He spent a decade in imaging labs learning to see inside the body. Now the Lumius co-founder is building a 3D ultrasound anyone can pick up - starting with the procedures where half of first tries miss.
Point a phone at a room and you get depth, edges, faces - a scene your eye reads instantly. Point an ultrasound probe at a patient and you get a flat gray slice, one plane at a time, and a clinician who has to rebuild the whole three-dimensional picture inside their own head. Tri Vu has spent enough years staring at those slices to find the gap absurd. His company, Lumius, exists to close it.
The pitch is compact enough to fit on a business card: a 3D camera for the body. Instead of thin 2D cross-sections, the operator sees volumetric anatomy in real time, with software that names what it is looking at - this is a vessel, that is a nerve, this is muscle - and suggests what to do next. Vu co-founded Lumius in 2024 with two friends from graduate school, and by the spring of 2026 the company had a spot in Y Combinator and a launch that described ultrasound, one of medicine's oldest imaging tools, as something closer to a camera you already know how to use.
Ultrasound is cheap, portable, and radiation-free, which is exactly why it is everywhere. Roughly 500 million scans happen every year. The catch is that almost all of them are 2D. The machine hands the operator a flat slice, and the operator supplies the missing dimension from memory and training. That works beautifully in expert hands. It works much less well when the hands are new.
Vascular access is where the difficulty shows up most plainly. Placing a central line means threading a catheter into a deep vein, guided by a live ultrasound image. The vein, the artery next to it, and the nerve running alongside all look like variations on gray on a 2D screen. Get the mental geometry wrong and the needle misses. For newer clinicians, roughly half of first attempts do. Each miss means another stick, more time, more risk, and a patient who feels every bit of the learning curve.
3D ultrasound is not science fiction. It already exists for cardiac and fetal imaging. But those systems cost about ten times what a standard 2D machine does, which keeps volumetric imaging locked inside specialties and big hospitals. Lumius is betting that the reason ordinary procedures still run on 2D is not that clinicians prefer it - it is that the affordable, portable, easy version was never built.
Look at the numbers side by side and the opportunity gets sharper. A skill that half of newcomers fail on their first try is not a skill problem so much as an interface problem. The information the clinician needs - where the vein sits relative to the needle in space - exists in the tissue. It is simply not being shown to them. Every mental reconstruction is a place where the process can break, and the current tool asks clinicians to do that reconstruction hundreds of times a shift, under time pressure, on real patients.
Vu did not arrive at this problem from a business school case study. He arrived from a lab bench. He earned his bachelor's in biomedical engineering at the University at Buffalo, then went to Duke for a master's in 2022 and a PhD in 2024, working in photoacoustic and ultrasound imaging - a field that uses light and sound together to see inside living tissue. His research pushed on the hardest version of the imaging problem: capturing fast, high-resolution 3D pictures of biology in motion.
The output was serious. His work has been cited more than 1,900 times, with an h-index of 21, and appears in journals like Nature Communications, Science, IEEE Transactions on Medical Imaging, and Light: Science & Applications. One project he contributed to imaged blood flow and oxygen levels across an entire mouse brain in real time, resolving both individual vessels and the whole organ at once. This is the kind of imaging depth most engineers spend a career chasing.
His publication record has a telling quirk. His single most-cited paper is not about ultrasound at all - it is a 2017 review of wearable food-intake monitoring technologies, a survey that has been leaned on more than 200 times. Then the work turns squarely toward imaging: real-time whole-brain photoacoustic microscopy, deep-learning methods for reconstructing undersampled scans, artifact removal with generative models. Read in order, it traces an engineer moving from cataloging what sensors can do toward making imaging systems faster, cheaper, and smarter - exactly the arc a portable 3D ultrasound would need.
He also holds a research appointment as an assistant professor at the University of Oklahoma's Stephenson School of Biomedical Engineering, where his stated focus is imaging technology for preclinical and translational work. Translational is the key word. It is one thing to publish a method that images a mouse brain; it is another to turn imaging into a device a busy clinician will reach for at two in the morning. Lumius is Vu's attempt to make the jump from the first to the second.
Lumius did not start as a business plan. It started as three graduate students who kept talking about the same thing. Vu, Luca Menozzi, and Chenhang Li met during their PhDs at Duke and bonded over ultrasound - the technology, its limits, and what a better version might look like. Menozzi, also a Duke PhD in ultrasound and photoacoustic imaging, became CTO. Li, a Duke PhD with a background in wearable electronics, took on the COO and CFO roles. Vu leads as CEO.
There is a practical wisdom in that origin. Deep-tech companies live or die on whether the founders truly understand the physics they are commercializing, and imaging is unforgiving - the difference between a clean 3D reconstruction and a smear of noise is buried in signal processing that takes years to internalize. A team assembled from a job board would have to learn to trust each other while also learning to trust the science. Vu's team had already done both, in the same building, on the same instruments.
The division of labor maps cleanly onto the problem. You need someone who understands the imaging physics deeply enough to make 3D fast and cheap. You need someone who has built compact, wearable hardware and can think about manufacturing and cost. And you need someone to hold the whole thing together and point it at a market. Three friends who spent years in the same lab already knew they could work through hard problems without the relationship cracking - which is often the part that actually kills early companies.
The smartest thing about the Lumius plan may be its restraint. A universal 3D imaging platform for the whole body is a vision, not a product. So the company is starting with one procedure where the pain is sharp, measurable, and constant: vascular access and central line placement. If a 3D view with automatic vessel detection can push that 50% first-attempt failure rate down, the value is obvious to anyone who has ever missed a vein or been on the receiving end of a second stick.
From that beachhead, the roadmap widens in deliberate steps - blood clot detection, nerve block imaging, tumor diagnosis, biopsy guidance. Each is a place where seeing the true 3D geometry, rather than inferring it, changes the odds. The through-line is consistency: turn a skill that today depends heavily on the operator's experience into something a device can help carry.
Starting narrow also solves a founder's hardest early question: who exactly is the first customer. By naming vascular access, Lumius knows precisely who to talk to - clinicians and hospital innovation teams who do central lines, medical device people who think about 3D guidance, and anyone who can advise on the hardware and regulatory path. That specificity is worth more than a broad vision at this stage. It turns an abstract camera for the body into a concrete answer to a question a nurse asks every shift: is the vein where I think it is.
The AI layer is where the framing of camera really earns its keep. A camera is not just a lens; it is a lens plus software that knows a face is a face. Lumius describes a system that automatically detects blood vessels, nerves, and muscle, then explains what the operator is seeing and suggests the next move. The goal is not to replace the clinician's judgment but to shorten the distance between a novice and a confident scan - to make the machine legible instead of cryptic.
There is a quiet audacity in describing a 70-year-old technology as a camera for the body. Ultrasound has a reputation - useful, ubiquitous, and hard to master. Calling it a camera resets expectations. Cameras are things everyone can operate. They give you a full picture, not a slice you have to interpret. They come with software that does the understanding for you. That single word carries the whole ambition of the company: not a marginally better scanner, but a different relationship between a person and the image.
It is also a tell about how Vu thinks. The academic version of his work is measured in citations and journal placements, and by that measure he has already succeeded. The founder version is measured in something blunter - whether a clinician who could not reliably hit a vein last month can hit it this month. Those are different scoreboards, and choosing the second one, while keeping a foot in the first, is the whole story of what he is doing now.
Lumius is early. It carries the standard disclaimer that its statements and products have not been evaluated by regulators and are not yet intended for medical use, and the path through clinical validation and FDA clearance is long for any medical device. But the founding logic is unusually clean: a real, quantified problem; a team that understands the underlying physics at a research level; and a wedge narrow enough to prove and broad enough to matter. Vu spent a decade learning to see inside the body. The company is his argument that everyone else should get to see it too.