AI · 2023 · Built for Luton and Dunstable University Hospital
AI-Powered Orthopedic Implant Identification
A surgeon is about to open this knee. What is already inside it?
Web platform where doctors upload an X-ray and a CNN identifies the orthopaedic implant by joint, so the model is known before revision surgery. Seven models on AWS, in use at Luton and Dunstable University Hospital and by orthopaedic surgeons worldwide.
Problem
Before revision surgery a surgeon needs to know which implant is already in the joint. Reading it off an X-ray by eye means catalogues, colleagues and guesswork.
Approach
- A web app where a doctor uploads the X-ray and picks the joint; the matching model returns the implant.
- Seven CNN models specialised by joint, covering knee, ankle, shoulder, wrist and elbow, trained and tested for classification accuracy and served from a FastAPI backend on AWS.
- A Next.js front end built from reusable Mantine UI components, with two-factor authentication and JWT sessions.
- Supabase records user activity and uploads for later analytics.
Stack
- FastAPI
- PyTorch
- Next.js
- AWS
- Supabase
Outcome
- In use at Luton and Dunstable University Hospital
- Used by orthopaedic surgeons worldwide
- Seven deployed deep learning models