Connectionists: Open PhD position at UCLouvain, Belgium (deep learning for medical images)

[John Lee]

The Université catholique de Louvain (UCLouvain) is hiring a PhD student 
within the framework of a project called "DeepContour" (Deep learning 
for automatic organ contouring of medical images in adaptive radiation 
therapy).

The project takes place at the crossing of machine learning and cancer 
treatment with photon or proton therapy).
To target the tumor and avoid organs at risk, radiation oncology heavily 
relies on medical images that must be annotated with the contours of 
these structures.
This difficult task is still performed mainly manually by physicians.
On the other hand, automatic recognition and delineation of organs in 
medical images is often turned into an image registration problem: a 
reference image, called atlas and manually segmented by an expert, is 
non-rigidly deformed to match the image to be segmented.
Registration can however suffer from inaccuracies, especially when 
trying to match anatomies of different patients.
DeepContour follows a different approach of the atlas, based on machine 
learning techniques.
Reference images will be used to train classifiers, namely, deep neural 
networks, which can afterwards process new images and label pixels or 
homogeneous groups of pixels, called superpixels.
A previous proof-of-concept has demonstrated the validity of this 
approach and the project aims to improve the methodology by:
* Defining new features to optimise prior segmentation into superpixels 
and therefore final, accurate delineation of organs (shape and texture 
features could be used in addition to gray level, distance and 
contiguity to organs).
* Managing unexpected or unpredictable objects (like medical devices, 
tumours, etc.), particularly difficult to deal with in image registration.
* Transpose from natural scenes to medical images an architecture that 
combines deep neural networks and superpixels.
* Make the methodology interactive (ask the user whenever the risk of 
error is significant) and adaptive (learn from new images segmented by 
the atlas and approved by physicians).
The project also includes the collection of reference data with a 
clinical partner and the comparison with registration-based atlases.

The ideal PhD student should have a master degree in engineering, 
computer science, or applied mathematics.
S/he should have a good background in mathematics, statistics, and 
programming (Python, Matlab, scripting langages), completed with 
scientific curiosity and good communication skills.
S/he will work in a multidisciplinary and dynamic environment 
(engineers, mathematicians, physicists, physicians), sharing its time 
between the ICTEAM Institute (Information and Communication 
Technologies, Electronics, and Applied Mathematics, Louvain-la-Neuve) 
and the IREC (Institut de Recherche Expérimentale et Clinique, Brussels).
S/he should speak at least English or French and be willing to learn the 
other language.

The contract is a 2+2 scholarship.
Two years are already secured, the 3rd and 4th depend on evaluation 
after the first two years.
The candidate should be eligible for a Belgian, tax-free PhD scholarship 
(EU citizen and holder of a Master degree).
The net salary is about 1700-1800 euros/month.

Applications will be considered as they are received; the position will 
remain open until filled.
The contract is planned to start in October 2016.

The Université catholique de Louvain (http://www.uclouvain.be) is the 
largest and highest-ranked university in the French-speaking part of 
Belgium, with more than 28000 students, 3600 researchers, and 2000 PhD 
students.
It is located mainly in Louvain-la-Neuve (25km in the South-East of 
Brussels) and the outskirts of Brussels (University hospital and 
medicine school).
The University provides easy access to language courses, as well as 
sport facilities and cultural events.

Contact information: john.lee at uclouvain.be (send a CV and motivation letter)