Context
Accurate segmentation of the prostate and its anatomical zones – Transitional Zone (TZ), Peripheral Zone (PZ), and the whole gland – from Magnetic Resonance (MR) images is essential for biopsy targeting, radiotherapy planning, and surgical decision-making. However, segmentation remains subjective: even experienced radiologists produce variable contours, particularly at the TZ–PZ interface where boundaries are unclear. This variability directly impacts treatment reproducibility and radiation dose accuracy. Current deep learning methods, often trained on a single “gold standard” generated by consensus or label smoothing, fail to capture inter-observer variability. As a result, models become biased and evaluations limited to accuracy, overlooking reliability and uncertainty, both essential for clinical deployment.
Internship topic
In this context, the MAPS-FM project has three main objectives :
- Embedding multi-annotator information using strategies such as crowd layer, annotator embeddings, mixture of experts, label-distribution learning, Bayesian modeling, or co-teaching.
- Evaluate foundation model backbones versus models trained from scratch, highlighting relative advantages in data efficiency, generalization across scanners and cohorts, and interpretability quality.
- Develop a comprehensive evaluation framework that integrates uncertainty quantification, examines its correlation with inter-annotator disagreement.
This work builds on an existing dataset collected at the University Hospital of Brest, France. The dataset includes multiple expert annotations, providing a rich resource for methodological development. It enables explicit modeling of inter-observer variability and supports the design of new approaches that use this variability to improve uncertainty estimation and interpretability in prostate MR segmentation.
Environnement
The internship will be carried out as part of a bilateral research partnership between IMT Mines Alès and IMT Atlantique, in continuity with an existing collaboration, and will be funded through the IMT call of the Data Analytics & AI community.
Candidate profile
We are looking for a M2 student motivated by image analysis with a particular interest in deep learning applications. A background in biomedical or medical imaging and an experience with Python programming language and Pytorch package are required. Good communication and team working skills are also required as the intern will work in close collaboration between IMT Mines Alès and IMT Atlantique. A good ability to communicate in English as well as a fluent English for reading and writing scientific articles are also required.
Practical information
- 6-month internship starting from February to April 2026
- Location : IMT Mines Alès – 6 avenue de Clavières 30319 Alès Cedex, France
- Advisors : G. Andrade-Miranda (IMT Mines Alès, SyCoIA), P.-H. Conze (IMT Atlantique, LaTIM) and V. Jaouen (IMT Atlantique, LaTIM)
- Applications by email to gustavo.andrade-miranda@mines-ales.fr, pierre-henri.conze@imt-atlantique.fr and vincent.jaouen@imt-atlantique.fr including: full curriculum vitæ, cover letter stating your motivation and fit for this project, latest grade transcripts, optional) recommendation letters or contacts from former teachers/advisors
