Postdoctoral at University of Lille

Research area: Deep Learning for Anomaly Detection in Human Lung Biopsies Using LIBS Hyperspectral Data.

Activities:
Today, we face a major public health challenge due to environmental and occupational exposure to inhaled mineral particles, metals, and dust, which are often difficult to diagnose. Respiratory diseases represent the majority of occupational illnesses, both in terms of the variety of adverse health effects and the number of cases. This is due to the susceptibility of the respiratory tract to dust or metallic particles in the air. The overall health effects of exposure to mineral particles, metals, and dust are significant in terms of prevalence, morbidity, and healthcare costs. In this context, we proposed, as part of the ANR dIAg-EM project (Medical Diagnosis by Artificial Intelligence Applied to Elemental LIBS Microscopy), the development of a diagnostic tool to detect anomalies in LIBS hyperspectral images acquired from human lung sections. The first step of this project involved collecting a large set of lung sections from sick patients and controls, followed by measuring the corresponding LIBS hyperspectral images. We wish to emphasize the exceptional nature of the data acquired in this project in several respects. Biologically, it is the largest cohort of patients ever studied on this issue, with about 100 patients and an equal number of controls from autopsies. From a data structure perspective, each hyperspectral image consists of several hundred thousand spectra, each defined by more than 2000 wavelengths. Your research will thus focus on developing deep networks capable of detecting anomalies in a LIBS hyperspectral image based on all these acquired data. We have no preconceptions about the deep network approaches to be used or even potential architectures, giving you complete freedom for this exploration. Finally, operationally, you will be the sole user of a GPU computing cluster installed in our laboratory, consisting of 8 Tesla V100 32GB cards, which will be ideal for the prototyping phase. If necessary, we will, of course, use more extensive
external computing resources (AWS, IDRIS, etc.).

Skills:
– PhD in Computer Science, Applied Mathematics, or Medical Image Analysis (other PhD disciplines will be considered if the candidate can demonstrate proven expertise in deep learning).
– Programming skills in Python, knowledge of deep learning frameworks such as PyTorch and/or TensorFlow.

Work Environment:
You will work within the DyNaChem team (Dynamics, Nanoscopy & Chemometrics, https://lasir.cnrs.fr/dynachem/) at LASIRE UMR CNRS 8516 (Laboratory of Spectroscopy for Interactions, Reactivity, and Environment, https://lasir.cnrs.fr)), which is internationally recognized for chemometrics and machine learning applied to spectroscopy in a broad sense. Our laboratory is located in Villeneuve d’Ascq in a green setting, 15 minutes from Lille by metro, a very dynamic student city in northern France. As this research is part of the dIAg-EM ANR project, you will work closely with Dr. Vincent Motto-Ros’s team, one of the world’s leading experts in LIBS imaging, located at ILM in Lyon (Institute Lumière Matière, https://ilm.univ-lyon1.fr/index.php), as well as with Professor Benoit Busser from the Grenoble hospital, a specialist in lung diseases. Since LASIRE is in a restricted area (ZRR), a prior access request must be granted by the defense official before the contract is signed.

How to apply:
You must apply only through the website:
https://emploi.cnrs.fr/Offres/CDD/UMR8516-LUDDUP-002/Default.aspx
If you have any specific questions, you can contact Professor Ludovic Duponchel
(Ludovic.duponchel@univ-lille.fr)

Courtesy

L. Duponchel