An ML Mobile App for Shoulder Implant Identification

The challenge

Identifying which manufacturer made a shoulder implant is a real and unglamorous problem. Faced with an image of an existing implant, working out whose it is can mean consulting reference material or relying on experience — and the answer matters, because different manufacturers' components are not interchangeable.

The goal was to put that identification in a phone: capture or upload an image, and get a classification across four manufacturers — Cofield, DePuy, Tornier and Zimmer.

What we built

We worked as the client's embedded IT team rather than as an outside vendor delivering a fixed scope — building alongside them, with their domain knowledge driving what the product needed to be.

• A classification model distinguishing four implant manufacturers from a supplied image.
• iOS and Android apps — capture an image or upload an existing one, get a result on the device the user already carries.
• A REST API on AWS — the backend the apps talk to, handling inference and the surrounding infrastructure.

About that 87.5%

The model classifies correctly around seven times in eight.

Against a four-way choice, where guessing gets you 25%, that is real signal. It is also, plainly, wrong about one image in eight — and both halves of that sentence are true at once.

We publish the figure rather than calling it "high accuracy" because a number a reader can evaluate is worth more than an adjective they cannot. Any engineer assessing whether to work with us can decide for themselves what 87.5% on four classes is worth.

What a given accuracy level is fit for — and any clinical positioning or regulatory pathway that follows from it — is the product owner's determination, not a software vendor's. We built and measured; the client owns what it becomes.

The results

The application was delivered and handed over in full, and our engagement ended there.

That is how a build-and-hand-over should finish. The client owns the model, the apps and the infrastructure, and needs nobody's permission — including ours — to run, change or commercialise it. A clean exit is a feature of the engagement, not the end of a relationship gone wrong.

Working this way

Not every engagement is a fixed-scope project. Sometimes what a company needs is engineering capacity inside their team, working to their domain expertise — particularly where the domain is specialised enough that the knowledge has to come from the client's side.

We do both. If you need custom software or an AI capability built alongside your team rather than handed over a wall, get a free consultation.