Aller au menu Aller au contenu Aller à la recherche
  • Français
  • English

Internship: generation of dynamic pairs of pristine/noisy-real PET images to train a deep learning algorithm

aA - +Imprimer la page

Internship: generation of dynamic pairs of pristine/noisy-real PET images to train a deep learning algorithm

Supervisors: Antoine Merlet, PhD student, Benoit Presles, PhD, Jean-Marc Vrigneaud, PhD

Starting date: 1st February 2021

Project description

18F-Fluorodeoxyglucose (FDG) Positron Emission Tomography (PET) is a gold standard for the evaluation of tumour metabolism and is now widely used in medical oncology for cancer detection, staging, and more recently therapy monitoring. The concomitant evaluation of tumour perfusion and tumour metabolism is promising for the monitoring of new therapies targeting both tumour perfusion and viability. However, the development of dynamic FDG PET in clinical practice is challenging, due to the poor spatial resolution and signal to noise ratio of dynamic PET images. For example, classical reconstruction algorithms (MLEM) have shown to produce bias in short-duration frames in dynamic PET studies. Such a bias is very problematic for quantitative imaging, particularly when trying to derive an image-derived input function. Nowadays, new denoising techniques based on deep learning can be used to improve the quality of the PET images, by using either supervised or unsupervised approaches. If supervised methods are used, pairs of pristine/noisy-real PET images are needed to train the network. This is the aim of this internship: generating dynamic pairs of pristine/noisy-real PET images. To do so, a GATE ( modelisation of our PET camera has already been done to simulate/generate static PET images. The main objective of the internship is therefore to make the simulations dynamic.

The student will be located in the “Imagerie et Vision Artificielle" (ImViA) in Dijon (France), and will work in close relation with the medical team of the Georges François Leclerc Centre in Dijon (France).

Person specification

Candidates must be engineering or master students in a relevant subject area such as physics, computer science, biomedical engineering or applied mathematics.


Applications (including a CV and covering letter outlining your motivation for the position) should be sent to Benoit Presles (benoit.presles at and Jean-Marc Vrigneaud (jmvrigneaud at

Closing date: 31th December 2020


Log In

Create an account