Post-doctoral position (23 month) on weakly supervised computer vision for mass spectrometry imaging (Univ. Lille, France)

2 Mars 2023

---

Catégorie : Post-doctorant

---

Post-doctoral position (23 month) on weakly supervised computer vision for mass spectrometry imaging (Univ. Lille, France)

Context

Mass spectrometry is a technique to analyze chemical compounds by vaporizing molecules and measuring the quantities of ions that constitutes them at different mass/charge ratios. A mass spectrometer typically produces a high-dimensional vector (in the order of 1K to 10K components), called spectrum, that acts as a chemical signature of a sample. Mass spectrometry imaging (MSI) techniques can produce images of biological samples where each pixel is a high-dimensional spectrum; these images can be seen as maps of the molecular content of the samples, that can be used to localize specific structures like cancerous cells. Such images contain very rich data about the samples but are difficult to interpret by human operators. Machine learning-based techniques are typically used to help in their understanding, including deep learning models; however, these models face two major challenges:
- the high dimensionality of the spectra,
- the limited amount of annotated data available to train them, which is difficult to obtain, especially in the fields of biology and medicine.

There is therefore a need for machine learning-based computer vision approaches that can classify high-dimensional samples in a weakly supervised setting or a few-shot learning setting.

This problem is addressed in the DEADPOOL project (In Vivo Ambient Water-Assisted Laser Desorption/Ionization Mass Spectrometry Imaging) funded by ANR (French national research agency) and run by PRISM (University of Lille / INSERM), CRIStAL (Univ. Lille / CNRS), and PhLAM (Univ. Lille / CNRS). The project aims at:
- producing a high-definition MSI device that can be used in vivo, e.g. on patients in the operation theater,
- developing computer vision techniques for the analysis of MSI images in near-real-time to map the content of the samples.

The application use case of the project is the semantic segmentation of MSI of tumors acquired in vivo, in the operation theater, as an assistive tool for surgeons. The project covers all aspects of the system, from its hardware and software design to its validation in clinical trials on dog patients.


Objectives

The objective of this post-doctoral position is to participate in the design of deep learning algorithms for the classification of high-dimensional data and with little annotated training data. The candidate will be expected to:
- design new neural network models, training algorithms, or training workflows to tackle the classification of high-dimensional data with little annotated data,
- validate the proposed models on standard datasets of the literature,
- validate the proposed models on real-world MSI data from the project,
- participate in the development of the prototype to be used in the clinical trial.
The scope of the contributions are not expected to be limited to the use case of the project.

The candidate will also participate in publishing the results of the project, producing deliverables, and attend project meetings.


Skills

Candidates must hold a Ph.D. in computer science, signal and image processing, or a related field, with a specialization in machine learning or computer vision. Ph.D. students that expect to graduate by the starting date of the contract can apply.

Experience with one or more of the following items is a plus, but not mandatory:
- deep neural networks,
- weakly supervised learning, few-shot learning, or self-supervised learning,
- semantic segmentation,
- mass spectrometry data,
- high-dimensional data (e.g. hyperspectral imaging).

The candidates must also have:
- good programming skills in Python and experience with machine learning libraries (pytorch, tensorflow/keras, scikit-learn, etc.),
- good scientific writing skills,
- scientific curiosity and the will to interact with researcher from other fields.

Knowledge of the French language is not mandatory.


Conditions

Contract duration: 23 months

Expected starting date: June 1st, 2023 (can be subject to negociation)

Salary: From 2350€/month to 2850€/month (gross salary) depending on experience, including health insurance, retirement fund, and 5 weeks/year of paid vacations.

Location: Our team is located in IRCICA, on the scientific campus of University of Lille in Villeneuve d'Ascq, France.

Equipment: The candidate will be provided a laptop computer for everyday tasks and given access to computing resources (GPUs), including hardware dedicated to the project only.


About CRIStAL, the FOX team, and Univ. Lille

CRIStAL is the laboratory in computer science, signal processing and automatic control of University of Lille. It gathers over 450 researchers in the field.

The FOX team is part of the Image research group of CRIStAL. We carry out research in computer vision, with a focus on (but not limited to) human behavior understanding. In recent years, we have developed machine learning-based techniques for data- and energy-efficient computer vision. We have a publication record in major venues in the field (IEEE TAFFC, IEEE TIP, Pattern recognition, IJCNN, WACV, etc.). We are also part of IRCICA, an interdisciplinary research institute of CNRS, where we conduct research on energy-efficient neural networks.

CRIStAL and IRCICA are located on the scientific campus of University of Lille, in Villeneuve d'Ascq (France). The campus is located near the city of Lille (15 minutes by car or subway). Lille is the largest city in Northern France. It can be easily accessed from Paris (1 hour by train), Bruxelles (30 minutes), and London (90 minutes), and is renowned for its cultural life, gastronomy, and friendly locals.


How to apply?

Application process: Applicants should send
- a curriculum vitae,
- the contact info (name, position, email address) of (at least) two references,
- (optionally) recommendation letters,
by email to Pierre Tirilly (pierre.tirilly@univ-lille.fr). Applications can be sent in English or in French. Shortlisted candidates will then be contacted for an online interview.

Timeline: Applications should be sent by 30 March 2023. Interviews will be programmed typically one to two weeks after reception of the application, and no later than April, 7th. Final decisions will be sent by April, 12th. This schedule may be extended if the position is not filled by April, 12th.


Contact

Feel free to email Pierre Tirilly (pierre.tirilly@univ-lille.fr) for additional information about the position.