RESEARCHER PROFILE: Postdoc / R2: PhD holders
RESEARCH FIELD(S)1: Mathematics
MAIN SUB RESEARCH FIELD OR DISCIPLINES1: Machine Learning & Cancer / Data Science / Biostatistics
JOB /OFFER DESCRIPTION
The primary objective of the postdoctoral position, “Targeted Deep Learning for Survival Prediction from Histopathological Images,” is to develop Targeted Learning approaches combined with deep learning models to predict patient survival duration from Whole Slide Images (WSIs). The goal is to build robust, interpretable, and statistically valid survival estimators by integrating the theoretical guarantees of Targeted Maximum Likelihood Estimation (TMLE) with the power of deep neural networks to extract complex visual representations.
The project will pay particular attention to managing censored data and potential biases. A key milestone will be adapting Multiple Instance Learning (MIL) techniques to the constraints of targeted learning. This work has the potential to reshape histopathological analysis by providing personalized predictions while adhering to the rigorous requirements of causal statistics.
In the context of survival prediction, the challenge lies in incorporating the time-dependent and censored nature of the data. Unlike binary classification or standard regression, survival modeling requires estimating a hazard function or survival probability while accounting for patients whose event (death, recurrence, progression, etc.) is not observed within the study period (censoring).
The MIL model must therefore be adapted to learn from bags (WSIs) with censored survival labels.
This may involve:
- Applying attention-based aggregation on tile embeddings to weigh the most informative regions (e.g., ABMIL [1])
- Using neural survival models (e.g., DeepSurv [2], Cox-Time [3]) suited for censored data
- Integrating the TMLE component of Targeted Learning to produce efficient and consistent estimators, particularly in scenarios with confounding treatment or covariates ([4] and [5])
This approach will help identify which histological and morphological patterns within WSIs are associated with better or worse survival, leveraging the richness of visual data while complying with the statistical constraints of survival data.
Working Environment
The Laboratory:
The Institut de Mathématiques de Marseille (I2M) is jointly operated by the CNRS, Aix-Marseille University (AMU), and Centrale Marseille. It comprises around 130 faculty members, 30 CNRS researchers, 15 technical and administrative staff, 60 PhD students, and 20 postdoctoral researchers. I2M includes five research groups spanning a wide range of pure and applied mathematics (analysis, geometry, topology, logic, number theory, dynamics, combinatorics, probability, statistics, applied analysis, etc.), as well as many scientific and industrial applications. The postdoctoral researcher will be affiliated with the ALEA group, within the statistics team.
The Company:
DiaDeep is a medtech company based in Lyon, founded in 2022 by two pathologists and two university professors specialized in artificial intelligence (AI). Its core mission is to improve cancer diagnostic accuracy by developing AI-based solutions. DiaDeep has developed deep learning models capable of analyzing histopathological slides and providing personalized risk assessments for patients with cancers such as breast, colon, and melanoma.
DiaDeep is a member of the Lyonbiopôle competitiveness cluster, reinforcing its presence in the life sciences ecosystem of the Auvergne-Rhône-Alpes region. The company is a France 2030 award winner and has secured funding for this postdoctoral position through this grant.
The recruited individual will work closely with the research team at DiaDeep, which includes 6 researchers, among them 2 university professors. The candidate will also be integrated into the Marseille statistics team and collaborate with the identified project researchers, especially D. Pommeret and L. Reboul as well as international collaborators.
TYPE OF CONTRACT: TEMPORARY / JOB STATUS: FULL TIME
APPLICATION DEADLINE: 31/08/2025
ENVISAGED STARTING DATE: 01/10/2025
ENVISAGED DURATION: 12 months, renewable
WORK LOCATION(S): Institut de Mathématiques de Marseille (I2M), UMR 7373, 3 place Victor Hugo, 13003 Marseille
WHAT WE OFFER: Salary: from 2616 to 3062 euros gross per month, depending on experience.
Additional information: The Euraxess Center of Aix-Marseille Université informs foreign visiting professors, researchers, postdoc and PhD candidates about the administrative steps to be undertaken prior to arrival at AMU and the various practical formalities to be completed once in France: visas and entry requirements, insurance, help finding accommodation, support in opening a bank account, etc. More information on AMU EURAXESS Portal
QUALIFICATIONS, REQUIRED RESEARCH FIELDS, REQUIRED EDUCATION LEVEL, PROFESSIONAL SKILLS, OTHER RESEARCH REQUIREMENTS (Field of qualification, level, years of research experience, etc.)
The candidate must hold a PhD in one of the following fields: biostatistics, statistics, machine learning, bioinformatics, data science, or applied mathematics.
Experience in data processing and analysis is required, along with a solid background in statistical modeling and programming (Python).
REQUESTED DOCUMENTS OF APPLICATION, ELIGIBILITY CRITERIA, SELECTION PROCESS
Academic CV + recommendation letter(s) + the Jury’s report of oral PhD defence: In a single pdf file
HOW TO APPLY: [email protected]
