. Systematic reviews consistently show fewer attempts, shorter procedure times, lower complication rates when ultrasound is deployed.
Yet widespread adoption runs headlong into a persistent reality: the technology demands skill acquisition that many clinicians find genuinely challenging. Medical educators speak of "needle tip visualization" and "dynamic guidance" as discrete competencies requiring dedicated training time. Even with simulation-based mastery learning programs—which themselves consume meaningful institutional resources—proficiency takes practice. Tracking a needle through a 2D cross-section, adjusting for oblique angles, avoiding the dreaded posterior wall puncture, all while maintaining sterile technique. It's a rate-limiting step in modern procedural medicine.
The Allure (and Challenge) of 3D
The global 3D ultrasound market has grown substantially, standing at several billion dollars with projections for continued expansion. Most of that growth, historically, has come from obstetrics and cardiac imaging—applications where volumetric reconstruction happens after image acquisition, often on high-end cart systems costing tens of thousands of dollars more than their 2D counterparts.
Procedural guidance represents an entirely different animal. You need real-time visualization. You need portability—a device a resident can grab at 2 a.m. without extensive setup. And ideally, you need a price point that doesn't require convening a hospital capital committee.
The technical barriers have been formidable, perhaps more than the industry initially appreciated. Conventional 3D ultrasound relies on matrix arrays—grids of thousands of transducer elements generating volumetric data but demanding complex electronics, high computational overhead, and significant power consumption. Research platforms illustrate both the capability and the complexity of current approaches: impressive in the laboratory, daunting to commercialize.
Recent years have seen alternative strategies emerge. Row-column arrays, sparse transducer configurations, computational reconstruction using AI models—all have shown promise in academic settings. Studies demonstrate that real-time 3D imaging is technically feasible with novel architectures. But translating research prototypes into FDA-cleared, cost-effective clinical devices? That's proven slow going.
A Durham Bet
Into this landscape comes a wave of startups betting that procedural ultrasound is ripe for disruption. One such company is Lumius, a Durham, North Carolina venture that emerged from research at Duke University and secured a spot in Y Combinator's cohort in 2026.
The technical foundation is an approach called diffractive acoustic tomography, or 3D-DAT, published in Nature Communications in early 2025. Rather than complex matrix arrays, the method employs a standard linear ultrasound probe paired with a diffractive slit—an optical-acoustic component enabling simultaneous 3D ultrasound and photoacoustic imaging. The paper, authored by researchers including Lumius co-founders Tri Vu and Luca Menozzi, demonstrated near-isotropic, large field-of-view volumetric imaging with off-the-shelf components.
Lumius was founded by Vu as CEO, Menozzi as CTO, and Chenhang Li as CFO/COO, with scientific advisor Junjie Yao. The company is targeting vascular access as its initial application. The pitch: affordable, portable, real-time 3D ultrasound that removes the mental reconstruction burden and accelerates the learning curve. Company materials have suggested existing 3D devices cost more than ten times their 2D equivalents; Lumius aims to narrow that gap substantially.
There's an important caveat worth stating plainly. The company's materials have explicitly noted that its products have not been evaluated by the FDA and are not intended for medical use at their current stage. The device remains pre-regulatory, a research prototype advancing through translational funding channels. Lumius has received funding from the NC Biotechnology Center and Duke-Coulter Translational Partnership, among other sources. Y Combinator's standard investment came with acceptance to the accelerator program.
The company operates with a small team from Durham, leveraging proximity to Duke's biomedical engineering ecosystem. Whether 3D-DAT technology can navigate FDA clearance, demonstrate clinical superiority in controlled trials, and achieve manufacturing scale at the claimed price points remains very much an open question.
The AI Angle
Lumius isn't the only player rethinking procedural ultrasound, and the competitive landscape has gotten considerably more crowded. Established manufacturers have begun layering artificial intelligence onto traditional 2D platforms. FUJIFILM Sonosite launched PIV Assist, an AI feature helping plan peripheral IV insertion. Clarius secured FDA clearance for Median Nerve AI targeting carpal tunnel diagnosis and has reported reaching profitability. Butterfly Network, whose semiconductor-based handheld devices and cloud AI platform have attracted significant attention, gained FDA clearance for a blind-sweep gestational age tool—reportedly the first such approval for automated acquisition guidance.
These AI enhancements, however, still operate within 2D frameworks. They improve vein identification, measure depth, suggest insertion angles. But they don't eliminate the core challenge of spatial translation—that mental leap from screen to tissue.
True volumetric guidance systems exist, though mostly in specialized contexts. QT Imaging, which has developed 3D transmission ultrasound for breast imaging, secured a new Category III CPT code—a signal that payers and coding bodies recognize novel 3D ultrasound modalities, even if reimbursement pathways remain uncertain. iSono Health's ATUSA wearable 3D breast ultrasound received FDA clearance and has been working toward broader commercialization.
For vascular access specifically, some research groups have demonstrated 3D needle tracking prototypes using photoacoustic beacons or fiber-optic sensors. Studies have shown that even brief training with real-time 3D guidance can reduce procedural failures—evidence that volumetric visualization may indeed compress the learning curve. None of these research systems, though, have reached widespread clinical deployment. The gap between laboratory prototype and hospital bedside remains stubbornly wide.
Regulatory Realities
The pathway for new ultrasound devices is well-trodden but hardly trivial. Most diagnostic ultrasound systems require FDA 510(k) clearance as Class II devices. Manufacturers must demonstrate substantial equivalence to predicate devices and address acoustic output safety limits—straightforward in principle, time-consuming in practice.
For AI-enabled features, the landscape shifted when FDA finalized guidance on Predetermined Change Control Plans—a framework allowing manufacturers to define and pre-clear certain algorithmic updates without resubmission. This matters for any procedural guidance system planning to incorporate machine learning for anatomy segmentation, needle trajectory prediction, or acquisition assistance. The PCCP pathway introduces lifecycle obligations but also creates competitive advantage for companies building AI natively into their regulatory strategy from the outset.
Reimbursement is more straightforward, at least for the procedure itself. CPT code 76937, an add-on for ultrasound guidance in vascular access, is widely recognized by payers. Documentation requirements are clear: real-time imaging performed personally by the clinician with permanent recording. Whether 3D imaging justifies differential reimbursement—or merely represents a better way to accomplish the same billable service—remains an open question.
Then there's the practical matter of hospital procurement. Healthcare systems increasingly demand integration with existing PACS infrastructure, support for institutional training protocols, and audit trails for compliance. Startups promising novel imaging modalities must prove not only clinical value but also compatibility with IT ecosystems designed around decades-old DICOM standards. That's a harder sell than it sounds.
The Training Question
Perhaps the most compelling argument for 3D procedural ultrasound isn't that it's better for experts—though it may well be—but that it could democratize expertise. Studies confirm that simulation-based mastery learning reduces time to competency for ultrasound-guided vascular access. Low-cost phantoms can be produced affordably, but training programs still require faculty time, curriculum development, repeated practice sessions. Even asynchronous or online training modules demand hours per learner.
Research on biplane imaging hints at what's possible. Studies with novice operators have shown that dual-plane visualization reduces needle redirections and posterior wall punctures compared to single-plane guidance. If two simultaneous 2D views help, the reasoning goes, a true 3D representation should help more.
Some research prototypes bear this out. Participants could visualize target anatomy and needle trajectory simultaneously rather than inferring position from slice-by-slice images. Whether such advantages translate to commercial devices in real clinical environments—with real patients, real time pressures, real consequences—is precisely what companies like Lumius will need to prove.
What Happens Next
The next couple of years will likely determine whether affordable, real-time 3D ultrasound for procedures moves from concept to clinical reality. Several parallel trends suggest the timing may be right, though perhaps that's always how it looks before a technology wave either crests or breaks.
First, the AI infrastructure is maturing. FDA approvals for acquisition-guidance algorithms operating without expert sonographers have established regulatory precedent that could extend to procedural planning and anatomy auto-segmentation. As AI models grow more sophisticated at recognizing vessels, nerves, and anatomic landmarks from volumetric data, the value proposition of 3D guidance strengthens—at least in theory.
Second, hardware platforms are catching up. Announcements of 3D research tools, while aimed at academic users, signal that underlying signal processing and computational architectures have matured. Matrix arrays, row-column arrays, and other novel transducer configurations are moving from one-off experiments to reproducible, potentially manufacturable designs.
Third, clinical demand persists. Despite training efforts, vascular access complications haven't disappeared. CDC alerts about contaminated ultrasound gel causing infections have underscored the need for better protocols—and perhaps, implicitly, for technology that makes procedures faster and less prone to protocol violations when time is short.
The market will ultimately decide whether the benefits of volumetric guidance justify the incremental complexity and cost. For high-stakes procedures—central venous catheters, arterial lines, regional anesthesia blocks—the adoption threshold is lower than for routine peripheral IVs. If 3D imaging genuinely cuts training time or reduces complication rates by meaningful percentages, the return on investment becomes defensible.
Lumius, with its Y Combinator pedigree and academic research foundation, represents a bet that the technology has reached an inflection point. The company must still navigate regulatory clearance, validate performance in clinical trials, establish manufacturing partnerships, and convince hospital procurement committees that the device delivers on its promise. Those aren't trivial gates, and plenty of well-funded medical device startups have stumbled at each one.
But the broader trajectory—from cart-based research systems to portable AI-assisted devices to, potentially, affordable real-time 3D guidance at the bedside—feels increasingly plausible. The point-of-care ultrasound market will continue growing regardless. The question is whether the next generation of procedural ultrasound will simply be smaller, smarter 2D devices, or whether volumetric imaging finally becomes accessible enough to change how vascular access is taught and performed.
For clinicians still struggling with mental 3D reconstruction in the middle of the night, the answer matters. The technology may not be ready yet. But after years of false starts and incremental progress, it's getting closer.
