Deep learning based reconstruction of highly under-sampled PC-MRI for the quantification of blood flow

Introduction

Magnetic resonance imaging (MRI) is a very powerful and non-invasive imaging technique, as it harmlessly allows detection and monitoring of morphological characteristics of internal organs. Phase-contrast MRI (PC-MRI) is an advanced use of the MRI capabilities that can be used to simultaneously visualize and quantify blood flow [1]. A major application of PC-MRI is the detection and quantification of blood flow in the aorta, which can help diagnose and monitor major pathologies affecting the artery, such as aneurysms and dissections.

MRI acquisitions are inherently long, particularly in the case of the PC-MRI technique. Acquiring less data reduces the total acquisition time, but result in artefacts in the obtained images. Reconstruction techniques have long been developed to compensate for this effect.

More recently, deep learning-based reconstruction techniques have shown promise in solving the ill-posed problem of image reconstruction from under-sampled measurements [2]. In addition, they also provide a much faster evaluation than any other non-deep learning based technique that provides a reconstruction quality of the same order.

Objectives

The goal of this internship is to use the provided deep learning training framework to implement relevant architectures, monitor the associated training on the acquired data and perform hyper-parameter optimization to reach the best reconstruction quality of blood flow.

Candidate profile

The candidate should have a background in at least one of the following areas, and be enthusiastic about the others:

  • Python programming language
  • Image processing
  • Machine learning.

The internship is set to start in between February and April, and to last 5-6 months. The associated gratuity of ~560 euros/month.

It should be noted that this project could be pursued as a PhD work, in combination with aspects on MR sequence programming, in the framework of an ANR project.

How to apply

Please send a CV and a short cover letter to:

Gaël Touquet gael.touquet@creatis.insa-lyon.fr

Monica Sigovan monica.sigovan@creatis.insa-lyon.fr

Odysee Merveille odyssee.merveille@creatis.insa-lyon.fr

 

References

[1] - Stankovic, Zoran; Allen, Bradley D.; Garcia, Julio; Jarvis, Kelly B.; Markl, Michael (2014). "4D flow imaging with MRI". Cardiovascular Diagnosis and Therapy. 4 (2): 173–192. doi:10.3978/j.issn.2223-3652.2014.01.02. PMC 3996243. PMID 24834414.

[2] - " Applications of deep learning to MRI images: A survey" Jin Liu et al. Big Data Mining and Analytics ( Volume: 1, Issue: 1, March 2018) 10.26599/BDMA.2018.9020001