Published on 29 Jul 2026
min read
Views
Every year, a billion diagnostic scans are read somewhere in the world. Most tell a story the eye can follow. A stubborn minority hides the finding that changes a life, in a shadow of a millimeter wide, in a branch of an airway the scope has never seen. Medical imaging is the backbone of modern healthcare, a market that touched USD 42 billion in 2024 and is on course to cross USD 68 billion by 2032. Yet the deeper truth of this decade is simpler. The image is no longer the bottleneck. What we do with it is.
That is why I keep coming back to one idea. The diagnostic twin.
The Real Gap in MedTech Today
AI in MedTech is no longer experimental. It is existential. Aging populations, overstretched radiologists, and datasets that grow faster than the humans who read them have pushed our industry past the point where a better model, alone, is the answer. The global surgical imaging market is projected to move from USD 5.76 billion in 2025 to USD 9.60 billion by 2034, and hospitals in the United States alone spend over USD 5 billion a year on surgical imaging equipment. The spending is rising. The intelligence gap is rising faster.
I have watched a bronchoscope thread a thousand airways toward a nodule about the width of a grain of rice, with limited spatial context and vessels it cannot see. That gap, between what the scan shows and what the device needs to know, is where the next generation of MedTech will be won or lost.
From Static Image to Living Twin
A digital twin is not a 3D reconstruction with a marketing skin. Done properly, it is a patient-specific model built from a single CT scan that evolves through four stages of intelligence: structural, diagnostic, procedural, and behavioral. Each stage adds a layer a device or clinician can act on, moving the scan from a static picture to a working reference, as approaches like LTTS' Lung Digital Twin | LTTS illustrate.
The structural layer maps every airway, vessel, and lobe automatically, over 90 percent AI-automated, no manual annotation. The diagnostic layer finds and localizes tumors, fibrosis, emphysema, obstruction, and shows how it reached each conclusion. The procedural layer turns the scan into a route before the scope moves, with vessel-proximity mapping and branch-by-branch guidance. The behavioral layer, next on our roadmap, simulates airflow, disease progression, and what-if scenarios before a decision reaches the patient.
One CT. Four layers of intelligence. A single reusable pipeline that already extends across the lung, liver, heart, brain, and eye.
Why the Model Race Is the Wrong Race
Our industry has spent five years chasing higher accuracy scores. That effort was not wasted. It was insufficient. The bottleneck was never modeled. It was the discipline of turning a clinician's question into a design spec, then turning that spec into a system a regulator will approve, a device can run inside its own housing, and a hospital IT team can support on Tuesday morning.
That discipline has a name. We call it Engineering Intelligence.
EI is the discipline of engineering intelligence into everything we build, and everything we build with. It understands engineering complexity and turns it into intelligence at a scale. It shows up in four forms, all four visible in a diagnostic twin: Engineering AI that reasons for physical systems, Agentic AI that acts, Physical AI that runs inside the device, and Industrial AI that runs across the fleet.
Applied to a twin, EI is the reason a segmentation output becomes a navigable route. It is the reason an inference becomes an explanation a clinician will trust, and a reviewer will approve. It is the reason the same architecture that maps a lung today maps a liver next quarter without starting over.
The Data Problem No One Solved with More Data
Medical AI has been starving at the right data since day one. Our response, as an industry, has been to ask for more of it. Bigger cohorts. Cleaner labels. More institutions. It has not closed the gap.
You cannot annotate your way to organ-level intelligence. You must engineer around the scarcity. Over 90 percent automated segmentation. Multi-source clinical training data. Explainability designed into the model, not stapled afterwards. These are engineering decisions before they are data decisions. They are the reason the twin is robust in the real world, not just accurate on a benchmark.
Why Engineering Heritage Matters
There is a kind of credibility in this field you cannot fake. It comes from having built the plants that make the devices, the software that runs the devices, and now the intelligence that sits inside them. It is the reason a MedTech CTO trusts a partner to co-own the intelligence layer of a device that will sit inside a patient. It is the reason our navigation method is patented. It is the reason NVIDIA chose our platform to demonstrate what accelerated AI infrastructure can do for respiratory diagnostics at RSNA 2025.
Most AI companies I meet arrived at MedTech from software. We arrived in software engineering. In this industry, that order matters.
One Organ Today, Five on the Roadmap
The lungs are one of the most mature environments for diagnostic twin deployment today, but the principles extend well beyond respiratory care, into liver disease, cardiovascular diagnostics, neurology, and ophthalmology.
What makes the concept compelling is not modeling a single organ, but the possibility of a reusable engineering framework that supports many clinical applications. By combining patient-specific imaging, AI-driven interpretation, procedural guidance, and simulation within a common architecture, diagnostic twins may offer a foundation that evolves alongside future diagnostic and therapeutic innovation, raising a strategic question for MedTech companies: build individual AI capabilities for each domain, or invest in scalable platforms that support a broader portfolio over time.
The Stitch in Time
A stitch in time saves nine, and in MedTech it can mean faster time to market, lower compliance cost, and better patient outcomes. The organizations likely to shape the next era of healthcare are the ones acting now: integrating AI across the full product lifecycle rather than in isolated pilots, aligning it with clinical workflow and regulation from day one, and choosing engineering partners equipped to work at the level of the device itself.
The next differentiator may not be the sensor or the model alone, but the diagnostic twin between them, guiding the device and giving the clinician the reasoning behind each step. Ultimately, the industry faces a strategic choice: keep optimizing for accuracy alone or invest in the broader discipline of engineering intelligence that turns imaging data into clinical insight.
