Self-Supervised Learning applied to non-invasive predictions of clinical measurements associated with liver surgery

Inria Palaiseau Palaiseau, France

About the position

Context

Key words:
representation learning, knowledge distillation, clinical images
The internship will be co-supervised by Irene Vignon-Clémentel (Directrice de recherche), Hugues Talbot (Directeur de recherche) and Francesco Songia (PhD student).
Portal hypertension is a central complication of chronic liver disease and a major risk factor in liver surgery. A high gradient between the portal vein and the vena cava is strongly associated with post-operative liver failure.
Deep learning techniques can be employed to predict the invasive measurements usually taken in the operating room, starting from preoperative exams (CTs, biological samples, MRI, …).
Foundation models trained on thousands of CT scans are available [1], and they can provide the starting embeddings representing the patient. The other modalities at disposal will be combined with this representation, relying on self-distillation strategies [2].
Don’t hesitate to apply online including your CV, motivation letter and grades, or reach out to us for more information.
Francesco Songia:
francesco.songia@inria.fr

Assignment

The main goal of this internship is to combine all the available sources of information describing the patient state, relying on self-supervised learning. The different data sources have to be included in a more general pipeline that can deal with the addition of new modalities and that can support different downstream tasks.
Our research team also includes clinicians; they can provide specific interpretations and guidelines to better understand the clinical-related questions and how to work with the data at disposal. This means the intern will also exchange directly with clinicians and see how the research could be applied in practice.
[1] Blankemeier, Louis, et al. "Merlin: A vision language foundation model for 3d computed tomography."
Research square
(2024): rs-3.
[2] Caron, Mathilde, et al. "Emerging properties in self-supervised vision transformers."
2021 IEEE/CVF international conference on computer vision (ICCV)
. IEEE, 2021.

Main activities

Main tasks:

  • Literature review on related works (relying on self-supervised learning) to predict clinical measurements non-invasively
  • Gather all the data sources available, from foundation models to segmentation models
  • Combine the data modalities available, relying on self-supervised learning
  • Participate in activities of the research group (seminars, meetings, social activities)
  • Write a report, present the results to the research group and in the ARTEMIs EU project

Skills

The core competences for an ideal candidate are:

  • computer vision basics (master level, M1 or M2)
  • experience in programming (Python)
  • good communication skills in English

Experience on working with clinical images will be considered as a valuable plus.

This listing was collected from a public source and is reproduced here for information only. Always confirm the details on the original posting before applying.
View the original posting

Similar positions