Vous êtes ici : Accueil » Kiosque » Annonce

Identification

Identifiant: 
Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?

Annonce

9 avril 2020

PhD Fellowship on Few-Shot Generative Modelling for medical image synthesis


Catégorie : Doctorant


PhD Fellowship on Few-Shot Generative Modelling for medical image synthesis. Send your application before April 30th.

 

Thesis location :

Laboratory of Medical Information Processing (LaTIM), French Institute of Health and Medical Research (INSERM UMR 1101), Brest, France,

Period :

3 years, starting on October 2020

Context and objectives :

Among deep learning approaches, generative models (GMs) in particular based on GANs (Generative Adversarial Networks) are gaining a lot of interest in medical imaging.

One of the current limitations of GMs in a medical context lies in the large volume of images necessary for their training, as access to large clinical datasets is generally made challenging. For this reason, it is often necessary to resort to synthetic image generation methods, the objective of which is to reach the highest achievable level of realism. One way to produce realistic simulated image datasets is the use of Monte Carlo, particularly within the context of radiation based imaging (PET, SPECT, CT). However, these simulations are not suitable for all imaging methods (such as magnetic resonance imaging (MRI), or ultrasound imaging) and they require very long, sometimes prohibitive, computing times.

The objective of this thesis will be to develop an alternative approach for realistic medical image generation based on the concept of "few shot learning", an emerging idea in computer vision which consists in restricting the learning of generative models to a small number or even a single image. These promising methodological developments have so far not been exploited for pathological image generation in medical imaging. One of the objectives of the thesis will be the development of few-shot GAN models for PET, CT and MRI cancer image simulation. The realism of the generated images will be evaluated with scrutiny through comparison with highly realistic Monte Carlo simulations and real clinical images.

Qualifications

Education :

The candidate must hold a Master’s degree in one of these domains : physics, electrical/electronic engineering, computer science, applied mathematics.

Scientific interests :

Good understanding of the physics of medical imaging and its challenges.

Programming skills :

Fluent data processing using scripting langages (UNIX shell/python/Matlab).

Languages :

English (complimentary), French (optional).

Contacts :

Send before April 30th (in French or in English) CV, grades/marks (whatever currently available if you are actually on a Master), and a brief statement of interest by email to

Vincent Jaouen : vjaouen@gmail.com and Dimitris Visvikis (dimitris.visvikis@inserm.fr).

 

Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2020.