Personalized multi-organ automatic segmentation and 3D morphological feature analysis of the human face

Scientific contribution:

The PhD research topic is part of the multidisciplinary PREDIT4FACE project, which was selected within the PEPR Digital Health Call for Projects of France 2030. PREDIT4FACE—"PREdictive DIgital Twins for FACial Expression"—aims to develop a multiscale predictive digital twin of the face to better understand, model, and rehabilitate facial expressions. The project is led by the BMBI laboratory (UTC-CNRS) and brings together a multidisciplinary consortium with the following partners: CHIMERE (UPJV-CHU Amiens, INSERM), LaMcube (Centrale Lille Institut, CNRS), LaTIM (IMT Atlantique, INSERM), and Roberval (UTC).

Expected contributions of the PhD: 

Understanding the mechanical function of facial muscles during facial expressions and mimetic movements is essential for establishing a quantitative diagnosis and defining a personalized functional rehabilitation strategy in patients who have undergone facial paralysis or face transplantation. Personalized finite element models have been commonly developed from medical imaging to investigate the activation, contraction, and coordination of facial muscles during facial expressions. However, fast and accurate segmentation and reconstruction of human facial structures (i.e., the skull, external skin-based envelope, and facial muscles) remain particularly challenging due to their anatomical complexity (small size, strong shape variability between subjects/patients). In particular, automatic segmentation and reconstruction of individual facial muscles with fiber-oriented architecture from low contrast magnetic resonance imaging (MRI) have not yet been achieved. Moreover, the development of a statistical shape model of the human face is clinically relevant and technically challenging (multi-object SSM, muscle attachment sites, correspondence establishment within and between muscles) for clinical translation of the model generation and exploitation.

The objectives of this PhD project are two-fold:

1. The first objective is to develop an automatic MRI-based segmentation pipeline to segment and reconstruct human facial structures (i.e., the skull, skin envelope, and 10 pairs of facial muscles), along with their associated anatomical features, using state-of-the-art deep learning approaches such as CNNs, U-Net, and their recent variants (e.g., nnU-Net and U-NetR). Learning database will be built from MRI images of normal subjects and facial palsy patients. Particular care will be taken to ensure the anatomical consistency of the segmented structures, especially the accurate positioning of muscle attachment points. To this end, fine-tuning strategies and Efficient Paired Attention (EPA) methods will be employed.

2. The second objective is to develop a multi-object statistical shape model (SSM) that includes the skull, fat and skin envelope and 10 pairs of facial muscles. Introduced in the 2000s, SSMs have proven to be a valuable way to describe shape variability within a studied dataset. SSMs are developed by creating 3D models for multiple versions of a shape (e.g. from multiple patients) and by establishing correspondence between them. Reduction dimension methods can then determine mutually independent patterns of shape variation and their contribution to the shape variability. The SSM will be augmented with the muscle insertion points for better accuracy. The model’s robustness will be validated using standard metrics like compactness, specificity, and generalization before to be used for the personalized 3D morphological feature analysis. 

Related publications of the team:

• Nguyen T. N. et al. (2021) https://doi.org/10.1016/j.cmpb.2020.105846

• Boutillon A. et al. (2022) https://doi.org/10.1016/j.media.2022.102556

• Fouefack J. R. et al. (2023) https://doi.org/10.1016/j.media.2022.102730 

• Tran V. D. et al. (2023) https://doi.org/10.3390/bioengineering10060737

• Guezou-Philippe A. et al. (2025) https://doi.org/10.1371/journal.pone.0325587 

Internationally recognized generalist engineering school of the IMT (Institut Mines-Télécom), leading French engineering school (Technological University), IMT Atlantique aims to support transitions, train responsible engineers, and use scientific excellence to serve teaching, research, and innovation. Recognized for the quality of its research, the scientists at IMT Atlantique support approximately 300 doctoral students.

The Data Science Department of IMT Atlantique conducts cutting-edge research in data science and health through the LaTIM INSERM 1110 laboratory (https://www.univ-brest.fr/latim/en/page/equipe-imagine?q=fr/page/equipe-imagine).  We develop multi-disciplinary research driven by members from IMT Atlantique, CHU Brest, University of Western Brittany and Inserm.

Centrale Lille is a renowned highly selective Graduate Engineering school ranked among the top “Grandes Ecoles” of the French higher education and research, with roots back to 1854. The research proposal aligns with strategic challenge 6, 'Life, Health, and Well-being,' for Centrale Lille.

As part of the PREDIT4FACE project, the doctoral researcher will be based at IMT Atlantique (Brest campus) and collaborate closely with Centrale Lille and the Hospital of Amiens, the clinical partner of the project.

Location: IMT Atlantique (Brest Campus)

Laboratory: LaTIM UMR INSERM 1101

Doctoral School: Sciences pour l’Ingénieur et le Numérique (SPIN) 

Domain: Sciences et Technologies de l'Information et de la Communication (STIC)

Specialty: Image, Signal, Vision

Partnership: Centrale Lille

Funding: PREDIT4FACE (PEPR Digital Health Call 2030)

Profile/required skills:

Engineer or MSc degree in computer vision, applied mathematics, and/or biomedical engineering

• Image processing and Statistical modelling

• 3D modelling and reconstruction

• Programming C++/Python and Deep learning, with commented and versioned code

• Analytical and synthesis skills for literature review

• Strong writing and communication ability to disseminate research results

• Ability to collaborate in an interdisciplinary team

• Interest in medical applications