Reliable deep learning for medical image segmentation
Segmentation methods that stay dependable across different data sources and limited annotations.
Background
Segmentation models often lose accuracy when data come from different scanners or sites, or when only a few annotated examples are available.
Research questions
- How can segmentation stay dependable across data sources?
- How can models learn well from limited annotations?
Objectives
- Study the robustness of segmentation across data sources
- Reduce the need for annotated data
Approach
Deep learning segmentation with a focus on robustness and limited supervision.
Related experience
- Medical image segmentation with U-Net
- 3D point cloud semantic segmentation (B.Sc. thesis)
The full proposal is shared privately on request. Request the full proposal · All proposals