The Problem
Reading CT scans takes trained specialists and time. I wanted to see how far a lightweight model on a mobile-friendly stack could go as a first-pass screening aid. It was built as a learning project, not a diagnostic tool.
What I Built
- Model: a MobileNetV2-based image classifier trained with MediaPipe Model Maker on lung CT images. MobileNetV2 is small enough to serve quickly without a heavy GPU.
- API: a Flask REST service that handles image preprocessing, queues requests, and returns predictions in real time.
- App: a cross-platform React Native (Expo) front end where a user uploads a scan and sees the result clearly.
Results
The classifier reached 87% accuracy on held-out CT scans, and the full loop runs end to end: upload, preprocessing, prediction, display.
Screenshots
Screenshots pending.