ML Medical Image Classification · November 2022
physicAIn
An end-to-end ML-powered system integrating an Android client with a cloud inference API for near real-time diagnostic predictions.
10,000 images
Training set
95%
Accuracy
MLH 3rd Place
Award
Demo
What it does
- Takes a medical image from an Android phone and returns a diagnostic prediction in near real time, pairing the mobile app with a cloud inference service.
- Makes its predictions with a deep-learning model trained on 10,000 images that reaches 95% accuracy on its test set.
- Serves that model as a Flask microservice on a Google Cloud VPS, which the Android client calls over the network.
- Built in a weekend at BCHACKS 4.0, where it won MLH 3rd Place Overall and the Best Use of GitHub award.