Post-Doctoral Research Visit F - M From Foetus To Elderly Universal Domain-Agnostic Segmentation Of Brain Mri Across The Human Lifespan. H/F

INRIA

Post-Doctoral Research Visit F/M From foetus to elderly: universal domain-agnostic segmentation of brain MRI across the human lifespan.
Le descriptif de l'offre ci-dessous est en Anglais
Type de contrat : CDD

Niveau de diplôme exigé : Thèse ou équivalent

Fonction : Post-Doctorant

A propos du centre ou de la direction fonctionnelle

Inria is the French National Institute for Research in Digital Science, of which the Inria Côte d'Azur University Center is a part. With strong expertise in computer science and applied mathematics, the research projects of the Inria Côte d'Azur University Center cover all aspects of digital science and technology and generate innovation. Based mainly in Sophia Antipolis, but also in Nice and Montpellier, it brings together 47 research teams and nine support services. It is active in the fields of artificial intelligence, data science, IT system security, robotics, network engineering, natural risk prevention, ecological transition, digital biology, computational neuroscience, health data, and more. The Inria Center at Université Côte d'Azur is a major player in terms of scientific excellence, thanks to the results it has achieved and its collaborations at both European and international level.

Contexte et atouts du poste

Thispostdoc offer (2 years)aims at developing an AI algorithm for thedomain-agnostic segmentation of brain MRI scans across the human lifespan. It will be conducted under the supervision of Dr. Benjamin Billot (Inria) and Dr. Henry Tregidgo (UCL).

As such, thispostdoc offers a unique opportunity to work with two of the world-leading research groups in medical image analysis: the Epione team at Inria (Antibes, France) and the Hawkes Institute at University College London (UK). The position will be primarily based in Epione, with planned research visits to UCL.

Epione is part of Inria (the French National Institute for Research in Digital Science and Technology), and is located in the technological cluster of Sophia-Antipolis. It is an internationally recognised team of around 60 researchers working at the intersection of machine learning, physics-based modelling, and geometric statistics for medical imaging.

The Hawkes Institute, based at the heart of UCL's AI center in central London, provides a highly interdisciplinary environment focused on advancing healthcare technologies through cutting-edge research in computational, engineering, and medical sciences.

Mission confiée

Brain MRI segmentation plays a prominent role in understanding the human brain by enabling an array of subsequent analyses such as volumetry, morphology, and connectivity. In this context, automated segmentation methods have been proposed to solve cost and reproducibility issues with manual contouring. As such, modern deep learning networks are now able to achieve state-of-the-art results and almost reach human-level performance. However, these networks suffer from a lack of generalisability, since their accuracy dramatically drops when they are presented with images outside of their training domain. This issue, known as the domain gap, is critical for the analysis of brain MRI scans, since these can present a huge variability in terms of scanners, sequences, resolutions, artefacts, population shifts, pathologies, etc.

Recently, we addressed the domain gap issue by introducing a new paradigm, domain randomisation, to brain MRI. This strategy trains a segmentation network using synthetic data sampled from a parametric generative model. Crucially, instead of fine-tuning this model to produce realistic images, we fully randomise all its parameters (resolution, MRI contrast, artefacts, etc.) to produce extremely variable images. When trained with such data, the downstream segmentation network, named SynthSeg [1], is forced to learn domain-agnostic features, such that it can then segment a wide range of real images without any retraining, or fine-tuning. SynthSeg is distributed with FreeSurfer (50,000+ active licenses), and has been highly impactful in neuroimaging, including adaptations to many tasks beyond segmentation.

While SynthSeg has shown unprecedented generalisation across a wide range of contrasts and resolutions in adult brain MRI, its performance greatly degrades on children, infants and foetuses. Indeed, the rapidly evolving anatomy combined with myelination-triggered contrast changes make these populations particularly challenging to analyse, albeit being central to understanding the human brain development. Interestingly, SynthSeg has been specifically retrained on foetuses [2] and infants [3], but the resulting models cannot generalise to age-ranges outside and often use simplified brain atlases.

Principales activités

This project aims to deliver a universal, fully automated tool for robust and accurate segmentation of brain MRI across the entire lifespan: from foetuses to elderly. Unifying such heterogeneous data within a single model is highly challenging, yet its success would be highly impactful for the neuroimaging community.

Building on our prior work (SynthSeg) , we will extend domain randomisation to capture the unique anatomical and contrast characteristics of foetal and infant brains. Rather than relying on limited foetal-specific atlases, we will deform adult atlases to younger anatomies using advanced non-linear cross-modal registration, yielding fine-grained topological maps beyond the resolution of current foetal resources. In parallel, we willintroduce a dynamic regrouping of anatomical regions to model developmental contrast shifts, leveraging Laplacian graph clustering at the regional level. These innovations will enable seamless integration of early-life and adult data into a single training framework, resulting in a unified SynthSeg-Lifespan model.

We will validate the model across diverse imaging domains, age groups, and clinical datasets to ensure robustness and generalisability. All developments will be released through the next-generation SynthSeg repository designed for much broader support and dissemination (currently under development). Moreover, if successful, the trained model will be integrated into the widely used FreeSurfer neuroimaging softwaresuite (https://surfer.nmr.mgh.harvard.edu).

Compétences

- PhD degree in medical image analysis.

- Advanced knowledge in deep learning for medical imaging, especially segmentation and registration.

- Experience in adult and/or fetal brain MRI.

- Advanced programming skills in Python, with experience in both PyTorch AND TensorFlow.

- Good writing skills.

- Good relational and communication skills to interact with professionals with either medical or engineering backgrounds.

Avantages

- Subsidized meals
- Partial reimbursement of public transport costs
- Leave: 7 weeks of annual leave + 10 extra days off due to RTT (statutory reduction in working hours) + possibility of exceptional leave (sick children, moving home, etc.)
- Possibility of teleworking and flexible organization of working hours
- Professional equipment available (videoconferencing, loan of computer equipment, etc.)
- Social, cultural and sports events and activities
- Access to vocational training
- Contribution to mutual insurance (subject to conditions)

Rémunération

Gross Salary: 2788 € per month