In silico, bench, ex vivo, in vivo: orchestrating medtech prototyping

A medtech engineer can now produce a functional prototype in 48 hours. Print a complex geometry in biocompatible resin, mould a small series in low-pressure silicone, run a CFD simulation without a supercomputer, generate topology-optimised variants, integrate a standardised handle rather than designing one from scratch. What used to require a factory ten years ago now fits in a small workshop.

This generational leap concerns the tools, the ones we use to design and to build. CAD, generative AI, simulation, 3D printing, low-pressure moulding, off-the-shelf components: they're faster, more accessible, cheaper and more relevant than ten years ago. A small team can now, frugally, get started and produce results that convince partners.

What hasn't changed are the tests, the ones that validate. Mechanics, physics, ergonomics, and above all patient safety and the regulatory framework that protects it: these realities haven't moved. Going faster and cheaper upstream isn't enough. You need to intelligently orchestrate tools and tests to move forward efficiently, without major rollbacks.

The real gain isn't speed, it's precision

The classic trap is to assume you're moving faster because you're prototyping faster. Rarely true. The real gain of this new generation of tools is that you can finally isolate one variable at a time. Test the effect of a bend radius on catheter pushability without rebuilding the whole instrument. Check biocompatibility without committing to a pre-series batch. Compare three distal-tip geometries in parallel.

That granularity of questioning changes everything, provided you know how to use it, and provided you send the question to the right test.

Four tests, four questions, four truths

A medtech project today draws on a palette of four complementary tests. Three are classic, one is (fairly) new. Each has its own area of expertise, and each answers a different question.

In silico (CFD, FEA, transient simulation, digital twins, in silico trials, generative AI) is the newcomer of the palette. It answers "is it plausible?". It helps you project the design forward, size it, make early trade-offs. You explore the solution space, predict orders of magnitude, eliminate dead ends. Fast and at near-zero marginal cost.

The bench answers "will it hold?". It confronts the geometry with the reality of tolerances, materials, cycles. Kinematics come alive, you leave the model for the material, the screen for the tangible feel. You produce data, verify, modify, improve.

Ex vivo (tissues, isolated organs, biological models) answers "how does it interact with anatomy?". You meet real friction, the succession of curves, the narrowness of the path, the fragility of tissues. You discover the anatomical surprises, a calcified artery doesn't behave like its mesh, and no simulation had factored that in.

In vivo answers "how does it behave inside a complete, living system?". Ergonomics, physiological conditions, motion, duration, temperature, the subject's resistance to the procedure. Real pressures, access difficulties, working under fluoroscopy. It's the final arbiter, in the operating room, you find out whether months of iteration led to the right solutions.

Well orchestrated, these four tests chain together and reinforce one another: what you learn at step N feeds the precision of step N+1. Poorly orchestrated, they leave blind spots and late-discovered contingencies that force you to backtrack and that drain morale and budgets. The closer you get to design freeze, the more expensive each rollback becomes.

The sequence isn't a linear checklist either. Sometimes you have to run an animal study between bench testing, just to clear an early anatomical risk. Sometimes an ex vivo surprise sends you back into simulation. The real discipline holds in a single question, repeated at every iteration: which test, with which tool, at which moment, for which question?

A lesson at 37 °C

On a recent project, an in vivo trial taught us something no thermal simulation had anticipated. At 37 °C the device shaft lost enough rigidity to coil on itself inside the aorta, instead of taking the curve we expected. The material had behaved exactly as predicted on the bench, at 23 °C. That's precisely why we added the temperature factor to our shaft stiffness calculator. (something we could have predicted thanks to our shaft stiffness calculator if it had existed earlier).

This kind of discovery isn't a failure. It's the very reason you run the tests. The point is that what you learn moves you forward rather than backward. And the earlier this learning shows up in the sequence, the more valuable it is.

The new key skill: assigning the right question to the right test

This also means distinguishing three objects we lazily lump together under "prototype":

• The POC is an object, sometimes a quick build, that confirms the plausibility of the idea. Its sole mission is to lift a conceptual uncertainty.

• The iteration prototype answers "what happens if I change X?". It's designed to isolate variables, understand their interactions, and push the system toward the best possible instrument.

• The representative prototype (close to design freeze) answers "will the industrialisable version hold its promises?".

Conflating the three, or skipping one, costs months. A POC that works has never validated industrialisation. An iteration prototype that fails has never condemned the concept.

Non deus ex silico: what "in silico" can and cannot do

In silico is probably the most striking innovation of the past decade for medtech prototyping. CFD, FEA, transient simulation, digital twins, in silico trials, generative AI: the field of what you can explore before machining anything has expanded spectacularly. The FDA itself now integrates modelling and simulation into regulatory submissions, in silico studies can reduce the scope of clinical studies, optimise their design, and strengthen safety-margin analyses.

A genuine revolution. But non deus ex silico: simulation isn't a god descending from the silicon to solve your problems for you. It predicts what you ask it to look at. It does not predict what you don't know to search for. A thermal model that doesn't account for shaft softening at 37 °C won't catch it. An arterial mesh that hasn't included calcification won't catch it either. The power of simulation is proportional to the quality of the questions you ask it, and its blind spots mirror the ones in your hypotheses.

This is why the sequence in silico → bench → ex vivo → in vivo isn't a legacy to bypass, but a discipline to optimise. In silico doesn't replace the other tests: it prepares them. And better prepared, they deliver their answers faster.

What this changes in practice

The medtech project that succeeds today isn't the one that lines up the most prototypes. It's the one that knows, before each iteration, which question it answers, with which test to validate it, with which tool to build it, and what deliverable to expect.

The new tools are fabulous. They save us time, money, and a great deal of relevance upstream. What hasn't changed is the demand of the real, biomechanics is a science, physiology is full of challenges. And what hasn't changed either is what separates a project that ships from one that stalls: methodology, rigour, creativity in problem-solving, and respect for the regulatory framework and patient safety. Believing you can cut corners just because you prototype faster is a mistake.

Everything plays out in the precise knowledge of the tools and the tests, of their respective limits, and in the discipline of orchestration. That's exactly what we've been bringing to our partners for twenty years in minimally invasive surgery, vascular and cardiac devices.

The more clearly you know what question you're asking at each step, the less you pay for the answers. That's how a good idea becomes a real device, and how it ends up serving the patients who were waiting for it.