Laboratoire de Traitement et Communication de l'Information
Role details
Job location
Tech stack
Job description
École doctorale : Ecole Doctorale de l'Institut Polytechnique de Paris Laboratoire de recherche : Laboratoire de Traitement et Communication de l'Information, This project turns towards reinforcement learning and generative AI to improve the performance of deep learning models of semantic segmentation of medical images under the usual constraints that (1) anatomical knowledge must be captured from limited numbers of annotations (data scarcity), and (2) segmentation is performed in challenging contexts: noisy images with artefacts, loss of contrast, shadows, a variety of sequences/modalities and complex anatomical structures. Diverse approaches have been proposed to address these challenges: (1) for anatomy-guided segmentation: specialized topologic or geometric losses for specific use cases [18-23], shape losses [24], shape edition in latent spaces [25]; (2) for few-shot segmentation: domain adaptation strategies or self-supervised learning strategies [26-31]. To contrast, LLM training involves additional reinforcement learning steps beyond self-supervised pre-training [1,2,3], referred to as 'aligning' to user preferences. RL strategies have considerably simplified and improved in this context [4-9]. There is (very) scarce literature in medical imaging adopting these concepts [32], and we aim to address this situation with this project.
The aim of the project is to develop novel deep learning models of medical image segmentation and generation that address two main bottlenecks towards adoption of automated segmentation algorithms by clinicians and industry: anatomically implausible segmentation outputs and data scarcity. We will consider a variety of use cases: cardiac segmentation (including 2D+t echocardiography), vessel segmentation in abdominal imaging, retinal vessel segmentation, brain structures and possibly lung segmentation. See the attached PDF.
Requirements
Idéalement, le candidat sera familier avec le domaine de l'imagerie médicale, du deep learning et des modèles génératifs.Le candidat devra parler et écrire couramment en anglais, y compris pour la présentation de ses travaux scientifiques.Le candidat devra être motivé par un environnement de recherche et posséder de bonnes qualités interpersonnelles. ","identifier":{"@type":"PropertyValue","name":"Institut Polytechnique de Paris Télécom Paris","value":"e33da28cf4e015d4ed63746b33a97d3a"},"url":"https://www.hellowork.com/fr-fr/emplois/76933336.html","datePosted":"2026-03-17T17:49:49Z","directApply":false,"educationRequirements":{"@type":"EducationalOccupationalCredential","credentialCategory":"postgraduate degree"},"employmentType":["TEMPORARY","FULL_TIME"],"experienceRequirements":"no requirements","hiringOrganization":{"@type":"Organization","name":"Institut Polytechnique de Paris Télécom Paris"},"industry":"Service public d'état","jobLocation":{"@type":"Place","address":{"@type":"PostalAddress","addressCountry":"FR","addressLocality":"Paris","addressRegion":"Île-de-France","postalCode":"75000"}},"occupationalCategory":"Audiovisuel","qualifications":"Idéalement, le candidat sera familier avec le domaine de l'imagerie médicale, du deep learning et des modèles génératifs. Le candidat devra parler et écrire couramment en anglais, y compris pour la présentation de ses travaux scientifiques. Le candidat devra être motivé par un environnement de recherche et posséder de bonnes qualités