17/06/2025
We are working on an applied research project around a crucial issue: the automatic prediction of information about war injuries from 3D scans of the human body. Matthieu Brossard Camille Aubry Jay Asodariya Pierre Clément
The project :
Thanks to 3D imaging of the human body, we can now visualize injuries from all angles. But there is still a lot of work to do to automatically identify the characteristics of the injured areas from these 3D models. It’s a challenge that is still little explored.
Our first objective: automatically segment wounds from 3D files in Wavefront format (.obj, .mtl, .png), to facilitate medical analysis.
The goal
- Accelerate medical evaluation of injuries in urgent contexts.
- Provide a diagnostic aid tool based on artificial intelligence.
- Provide an interactive demo of the project, accessible via the web.
How does it work?
We set up a method based on 2D projections of the 3D model:
- The model is projected from different angles to create 2D views.
- These views are analyzed by DeepSkin (https://github.com/Nico-Curti/Deepskin), a neural network specialized in wound segmentation.
- The results are reprojected in 3D to mark the injured areas.
An interactive web interface has been developed to test this approach:
- The user uploads a 3D model from their computer or chooses a pre-loaded one.
- They can visualize and compare the original model and the segmented model directly in the browser.
This project shows that a bridge is possible between medicine, AI, and 3D visualization. We will continue to improve it, with a more intuitive interface and finer segmentation.
tags :