Ingenieur R&D Cifre, Adversarial Neural Networks for Tumor Micro-environment Modeling

Ingenieur R&D Cifre, Adversarial Neural Networks for Tumor Micro-environment Modeling

ORGANISATION/COMPANY TheraPanacea – Centre Leon Berard – Cnrs

RESEARCH FIELD deep learning, medical image analysis, radiomics, dose painting


- Professor Vincent Gregoire, Radiation Oncology Dept, Centre Léon Bérard, Lyon

- Dr. David Sarrut, Directeur de Recherce, Cnrs, Centre Léon Bérard, Lyon

- Professor Nikos Paragios, TheraPanacea, Paris STARTING

DATE November 1st 2019

TheraPanacea at glance

We are a fast growing medical start-up company (Paris Region AI Challenge 1st Prize in 2018) that develops artificial intelligence solutions to unlock the full potential of cancer treatment by radiation therapy. For more information about the company please visit

We are looking for a PhD student...

Determining cancer/tumor micro and macro environment infiltration is one of the most important factors for planning radiotherapy treatment. Currently, tumor volume delineation is typically done on the basis of clinical examination and imaging, and in general homogeneous margins are applied to determine its microscopic infiltration prior to treatment. Such margins are based on clinical experience, and do not necessarily correspond to the biological infiltration, thus potentially hampering treatment efficiency. In the context of this thesis, we would like to couple clinical and imaging information, and biological evidence towards learning automatically from images the micro and macro tumor infiltration level. In particular, using the latest advances of machine learning we would like to develop a predictive mechanism that could produce automatically the level of infiltration and the characterization of tumor components from multi-parametric imaging. The method will use as basis pairs of multiparametric imaging (prior to tumor resection through surgery) and histological images (obtained from the surgical specimen after resection) on which explicit understanding of the level of infiltration is feasible. Generative adversarial networks will be employed to couple evidence driven observations from images with proper biological assessment of tumor infiltration towards developing the model that subsequently will be used in clinical practice: (i) to determine the extent of tumor infiltration within surrounding healthy tissue towards automatic determination of tumor margins in the context of radiation therapy, (ii) to correlate multi-parametric radiomics driven imaging characteristics with biological assessment of tumor aggressiveness, (iii) to determine dose prescription and biologically driven component-based plan adaptation in the context of radiation therapy; reinforcement learning to handle the problem of selecting the plan with optimal parameters to ensure the efficacy of the plan will also be developed.

Head and neck tumors will be used as basis for the evaluation of this ambitious study on which pretreatment multi-parametric imaging will be coupled with full 3D digital histology after surgery and correlated with long term survival. The successful applicant will work in an international team supervised by leading experts in the field of radiation oncology, medical imaging and medical physics. The results of the research will be published in high impact journals.

Required skills

- Master’s degree in computer science or applied mathematics

- Ability to work independently and as a member of a research team

- Experience in optimization, deep learning and python is expected

- English at a conversational and written level (international team)

Required documents

- Cover letter and curriculum vitae

- Academic transcripts, duplicate of the Master’s diploma if available

- Recommendation letter of one referee

To apply, please submit your application at the following address: