hackaton 2025
I recently took part in a hackathon at work. Here’s the idea I presented. I called my project Dentaface.
The goal of this project is to automate the detection of dental cavities with X-ray images or pictures using a deep learning model (Detectron2 RetinaNet) and expose it via a FastAPI-based backend.
It enables real-time caries detection through image uploads, combined with location-aware cost estimation based on geocoding.
As far as we know, this product has no competitor.
The application is managed using two Docker images, one for the backend-training and one for the frontend, making it easily migratable to Amazon ECS.
—————————————
Phase 1: Training RetinaNet
- Train RetinaNet using open-source datasets such as Roboflow and Mendeley
- Model is optimized for single class detection:
carie
The first phase of the project consisted in training the model on publicly available datasets, namely Mendeley and Roboflow. The fine-tuning involved adjusting the parameters and allowing the model to train over time.
Phase 2: User Interaction
- Upload X-ray images through the website
- Enter postal address
- If caries are detected:
- Receive a list of nearby dentists (via OpenStreetMap)
- Receive treatment quotes
The second phase involved providing a backend application and a frontend interface that allow users to upload their X-ray images and, if needed, receive treatment quotes based on the geolocation of both the user and nearby dentists.