Complete description: https://www.lis-lab.fr/wp-content/uploads/2026/10/internship_proposal.pdf
Context
The electrocardiogram (ECG) is a diagnostic tool commonly used in medical practice. A growing number of studies investigate its prognostic value, notably for short- or medium-term mortality risk. Thanks to large databases, several foundation models for the ECG, trained on millions of signals, have been proposed. Recent work by the team has shown that useful and well-generalising representations can be learned from a thousand times fewer ECGs than existing foundation models. One factor has a strong impact on the results: how the 12 leads of the ECG are combined: They are highly correlated, partly because of the acquisition setup (electrode placement), which encourages overfitting. Current methods regularise strongly by processing the leads independently before fusing the results.
Internship Objectives
The internship extends this work along two axes:
- Lead fusion: Propose and evaluate new fusion methods, for instance taking into account the spatial organisation of the leads through graph neural networks and/or attention mechanisms.
- Survival analysis: Study mortality beyond binary classification, using survival analysis methods (e.g.\ DeepSurv, DeepHit), in order to assess the benefit of the proposed methods in this setting.
Depending on the skills and interests of the selected candidate, one of the two axes may be emphasised. The project relies on public cohorts, and will be carried out in coordination with the ongoing PhD thesis in the team on the same topic. Depending on the results, application to medical data from Hôpital Nord in Marseille may be considered, subject to administrative approval. The results may lead to a scientific publication.
Conditions
- Supervisors: Paul Chauchat, Luca Thiebaud (LIS)
- Collaboration: Mustapha Ouladsine (LIS), Stéphane Delliaux (C2VN)
- Location: LIS, Campus Saint-Jérôme, 52 av Escadrille Normandie Niemen, Marseille
- Duration: 5 months, flexible start date, first half of 2027
- Funding: Institut Laënnec (AMU), standard internship stipend
Candidate profile
- Master 2 or final-year engineering student, with a strong background in statistics and/or machine learning, and a keen interest in biomedical applications
- Good programming skills in Python (ideally with experience in PyTorch)
- Prior knowledge of the ECG, or experience in signal processing or survival analysis, is a plus
- Critical thinking about the use gen AI (pros and cons)
Contact / How to apply
Send an e-mail to paul[dot]chauchat[at]lis-lab[dot]fr and luca[dot]thiebaud[at]lis-lab[dot]fr:
- Describe your interest in the topic and how your profile fits it directly in the email;
- Attach your CV and your academic transcripts.
Application deadline: 30 November. Applications will be reviewed on a rolling basis.
