Location: Clermont-Ferrand, France
Institution: University Hospital of Clermont-Ferrand (CHU de Clermont-Ferrand)
Research group: DIA2M (Department of AI Research and Medical Applications), jointly organised with EnCoV (CNRS / Université de Clermont Auvergne)
Duration: 12 months with possibility of extension
Starting date: Flexible
Project: France 2030 PEPR Women’s health https://pepr-sante-femmes-et-couples.fr/endometriose/
Project Overview
We are seeking a highly motivated Postdoctoral Researcher to join our interdisciplinary team working at the intersection of computer vision, machine learning, and minimally invasive surgery.
The project focuses on the automatic analysis (eg, segmentation) of medical images, with application to endometriosis lesions seen in laparoscopic images. The long-term objective is developing general AI methods and applying them for AI-assistance tools to support surgeons in recognising and localising lesions during surgery.
The project is conducted in collaboration with the LTSI Inserm laboratory in Rennes. A large annotated clinical dataset has already been collected in a previous project, and additional data will be acquired throughout the project.
Scientific Background
Deep learning methods have achieved remarkable performance in medical image segmentation. However, in many clinical applications, obtaining reliable annotations is challenging. Expert annotations often exhibit significant variability due to differences in interpretation, uncertainty regarding lesion boundaries, and inherent ambiguities in medical images.
The primary research objective is to make methodological advances in learning from imperfect supervision by developing novel deep learning strategies that explicitly account for annotation uncertainty and noisy labels. While these methods are intended to be broadly applicable across medical imaging, they will be developed and validated using clinically relevant imaging datasets, with a particular focus on endometriosis.
Endometriosis provides an ideal test case for this research, as our preliminary analyses reveal substantial inter-expert variability in the annotation of endometriosis lesions. This variability reflects the challenges of imperfect supervision commonly encountered in clinical imaging. Moreover, for a subset of the dataset, consensus annotations are available, providing a unique opportunity to investigate the effects of annotation uncertainty on model training and performance and to evaluate the effectiveness of the proposed methods.
The project builds upon our recent work on learning from noisy labels and uncertainty estimation while extending these ideas to clinically relevant image segmentation problems.
Research Objectives
The successful candidate will contribute to several aspects of the project, including:
- Analysis of inter-expert variability in medical image annotations.
- Development of robust deep learning methods for image analysis, including semantic segmentation.
- Confidence estimation and uncertainty-aware training.
- Evaluation of models using single expert, multi-expert and consensus annotations.
- Publication of research results in leading international conferences and journals.
Depending on the candidate’s interests and the progress of the project, additional research directions in surgical computer vision may also be explored.
Candidate Profile
We are looking for applicants with:
- A PhD in Computer Science, Artificial Intelligence, Computer Vision, Machine Learning, or a closely related field.
- Good communication skills and proficiency in English, both written and spoken.
- Strong programming skills in Python and experience with PyTorch.
- Experience in deep learning and modern optimisation methods.
- Excellent research and scientific writing skills.
- Ability to work independently while collaborating within a multidisciplinary research team.
The following qualifications are considered an advantage:
- Experience in image segmentation.
- Experience with medical image analysis or biomedical AI.
- Research on uncertainty estimation, robust learning, weak supervision, or learning from noisy labels.
- Publications in recognised international conferences or journals.
Research Environment
The postdoctoral researcher will join DIA2M at University Hospital of Clermont-Ferrand.
The project will be supervised by: Navid Rabbani, Yamid Espinel and Adrien Bartoli
The successful candidate will work closely with computer vision researchers, clinicians, and PhD students in a collaborative and internationally recognised research environment.
What We Offer
- A competitive salary.
- A stimulating interdisciplinary research project with direct clinical impact.
- Access to a unique and growing clinical dataset.
- Collaboration with surgeons, clinicians, and AI researchers.
- Opportunities to publish in leading conferences and journals.
- A dynamic and supportive research environment.
Application
Applications should include:
- Curriculum Vitae
- Cover letter describing research interests and motivation
- Publication list
- Contact information for academic references
Applications and any questions regarding the position should be sent to Navid Rabbani . Please use the following subject line when sending your application emails: “InENDO postdoc – Your Name”
