EPIONE

EPIONE - 2025

2025Activity reportProject-TeamEPIONE‌

RNSR: 201822641L

Research center Inria Centre at Université‌ Côte d'Azur
Team name: E-Patient: Images, Data &‌ MOdels for e-MediciNE

Creation of the Project-Team: 2018‌ May 01

Each year, Inria research teams publish‌ an Activity Report presenting their work and results‌ over the reporting period. These reports follow a‌ common structure, with some optional sections depending on‌ the specific team. They typically begin by outlining‌ the overall objectives and research programme, including the‌ main research themes, goals, and methodological approaches. They‌ also describe the application domains targeted by the‌ team, highlighting the scientific or societal contexts in‌ which their work is situated.

The reports then‌ present the highlights of the year, covering major‌ scientific achievements, software developments, or teaching contributions. When‌ relevant, they include sections on software, platforms, and‌ open data, detailing the tools developed and how‌ they are shared. A substantial part is dedicated‌ to new results, where scientific contributions are described‌ in detail, often with subsections specifying participants and associated keywords.

Finally, the‌ Activity Report addresses funding,‌ contracts, partnerships, and collaborations‌‌ at various levels, from industrial agreements to international‌ cooperations. It also covers‌ dissemination and teaching activities,‌‌ such as participation in scientific events, outreach, and‌ supervision. The document concludes‌ with a presentation of‌‌ scientific production, including major publications and those produced‌ during the year.

Keywords‌

Computer Science and Digital‌‌ Science

A3.3. Data and knowledge analysis
A3.4. Machine‌ learning and statistics
A4.3.‌ Cryptography
A4.4. Security of‌‌ equipment and software
A4.8. Privacy-enhancing technologies
A5.2. Data‌ visualization
A5.3. Image processing‌ and analysis
A5.6. Virtual‌‌ reality, augmented reality
A5.9. Signal processing
A6.1. Methods‌ in mathematical modeling
A6.2.‌ Scientific computing, Numerical Analysis‌‌ & Optimization
A6.3. Computation-data interaction
A8.3. Geometry, Topology‌
A9. Artificial intelligence
A9.2.‌ Machine learning
A9.3. Signal‌‌ processing
A9.6. Decision support
A9.7. AI algorithmics
A9.9.‌ Distributed AI, Multi-agent
A9.10.‌ Hybrid approaches for AI‌‌
A9.12. Computer vision

1‌‌ Team members, visitors, external collaborators

Research Scientists

Nicholas‌ Ayache [Team leader‌, Inria, Senior‌‌ Researcher, HDR]
Irene Balelli [Inria‌, ISFP]
Benjamin‌ Billot [Inria,‌‌ Researcher]
Hervé Delingette [Inria, Senior‌ Researcher, HDR]‌
Marco Lorenzi [Inria‌‌, Senior Researcher, from Oct 2025,‌ HDR]
Marco Lorenzi‌ [Inria, Researcher‌‌, until Sep 2025, HDR]
Guillaume‌ Olikier [Inria,‌ Starting Research Position,‌‌ until Aug 2025]
Xavier Pennec [Inria‌, Senior Researcher,‌ HDR]
Maxime Sermesant‌‌ [Inria, Senior Researcher, HDR]‌

Post-Doctoral Fellows

Safaa Al‌ Ali [Inria,‌‌ until Jun 2025]
Francesco Cremonesi [Inria‌]
Bernhard Follmer [‌Inria, Post-Doctoral Fellow‌‌, from May 2025]
Jia Guo [‌CHU Nice]
John‌ Kalkhof [Inria,‌‌ Post-Doctoral Fellow, from Jul 2025]
Huiyu‌ Li [IHU Respirera‌, from May 2025‌‌]
Buntheng Ly [IHU Lyric]
Ghiles‌ Reguig [Inria]‌
Jesus Jairo Rodriguez Padilla‌‌ [Inria]
Alessandro Viani [Inria,‌ Post-Doctoral Fellow, until‌ Feb 2025]

PhD‌‌ Students

Amel Bakhouche [Université Côte d'Azur,‌ from Feb 2025]‌
Olivier Bisson [Inria‌‌]
Florencia Boccarato [Inria, from Feb‌ 2025]
Fahym Bounazou‌ [AP-HP, from‌‌ Feb 2025]
Alix De Langlais [Inria‌]
Nicolas Drettakis [‌Inria]
Ezem Sura‌‌ Ekmekci [Inria]
Federica Facente [Inria‌]
Camilla Ferrario [‌Inria]
Giulia Foroni‌‌ [Inria, from Oct 2025]
Sebastien‌ Goffart [CHU Nice‌]
Lisa Guzzi [‌‌Université Côte d'Azur, until Oct 2025]‌
Manasi Kattel [Inria‌]
Wassila Khatir [‌‌Université Côte d'Azur]
Arnaud Lang [Inria‌, from Dec 2025‌]
Maelis Morier [‌‌Inria]
Huyen Trang‌ Nguyen [Inria]
Evariste Njomgue Fotso [‌Inria]
Giuseppe Orlando [Inria, from‌ Jun 2025]
Rafael Luis Soares Da Costa‌ E Silva [Inria]
Tom Szwagier [‌Inria]
Adrien Tchuem Tchuente [Inria,‌ from Oct 2025]
Elie Thellier [Inria‌, from Mar 2025]
Tony Zaayter [‌Inria, from Nov 2025]

Technical Staff‌

Nicolas Cedilnik [Inria, Engineer, from‌ Mar 2025]
Lucie Chambon [Inria,‌ Engineer]
Gaetan Desrues [Inria, Engineer‌, from Mar 2025]
Hye Lim Lee‌ [Inria, Engineer]
Marco Milanesio [‌Université Côte d'Azur]
Mihaela Pop [Inria‌, Engineer, from Mar 2025 until Oct‌ 2025]
Hari Sreedhar [Université Côte d'Azur‌, Engineer, from Mar 2025]

Interns‌ and Apprentices

Prabal Ghosh [Inria, Intern‌, from Apr 2025 until Sep 2025]‌
Tuan Hoang [Inria, Intern, from‌ Apr 2025 until Aug 2025]
Arnaud Lang‌ [Inria, Intern, from Apr 2025‌ until Sep 2025]
Tuan Anh Nguyen [‌Inria, Intern, from Apr 2025 until‌ Aug 2025]
Giuseppe Orlando [Eurecom,‌ from Feb 2025 until May 2025]

Administrative‌ Assistant

Nathalie Nordmann [Inria]

Visiting Scientists‌

Alessandra Corda [Politecnico de Milano, from‌ Nov 2025]
Hervé Lombaert [Polytechnique Montréal‌, from Dec 2025]

External Collaborators

Sébastien‌ Frey [CHU Nice]
Eleonore Haupaix-Birgy [‌CHU Nice, from Feb 2025]
Cécile‌ Rouzier [CHU Nice]

2 Overall objectives‌

2.1 Description

Our long-term goal is to contribute‌ to the development of what we call the‌ e-patient (digital patient) for e-medicine (digital medicine) (Fig.‌ 1).

the e-patient (or‌ digital patient) is a set of computational models‌ of the human body able to describe and‌ simulate the anatomy and the physiology of the‌ patient's organs and tissues, at various scales, for‌ an individual or a population. The e-patient can‌ be seen as a framework to integrate and‌ analyze in a coherent manner the heterogeneous information‌ measured on the patient from disparate sources: imaging,‌ biological, clinical, sensors, ...
e-medicine (or digital medicine)‌ is defined as the computational tools applied to‌ the e-patient to assist the physician and the‌ surgeon in their medical practice, to assess the‌ diagnosis/prognosis, and to plan, control and evaluate the therapy.

The models that‌ govern the algorithms designed‌ for e-patients and e-medicine‌‌ come from various disciplines: computer science, mathematics, medicine,‌ statistics, physics, biology, chemistry,‌ etc. The parameters of‌‌ those models must be adjusted to an individual‌ or a population based‌ on the available images,‌‌ signals and data. This adjustment is called personalization‌ and usually requires solving‌ difficult inverse problems. The‌‌ overall picture of the construction of the personalized‌ e-patient for e-medicine was‌ presented at the College‌‌ de France through an inaugural lecture and a‌ series of courses and‌ seminars, concluded by an‌‌ international workshop.

2.2 Organization

The research organization in‌ our field is often‌ built on a virtuous‌‌ triangle (Fig. 2). On one vertex, academic‌ research requires multidisciplinary collaborations‌ associating informatics and mathematics‌‌ to other disciplines: medicine, biology, physics, chemistry ...‌ On a second vertex,‌ a clinical partnership is‌‌ required to help defining pertinent questions, to get‌ access to clinical data,‌ and to clinically evaluate‌‌ any proposed solution. On the third vertex, an‌ industrial partnership can be‌ introduced for the research‌‌ activity itself, and also to transform any proposed‌ solution into a validated‌ product that can ultimately‌‌ be transferred to the clinical sites for an‌ effective use on the‌ patients.

Figure 2:‌‌ A pluridisciplinary research triangle.

Keeping this triangle in‌‌ mind, we choose our research directions within a‌ virtuous circle: we look‌ at difficult problems raised‌‌ by our clinical or industrial partners, and then‌ try to identify some‌ classes of generic fundamental/theoretical‌‌ problems associated to their resolution. We also study‌ some fundamental/theoretical problems per‌ se in order to‌‌ produce fundamental scientific advances that can help in‌ turn to promote new‌ applications.

3 Research program‌‌

3.1 Introduction

Our research objectives are organized along‌ 5 scientific axes (Fig.‌ 3):

Biomedical Image‌‌ Analysis & Machine Learning
Imaging & Phenomics, Biostatistics‌
Computational Anatomy, Geometric Statistics‌
Computational Physiology & Image-Guided‌‌ Therapy
Computational Cardiology & Image-Based Cardiac Interventions

Figure‌ 3: Epione's five‌ main research axes

For‌ each scientific axis, we‌ introduce the context and‌‌ the long term vision of our research.

3.2‌ Biomedical Image Analysis &‌ Machine Learning

The long-term‌‌ objective of biomedical image‌ analysis is to extract, from biomedical images, pertinent‌ information for the construction of the e-patient and‌ for the development of e-medicine. This relates to‌ the development of advanced segmentation and registration of‌ images, the extraction of image biomarkers of pathologies,‌ the detection and classification of image abnormalities, the‌ construction of temporal models of motion or evolution‌ from time-series of images, etc.

In addition, the‌ growing availability of very large databases of biomedical‌ images, the growing power of computers and the‌ progress of machine learning (ML) approaches have opened‌ up new opportunities for biomedical image analysis.

This‌ is the reason why we decided to revisit‌ a number of biomedical image analysis problems with‌ ML approaches, including segmentation and registration problems, automatic‌ detection of abnormalities, prediction of a missing imaging‌ modality, etc. Not only those ML approaches often‌ outperform the previous state-of-the-art solutions in terms of‌ performances (accuracy of the results, computing times), but‌ they also tend to offer a higher flexibility‌ like the possibility to be transferred from one‌ problem to another one with a similar framework.‌ However, even when successful, ML approaches tend to‌ suffer from a lack of explanatory power, which‌ is particularly annoying for medical applications. We also‌ plan to work on methods that can interpret‌ the results of the ML algorithms that we‌ develop.

3.3 Imaging and Phenomics, Biostatistics

The human‌ phenotype is associated with a multitude of heterogeneous‌ biomarkers quantified by imaging, clinical and biological measurements,‌ reflecting the biological and patho-physiological processes governing the‌ human body, and essentially linked to the underlying‌ individual genotype. In order to deepen our understanding‌ of these complex relationships and better identify pathological‌ traits in individuals and clinical groups, a long-term‌ objective of e-medicine is therefore to develop the‌ tools for the joint analysis of this heterogeneous‌ information, termed Phenomics, within the unified modeling‌ setting of the e-patient.

To date the most‌ common approach to the analysis of the joint‌ variation between the structure and function of organs‌ represented in medical images, and the classical -omics‌ modalities from biology, such as genomics or lipidomics,‌ is essentially based on the massive univariate statistical‌ testing of single candidate features out of the‌ many available. This is for example the case‌ of genome-wide association studies (GWAS) aimed at identifying‌ statistically significant effects in pools consisting of up‌ to millions of genetics variants. Such approaches have‌ known limitations such as multiple comparison problems, leading‌ to underpowered discoveries of significant associations, and usually‌ explain a rather limited amount of data variance.‌ Although more sophisticated machine learning approaches have been‌ proposed, the reliability and generalization of multivariate methods‌ is currently hampered by the low sample size‌ relatively to the usually large dimension of the‌ parameters space.

To address these issues this research‌ axis investigates novel methods for the integration of‌ this heterogeneous information within a parsimonious and unified‌ multivariate modeling framework. The cornerstone of the project consists in achieving an‌ optimal trade-off between modeling‌ flexibility and ability to‌‌ generalize on unseen data by developing statistical learning‌ methods informed by prior‌ information, either inspired by‌‌ "mechanistic" biological processes, or accounting for specific signal‌ properties (such as the‌ structured information from spatio-temporal‌‌ image time series). Finally, particular attention will be‌ paid to the effective‌ exploitation of the methods‌‌ in the growing Big Data scenario, either in‌ the meta-analysis context, or‌ for the application in‌‌ large datasets and biobanks.

Federated learning in multi-centric‌ studies. The current research‌ scenario is characterized by‌‌ medium/small scale (typically from 50 to 1000 patients)‌ heterogeneous datasets distributed across‌ centers and countries. The‌‌ straightforward extension of learning algorithms successfully applied to‌ big data problems is‌ therefore difficult, and specific‌‌ strategies need to be envisioned in order to‌ optimally exploit the available‌ information. To address this‌‌ problem, we focus on learning approaches to jointly‌ model clinical data localized‌ in different centers. This‌‌ is an important issue emerging from recent large-scale‌ multi-centric imaging-genetics studies in‌ which partners can only‌‌ share model parameters (e.g. regression coefficients between specific‌ genes and imaging features),‌ as represented for example‌‌ by the ENIGMA imaging-genetics study, led by the‌ collaborators at University of‌ Southern California. This problem‌‌ requires the development of statistical methods for federated‌ model estimation, in order‌ to access data hosted‌‌ in different clinical institutions by simply transmitting the‌ model parameters, that will‌ be in turn updated‌‌ by using the local available data. This approach‌ is extended to the‌ definition of stochastic optimization‌‌ strategies in which model parameters are optimized on‌ local datasets, and then‌ summarized in a meta-analysis‌‌ context. Finally, this project studies strategies for aggregating‌ the information from heterogeneous‌ datasets, accounting for missing‌‌ modalities due to different study design and protocols.‌ The developed methodology finds‌ important applications within the‌‌ context of Big Data, for the development of‌ effective learning strategies for‌ massive datasets in the‌‌ context of medical imaging (such as with the‌ UK biobank), and beyond.‌

3.4 Computational Anatomy and‌‌ Geometric Statistics

Computational anatomy is an emerging discipline‌ at the interface of‌ geometry, statistics and image‌‌ analysis which aims at developing algorithms to model‌ and analyze the biological‌ shape of tissues and‌‌ organs. The goal is not only to establish‌ generative models of organ‌ anatomies across diseases, populations,‌‌ species or ages but also to model the‌ organ development across time‌ (growth or aging) and‌‌ to estimate their variability and link to other‌ functional, genetic or structural‌ information. Computational anatomy is‌‌ a key component to support computational physiology and‌ is evidently crucial for‌ building the e-patient and‌‌ to support e-medicine.

Pivotal applications include the spatial‌ normalization of subjects in‌ neuroscience (mapping all the‌‌ anatomies into a common reference system) and atlas‌ to patient registration to‌ map generic knowledge to‌‌ patient-specific data. Our objectives will be to develop‌ new efficient algorithmic methods‌ to address the emerging‌‌ challenges described below and‌ to generate precise specific anatomical model in particular‌ for the brain and the heart.

The objects‌ of computational anatomy are often shapes extracted from‌ images or images of labels (segmentation). The observed‌ organ images can also be modeled using registration‌ as the random diffeomorphic deformation of an unknown‌ template (i.e. an orbit). In these cases as‌ in many other applications, invariance properties lead us‌ to consider that these objects belong to non-linear‌ spaces that have a geometric structure. Thus, the‌ mathematical foundations of computational anatomy rely on statistics‌ on non-linear spaces.

Geometric Statistics aim at studying‌ this abstracted problem at the theoretical level. Our‌ goal is to advance the fundamental knowledge in‌ this area, with potential applications to new areas‌ outside of medical imaging. Beyond the now classical‌ Riemannian spaces, we aim at developing the foundations‌ of statistical estimation on affine connection spaces (e.g.‌ Lie groups), quotient and stratified metric spaces (e.g.‌ orbifolds and tree spaces). In addition to the‌ curvature, one of the key problem is the‌ introduction of singularities at the boundary of the‌ regular strata (non-smooth and non-convex analysis).

A second‌ objective is to develop parametric and non-parametric dimension‌ reduction methods in non-linear space. An important‌ issue is to estimate efficiently not only the‌ model parameters (mean point, subspace, flag) but also‌ their uncertainty. We also want to quantify the‌ influence of curvature and singularities on non-asymptotic estimation‌ theory since we always have a finite (and‌ often too limited) number of samples. A key‌ challenge in developing such a geometrization of statistics‌ will not only be to unify the theory‌ for the different geometric structures, but also to‌ provide efficient practical algorithms to implement them.

A‌ third objective is to learn the geometry from‌ the data. In the high dimensional but‌ low sample size (small data) setting which is‌ the common situation in medical data, we believe‌ that invariance properties are essential to reasonably interpolate‌ and approximate. New apparently antagonistic notions like approximate‌ invariance could be the key to this interaction‌ between geometry and learning.

Beyond the traditional statistical‌ survey of the anatomical shapes that is developed‌ in computational anatomy above, we intend to explore‌ other application fields exhibiting geometric but non-medical data.‌ For instance, applications can be found in Brain-Computer‌ Interfaces (BCI), tree-spaces in phylogenetics, Quantum Physics, etc.‌

3.5 Computational Physiology and Image-Guided Therapy

Computational Physiology‌ aims at developing computational models of human organ‌ functions, an important component of the e-patient, with‌ applications in e-medicine and more specifically in computer-aided‌ prevention, diagnosis, therapy planning and therapy guidance. The‌ focus of our research is on descriptive (allowing‌ to reproduce available observations), discriminative (allowing to separate‌ two populations), and above all predictive models which‌ can be personalized from patient data including medical‌ images, biosignals, biological information and other available metadata.‌ A key aspect of this scientific axis is‌ therefore the coupling of biophysical models with patient data which implies that‌ we are mostly considering‌ models with relatively few‌‌ and identifiable parameters. To this end, data assimilation‌ methods aiming at estimating‌ biophysical model parameters in‌‌ order to reproduce available patient data are preferably‌ developed as they potentially‌ lead to predictive models‌‌ suitable for therapy planning.

Previous research projects in‌ computational physiology have led‌ us to develop biomechanical‌‌ models representing quasi-static small or large soft tissue‌ deformations (e.g. liver or‌ breast deformation after surgery),‌‌ mechanical growth or atrophy models (e.g. simulating brain‌ atrophy related to neurodegenerative‌ diseases), heat transfer models‌‌ (e.g. simulating radiofrequency ablation of tumors), and tumor‌ growth models (e.g. brain‌ or lung tumor growth).‌‌

To improve the data assimilation of biophysical models‌ from patient data, a‌ long term objective of‌‌ our research will be to develop joint imaging‌ and biophysical generative models‌ in a probabilistic framework‌‌ which simultaneously describe the appearance and function of‌ an organ (or its‌ pathologies) in medical images.‌‌ Indeed, current approaches for the personalization of biophysical‌ models often proceed in‌ two separate steps. In‌‌ a first stage, geometric, kinematic or functional features‌ are first extracted from‌ medical images. In a‌‌ second stage, they are used by personalization methods‌ to optimize model parameters‌ in order to match‌‌ the extracted features. In this process, subtle information‌ present in the image‌ which could be informative‌‌ for biophysical models is often lost which may‌ lead to limited personalization‌ results. Instead, we propose‌‌ to develop more integrative approaches where the extraction‌ of image features would‌ be performed jointly with‌‌ the model parameter fitting. Those imaging and biophysical‌ generative models should lead‌ to a better understanding‌‌ of the content of images, to a better‌ personalization of model parameters‌ and also better estimates‌‌ of their uncertainty.

3.6 Computational Cardiology and Image-Based‌ Cardiac Interventions

Computational Cardiology‌ has been an active‌‌ research topic within the Computational Anatomy and Computational‌ Physiology axes of the‌ previous Asclepios project, leading‌‌ to the development of personalized computational models of‌ the heart designed to‌ help characterizing the cardiac‌‌ function and predict the effect of some device‌ therapies like cardiac resynchronization‌ or tissue ablation. This‌‌ axis of research has now gained a lot‌ of maturity and a‌ critical mass of involved‌‌ scientists to justify an individualized research axis of‌ the new project Epione,‌ while maintaining many constructive‌‌ interactions with the 4 other research axes of‌ the project. This will‌ develop all the cardiovascular‌‌ aspects of the e-patient for cardiac e-medicine.

The‌ new challenges we want‌ to address in computational‌‌ cardiology are related to the introduction of new‌ levels of modeling and‌ to new clinical and‌‌ biological applications. They also integrate the presence of‌ new sources of measurements‌ and the potential access‌‌ to very large multimodal databases of images and‌ measurements at various spatial‌ and temporal scales.

4‌‌ Application domains

The main applications of our research‌ are in the field‌ of healthcare and more‌‌ precisely the domain of‌ digital medicine and biomedical data analysis. The axes‌ of research presented above are related to many‌ branches of medicine including cardiology, oncology, urology, neurology,‌ otology, pneumology, radiology, surgery, dermatology, nuclear medicine. Within‌ those branches, the applications cover the following different‌ stages of medicine: prevention, diagnosis, prognosis, treatment.

5‌ Social and environmental responsibility

5.1 Footprint of research‌ activities

An important activity of Epione is to‌ introduce priors from clinical knowledge within data analysis,‌ through geometric information, biophysical models, causality, etc. This‌ enables to develop AI method requiring less data‌ and computations, therefore with a positive impact on‌ the environmental footprint of epione research activity.

6‌ Highlights of the year

6.1 Awards

Xavier Pennec‌ is the laureate of the GSI Achievement Award‌ recognizing outstanding achievement in geometric science of information,‌ received at the 7th international conference on Geometric‌ Science of Information in Saint-Malo on October 31,‌ 2025. This award recognizes his long-term scientific contributions‌ in geometric statistics.
Benjamin Billot received an Outstanding‌ Reviewer Award at the MICCAI 2025 conference. This‌ distinction highlights the quality and dedication of his‌ contributions to the scientific community through peer review.‌
Riccardo Taiello , received the award “Prix de‌ la Victoire de la recherche de la Ville‌ de Nice” from the City of Nice for‌ his thesis “Privacy-preserving machine learning for large-scale collaborative‌ healthcare data analysis”, supervised by Marco Lorenzi and‌ Melek Onen .
Wassila Khatir , co-supervised by‌ Irene Balelli and Marco Lorenzi , received the‌ Outstanding Poster Award at the MEI Center and‌ University Côte d'Azur International Symposium in Osaka,‌ Japan, for her work on "A Multi-omic Integration‌ Approach to Understand the Pathophysiology of Fragile X‌ Syndrome”. She has also been granted for a‌ student fellowship to attend CompSysBio 2025 at Aussois‌ Ski Resort, France, as part of her PhD‌ research.
Maëlis Morier received a Best Poster Award‌ at the SophI.A Summit 2025 Conference, at Sophia‌ Antipolis, for the work “Learning Cardiac Electrophysiology with‌ Graph Neural Networks for Fast Data-driven Personalized Predictions”‌ 51.
Rafael Silva , PhD student under‌ the supervision of Maxime Sermesant , ranked 2nd‌ for the 9th Prix Pierre Laffitte organized by‌ Mines Paris - PSL - Fondation Mines Paris‌. This prize rewards excellence and innovation in‌ research done in partnership with industry.
Best Paper‌ Awards for both Rafael Silva 43 and Buntheng‌ Ly 41 at the Functional Imaging and Modeling‌ of the Heart (FIMH 2025) conference in Dallas,‌ Texas.
Olivier Bisson and Tom Szwagier were nominated‌ for the best papers awards at 7th international‌ conference on Geometric Science of Information for their‌ papers respectively on Log-Euclidean Frameworks for Smooth Brain‌ Connectivity Trajectories 34 and Eigengap Sparsity for Covariance‌ Parsimony 44.

6.2 Promotions

Marco Lorenzi was‌ promoted Research Director (directeur de recherche de deuxième‌ classe DR2).

6.3 Others

Nicholas Ayache was invited‌ to give a plenary talk at the “AI,‌ Science, and Society” conference organized by the French Government as part of‌ the international AI Action‌ Summit program.
This‌‌ year, the longstanding collaboration between Nicholas Ayache and‌ the Brain Institute at‌ Pitié Salpétrière (Dr. O.‌‌ Colliot and Pr. B. Stankoff) on neuroimaging and‌ multiple sclerosis led to‌ the approval of a‌‌ new patent (MyeliGAN) for the generation of synthetic‌ 3D representations of myelin‌ content 64.
Following‌‌ his recruitement last year as Chargé de recherche‌ de classe normale (CRCN)‌ Benjamin Billot was the‌‌ recipient of the Idex attractivity package (R2D2 -‌ Talents Welcome Package) by‌ the Université Côte d'Azur‌‌ for his research on domain randomization for medical‌ image analysis.
Marco Lorenzi‌ published with Maria Zualuaga‌‌ (Eurecom) the book “Trustworthy AI in Medical Imaging”,‌ Elsevier, January 2025 52‌. This book brings‌‌ together researchers, medical experts, and industry partners and‌ aims at tackling trustworthiness‌ while bridging the gap‌‌ between AI research and concrete medical applications.
Maxime‌ Sermesant is a founding‌ member of the newly‌‌ created Society for Artificial Intelligence in Biomedical Imaging‌ (IABM).

7‌ Latest software developments, platforms,‌‌ open data

7.1 Latest software developments

7.1.1 MedINRIA‌

Name:
medInria Suite
Keywords:‌
Visualization, DWI, Health, Segmentation,‌‌ Medical imaging, Python, Web Application, Image registration
Scientific‌ Description:
MedInria aims at‌ creating an easily extensible‌‌ platform for the distribution of research algorithms developed‌ at Inria for medical‌ image processing. This project‌‌ has been funded by the D2T (ADT MedInria-NT)‌ in 2010, renewed in‌ 2012. A fast-track ADT‌‌ was awarded in 2017 to transition the software‌ core to more recent‌ dependencies and study the‌‌ possibility of a consortium creation.The Empenn team leads‌ this Inria national project‌ and participates in the‌‌ development of the common core architecture and features‌ of the software as‌ well as in the‌‌ development of specific plugins for the team's algorithm.‌
Functional Description:
medInria is‌ a software platform for‌‌ the visualisation and processing of medical images. MedInria‌ aims to disseminate Inria's‌ research results in medical‌‌ imaging to clinical, industrial and academic circles. It‌ is now a suite‌ that also includes a‌‌ Python command line version and a web version.‌
Release Contributions:
Python integration,‌ updated dependencies
News of‌‌ the Year:
New release this year with Python‌ integration and updated dependencies.‌
URL:
https://med.inria.fr
Contact:
Maxime‌‌ Sermesant
Participants:
Maxime Sermesant, Olivier Commowick
Partners:
HARVARD‌ Medical School, IHU -‌ LIRYC, NIH

7.1.2 Music‌‌

Name:
Multi-modality Platform for Specific Imaging in Cardiology‌
Keywords:
Medical imaging, Cardiac‌ Electrophysiology, Computer-assisted surgery, Cardiac,‌‌ Health
Functional Description:
MUSIC is a software developed‌ by the Asclepios research‌ project in close collaboration‌‌ with the IHU LIRYC in order to propose‌ functionalities dedicated to cardiac‌ interventional planning and guidance.‌‌ This includes specific tools (algorithms of segmentation, registration,‌ etc.) as well as‌ pipelines. The software is‌‌ based on the MedInria platform.
News of the‌ Year:
new version with‌ new visualisation tools and‌‌ updated python integration
URL:
https://team.inria.fr/asclepios/software/music/
Contact:
Maxime Sermesant‌
Participants:
Florent Collot, Mathilde‌ Merle, Maxime Sermesant
Partner:‌‌
IHU- Bordeau

7.1.3 geomstats‌

Name:
Computations and statistics on manifolds with geometric‌ structures
Keywords:
Geometry, Statistic analysis
Scientific Description:

Geomstats‌ is an open-source Python package for computations and‌ statistics on manifolds. The package is organized into‌ two main modules: “geometry“ and “learning“.

The module‌ `geometry` implements concepts in differential geometry, and the‌ module `learning` implements statistics and learning algorithms for‌ data on manifolds.

The goal is to provide‌ an easily accessible library for learning algorithms on‌ Riemannian manifolds.
Functional Description:
GeomStats is a Python‌ package that performs computations on manifolds such as‌ hyperspheres, hyperbolic spaces, spaces of symmetric positive definite‌ matrices and Lie groups of transformations. It provides‌ efficient and extensively unit-tested implementations of these manifolds,‌ together with useful Riemannian metrics and associated Exponential‌ and Logarithm maps. The corresponding geodesic distances provide‌ a range of intuitive choices of Machine Learning‌ loss functions. The operations implemented in GeomStats are‌ available with different computing backends such as numpy,‌ autograd, pytorch, and tensorflow.
Release Contributions:
- addition‌ of several metrics on the space of full-rank‌ correlation matrices taking advantage of diffeomorphism class, existing‌ Riemannian manifolds, and/or quotient space structure - refactoring‌ of quotient structure in order to treat landmarks,‌ curves, and shapes in an homogenized way, improvement‌ of alignment algorithms in those spaces - addition‌ of varifold metric (on surfaces) by leveraging pykeops‌ - full refactoring of geodesic metric spaces: graph‌ space, wald and BHV spaces, and spider (NB:‌ only BHV explicitly takes advantage of quotient structure,‌ so the renaming) - improvement of numerics: better‌ objects to handle optimization, initial/boundary value problems, finite‌ differences, and interpolation
News of the Year:
The‌ python package geomstats has been enhanced in 2025‌ with several types of log-Euclidean metrics on full‌ rank correlation matrices. This involved an important restructuration‌ to allow the efficient use of pull-back and‌ push-forward metrics between different spaces. Application results to‌ the modeling of the functional brain connectomes were‌ published at GSI 2025 (O. Bisson, Y. Aeschlimann,‌ S. Deslauriers-Gauthier, and X. Pennec. Log-Euclidean Frameworks for‌ Smooth Brain Connectivity Trajectories. In GSI'25 - Int.‌ Conf. on Geometric Science of Information, Saint-Malo (France),‌ France, October 2025).
URL:
https://github.com/geomstats/
Publications:
hal-05165921,‌ hal-04609816, hal-03766900, hal-02536154, hal-02908006,‌ hal-03505132, hal-03160677
Contact:
Xavier Pennec
Participants:
Olivier‌ Bisson, Xavier Pennec, Yann Thanwerdas, Luis Pereira, Anna‌ Calissano, Elodie Maignant, Nina Miolane, Alice Le Brigant‌
Partners:
University of California Santa Barbara, Université Panthéon-Sorbonne‌

7.1.4 Fed-BioMed

Name:
A general software framework for‌ federated learning in healthcare
Keywords:
Federated learning, Medical‌ applications, Machine learning, Distributed Applications, Deep learning
Scientific‌ Description:
While data in healthcare is produced in‌ quantities never imagined before, the feasibility of clinical‌ studies is often hindered by the problem of‌ data access and transfer, especially regarding privacy concerns.‌ Federated learning allows privacy-preserving data analyses using decentralized‌ optimization approaches keeping data securely decentralized. There are‌ currently initiatives providing federated learning frameworks, which are‌ however tailored to specific hardware and modeling approaches, and do not provide‌ natively a deployable production-ready‌ environment. To tackle this‌‌ issue, Fed-BioMed proposes an open-source federated learning frontend‌ framework with application in‌ healthcare. Fed-BioMed framework is‌‌ based on a general architecture accommodating for different‌ models and optimization methods‌
Functional Description:

Fed-BioMed software‌‌ offers a distributed architecture enabling machine learning in‌ healthcare multi-centric studies with‌ a specific focus on‌‌ real world use cases requirements :

- usability‌ : compatible with PyTorch,‌ scikit-learn, MONAI , easy‌‌ control via Jupyter notebook interactive console , experiment‌ control - security :‌ secured communications, model verification,‌‌ secure aggregation, differential privacy - hospital control and‌ governance
News of the‌ Year:
- Deployment of‌‌ Secured Aggregation Schemes - Redesign of Federated Dataset‌ - Improve packaging and‌ portability
URL:
https://fedbiomed.org
Publication:‌‌
hal-02966789v1
Contact:
Marco Lorenzi
Partner:
Université Côte d'Azur‌ (UCA)

8 New results‌

8.1 Medical Image Analysis‌‌ & Machine Learning

8.1.1 Prostate Cancer Detection and‌ Characterization from multiparametric MRI‌

This work was funded‌‌ by the AICOO and DAICAP project in the‌ scope of the Inria‌ APHP joint Bernouilli laboratory‌‌ .

Keywords: Prostate cancer detection; prostate cancer characterization‌.

Participants: Florencia Boccarato‌, Fahym Bounazou,‌‌ Hye Lim Lee, Raphaele Renard-Penna, Hervé‌ Delingette [Correspondant].

Preprocessing‌ of multiparametric MRI is‌‌ essential for automated prostate cancer detection. We propose‌ a method to synthesize‌ high b-value diffusion weighted‌‌ images (DWI) from standard multi-b DWI using an‌ optimized fusion of the‌ apparent diffusion coefficient (ADC)‌‌ maps (see Fig. 4). Evaluations show improved‌ lesion contrast, anatomical detail,‌ and overall image quality‌‌ compared with vendor-synthesized images, while remaining simple, interpretable,‌ and suitable for multicenter‌ datasets.
Accurate localization of‌‌ prostate tumors on multiparametric MRI (mpMRI) is important‌ for diagnosis, treatment planning‌ and communication between urologists,‌‌ radiologists and pathologists. We propose a data-driven approach‌ to automatically determine the‌ main prostate sector associated‌‌ with a given lesion. We evaluate the optimized‌ sectorization against the PI-RADS‌ v1 and v2.1 standards.‌‌
We have worked on the collection and curation‌ of the multi-centric database‌ DAICAP involving 8 different‌‌ university hospitals in France. We have developed quality‌ control processes and started‌ processing the data from‌‌ the Health Data Hub.

Figure 4: Pipeline‌ of acquisition-aware high- $b‌$ DWI synthesis.

8.1.2 Analysis of European National Health‌ data to study the‌ outcomes of patients with‌‌ vascular diseases

This work was partially funded by‌ 3IA Côte d'Azur.

Keywords:‌ Data extraction; SNDS; Causal‌‌ inference.

Participants: Amel Bakhouche [Correspondant], Hervé‌ Delingette, Juliette Raffort-Lareyre‌, Irene Balelli.‌‌

Clinical outcomes after varicose‌ vein surgery remain heterogeneous and difficult to predict.‌ We developed ML models on the QUALIVEIN cohort,‌ a French vascular database that prospectively collects detailed‌ operative information on patients treated for chronic venous‌ insufficiency (CVI), to predict 90-day clinical improvement and‌ identify key predictors (see Fig. 5).

We‌ are now analyzing data from the Système National‌ des Données de Santé (SNDS), extracting a 10-year‌ cohort to predict short- and long-term outcomes, study‌ the evolution of surgical techniques, and assess their‌ impact using causal inference methods.

Figure 5:‌ Feature importance plot showing the most influential variables‌ for the Random Forest model. BMI= Body Mass‌ index; CEAP= Clinical, Etiological, Anatomical, and Pathophysiological (CEAP)‌ classification; VCSS= Venous Clinical Severity score.

8.1.3 Spatial regularization‌ for improved accuracy and interpretability in keypoint-based registration‌

This work has been funded by the French‌ government, through the 3IA Cote d'Azur Investments in‌ the project managed by the National Research Agency‌ (ANR) with the reference number ANR-23-IACL-0001. Further support‌ has come from NIH NIBIB 1R01EB036945, NIH NICHD‌ 1R01HD114338, NIH NIBIB 1R01EB032708, MIT Jameel Clinic, MIT‌ CSAIL-Wistron Program.

Keywords: Spatial regularization; Interpretability; Unsupervised image‌ registration.

Participants: Benjamin Billot [Correspondant], Ramya‌ Muthukrishnan, Esra Abaci Turk, Ellen Grant‌, Nicholas Ayache, Hervé Delingette, Polina‌ Golland.

Unsupervised keypoint-based registration seeks to improve‌ interpretability while alleviating supervision requirements. Yet, the extracted‌ features often are hardly interpretable, thus undermining the‌ purpose of this very method.
We propose a‌ three-fold loss to regularize the features' spatial distributions:‌ a Kullback-Leibler (KL) divergence to model features as‌ interpretable point spread functions, a Frobenius norm on‌ the spatial covariance for sharpness, and a novel‌ repulsive loss to encourage spatial diversity in keypoints‌ (Fig. 6).
This regularization greatly improves the‌ interpretability of the keypoints, as well as the‌ overall accuracy of unsupervised keypoint-based registration by now‌ bridging the gap with state-of-the-art supervised methods 33‌.
This work is a collaboration between Inria,‌ MIT CSAIL, and the Boston Children's Hospital.

Figure‌ 6: (A) Registration by unsupervised keypoint detection.‌ Beyond the classical similarity loss between fixed and‌ moved volumes (leading to features comparable to B,left),‌ we propose a three-fold spatial regularization, whose effects‌ on the features/keypoints are depicted in (B-D): interpretability,‌ precision, and diversity.

8.1.4 Resource-efficient Automatic Refinement‌ of Segmentations via Weak‌ Supervision from Light Feedback‌‌

This work was supported by an Inserm-Inria funding.‌

Keywords: Weak supervision; Efficient‌ segmentation correction; Reaching clinical‌‌ accuracy.

Participants: Alix de Langlais [Correspondant],‌ Benjamin Billot, Marc-Olivier‌ Gauci, Hervé Delingette‌‌.

Foundation models enable automated segmentation but may‌ sometimes fail to reach‌ clinical accuracy. Existing refinement‌‌ methods, however, require either strong supervision or extensive‌ user interaction. To address‌ this, we present SCORE,‌‌ a weakly supervised framework that refines segmentations using‌ only light feedback based‌ on region wise quality‌‌ scores and segmentation error labels (Fig. 7).‌ On humerus CT scans,‌ SCORE improves TotalSegmentator performance‌‌ while greatly reducing annotation effort.

Figure 7:‌ Overview of SCORE, our‌ weakly supervised framework to‌‌ refine segmentation from an external tool using only‌ light feedback. The network‌ takes as input a‌‌ 3D image, its initial segmentation, and a probability‌ map for additional edge‌ priors.

8.1.5‌‌ Segmentation of Fractured Bones from CT

This work‌ was funded by the‌ French National Research Agency‌‌ (ANR), through the project RHU ReBone ANR-23-RHUS-0011

Keywords:‌ Bone fractures; CT imaging‌.

Participants: Hervé Delingette‌‌ [Correspondant], Hari Sreedhar, Marc-Olivier Gauci.‌

The project RHU ReBone‌ is a research initiative‌‌ aiming to improve pre-operative planning of surgical reduction‌ in complex fracture cases‌ of traumatic bone fractures.‌‌ The contributions of Work Package 1 in 2025‌ were to:

Help establish‌ the inclusion criteria and‌‌ descriptions of the three fracture types of interest‌ (distal radius fractures, tibial‌ plateau fractures, and acetabular‌‌ fractures)
Prepare the documentation and data management plans‌ for compliance with GDPR‌ regulations concerning the use‌‌ of medical data
Initiate the automatic segmentation algorithms‌ of bone fragments and‌ for initial fracture reduction‌‌ planning

8.1.6 TBDM: Temporal Boundary Distillation Module for‌ Surgical Gesture Segmentation

This‌ work was funded by‌‌ 3IA Cote d'Azur.

Keywords: Surgical gesture segmentation; Action‌ boundary modeling; Knowledge distillation‌.

Participants: Ezem Sura‌‌ Ekmekci [Correspondant], Sebastien Frey, Snehashis Majhi‌, Khodor Hamadi,‌ Hervé Delingette, Matthieu‌‌ Durand, Pierre Berthet-Rayne, François Bremond,‌ Nicholas Ayache.

Collaboration:‌ This work was conducted‌‌ in collaboration between Inria, CHU Nice, and Caranx‌ Medical.
Challenge Addressed: Developed‌ a solution for precise‌‌ temporal localization of surgical‌ gesture boundaries and transitions in robot-assisted surgery videos‌ 19.
Key Innovation: Introduced the Temporal Boundary‌ Distillation Module (TBDM), a framework that explicitly models‌ action transitions using RGB-only video without requiring kinematic‌ data or additional annotations.
Technical Approach: Implemented knowledge‌ distillation with cross-attention mechanisms to learn boundary-aware features‌ during training, with zero computational overhead at inference‌ (Fig. 8).
Validation and Results: Our method‌ achieved up to 8.5 edit score improvement on‌ CholecT50 dataset. It also obtained state-of-the-art performance on‌ RARP-45 (81.4 edit score, 77.9 F1-50), and demonstrated‌ consistent improvements across multiple baseline architectures.
Impact: Created‌ a generalizable, plug-and-play framework applicable to various surgical‌ video analysis tasks.

Figure 8: Overview of‌ TBDM framework for Surgical Gesture Segmentation. During training,‌ the boundary blocks generate boundary-aware features to guide‌ the projection layer. During inference, only the trained‌ projection layer is used, adding no extra cost,‌ while achieving significant boundary precision in segmentation.

8.1.7 Multi-stage CNN for fast registration of‌ 3D preoperative CTs to 2D intraoperative X-rays

This‌ work was funded by 3IA Cote d'Azur.

Keywords:‌ Pose estimation; Multi-stage CNN; Image guidance.

Participants:‌ Federica Facente [Correspondant], Benjamin Billot, Vivek‌ Gopalakrishnan, Manasi Kattel, Wen Wei,‌ Polina Golland, Hervé Delingette, Nicholas Ayache‌, Pierre Berthet-Rayne.

In this work, we‌ present LXPose (Live X-Ray Pose Estimation) 50,‌ a real-time multi-stage convolutional neural network (CNN) for‌ accurate registration of 3D preoperative CTs with 2D‌ intraoperative X-rays. The method estimates the X-ray source‌ pose to generate a synthetic X-ray from the‌ CT that best aligns with the input image,‌ as illustrated in Fig. 9A. This is‌ achieved using a multi-stage CNN that progressively estimates‌ the pose of the C-arm in the CT‌ scanner's coordinate system (Fig. 9B).

Figure 9‌: (A) Method overview. LXPose performs CT/X-ray registration‌ by progressively estimating the pose of the X-ray‌ source from which 2D projections of the CT‌ best align with the input X-ray. (B) LXPose‌ architecture. A first CNN regresses the pose parameters‌ from a 2D input image. This initial pose‌ is then used to simulate a corresponding synthetic‌ X-ray. Finally, the latter is used by a‌ second CNN along with the input image to‌ predict a corrective term for final pose estimation.‌

8.1.8‌‌ Exploring Variability in Medical Image Segmentation: New Metrics‌ and Frameworks

This work‌ was funded by the‌‌ MediTwin project and supported by the French government‌ under the France 2030‌ initiative.

Keywords: Interobserver variability;‌‌ Medical image segmentation; Deep learning.

Participants: Bernhard‌ Föllmer [Correspondant], Hervé‌ Delingette.

The aim‌‌ of this project is to develop a statistical‌ framework and novel metrics‌ for assessing interobserver variability‌‌ in medical image segmentation. The project consists of‌ three pillars (Fig. 10‌):

Develop novel segmentation‌‌ metrics for multi-rater comparison
Develop MaskStat—an open-source toolkit‌ for segmentation statistics
Evaluate‌ metrics in real-world use‌‌ cases and provide practical recommendations

The project will‌ enable rigorous performance and‌ quality assessment for medical‌‌ image segmentation in the presence of multiple observers.‌

Figure 10: Overview‌ of the three project‌‌ pillars: (1) development of novel segmentation metrics for‌ interobserver comparison, (2) MaskStat,‌ an open-source toolkit for‌‌ segmentation statistics, and (3) evaluation in practical applications‌ and formulation of recommendations.‌

8.1.9 AI-based precision oncology‌ to monitor response of‌ metastatic cancer to immunotherapy‌‌ using PET/CT imaging

This work was funded by‌ 3IA Côte d'Azur.

Keywords:‌ Immunotherapy; PET/CT imaging; Precision‌‌ oncology; Artificial intelligence.

Participants: Giulia Foroni [Correspondant]‌, Marco Lorenzi,‌ Olivier Humbert.

This‌‌ is a collaborative project between INRIA and CAL.‌ The goal is to‌ predict survival outcomes of‌‌ metastatic cancer patients treated with immunotherapy (Fig. 11‌). During the first‌ part of this project,‌‌ we reviewed classical survival models in both static‌ and longitudinal settings using‌ data from CAL center,‌‌ and we started examining advanced approaches to survival‌ analysis:

Development of end-to-end‌ clustering pipelines integrated with‌‌ survival analysis.
Structuring the data through graph-based representations.‌

Figure 11: Pipeline‌ for cox proportional hazards‌‌ model on static and longitudinal tabular data.

8.1.10‌ Development of predictive models in patients with Peripheral‌ Artery Disease

This work was partially funded by‌ the Horizon-Europe project VASCUL-AID (ID 101080947).

Keywords: Peripheral‌ artery disease; Angiography; Competing survival modeling.

Participants:‌ Sébastien Goffart [Correspondant], Odette Hart, Fabien‌ Lareyre, Lisa Guzzi, Kak Khee Yeung‌, Hervé Delingette, Manar Khashram, Juliette‌ Raffort.

We benchmarked six machine learning and‌ deep learning survival models to predict amputation-free survival‌ in 2,366 patients with peripheral artery disease undergoing‌ revascularization 20. Non-linear models achieved similar discrimination‌ to Cox regression, while the DeepSurv model (NLCH)‌ showed improved calibration. Competing risk approaches were evaluated,‌ and key clinical predictors were identified using SHAP-based‌ feature importance 21. Model performance is summarized‌ in Figure 12.

Figure 12: Survival‌ model performance across five-fold cross-validation in 483 patients:‌ (A) discrimination assessed by the concordance index; (B)‌ calibration assessed by the integrated Brier score. Model‌ comparisons were performed using Student's t-test (p >‌ 0.05).

8.1.11 Automatic‌ Segmentation of Lower-Limb Arteries on CTA for Pre-surgical‌ Planning of Peripheral Artery Disease.

This work was‌ funded by 3IA Côte d'Azur.

Keywords: Peripheral artery‌ disease; Image segmentation; Computed tomography angiography; Deep learning‌.

Participants: Lisa Guzzi [Correspondant], Maria A.‌ Zuluaga, Fabien Lareyre, Gilles Di Lorenzo‌, Sébastien Goffart, Andrea Chierici, Riccardo‌ Taiello, Juliette Raffort, Hervé Delingette.‌

Our work aims to develop and evaluate deep‌ learning methods to segment lower-limb arteries in peripheral‌ artery disease (PAD) using computed tomography angiography (CTA)‌ to support pre-surgical planning (see Fig. 13).‌

In 38, we proposed regional Hausdorff distance‌ loss functions for medical image segmentation, achieving state-of-the-art‌ performance without auxiliary losses.
In 37, we‌ applied state-of-the-art segmentation methods to automatically segment lower-limb‌ arteries in CTA images for PAD pre-surgical planning.‌

Figure 13: Overview of the automatic segmentation‌ of PAD-relevant structures on CTA images.

8.1.12 Segmentation of Abdominal‌ Aortic Aneurysm from CT‌ Angiography

This work was‌‌ partially funded by the ANR project PREDICTA ANR-22-CE45-0023‌

Keywords: Peripheral artery disease;‌ Angiography; Competing survival modeling‌‌.

Participants: Jia Guo [Correspondant], Fabien Lareyre‌, Hervé Delingette,‌ Juliette Raffort.

We‌‌ are developing an algorithm for the segmentation of‌ abdominal aortic aneurysm from‌ CT angiography which is‌‌ essential for diagnosis, risk stratification and endovascular aortic‌ repair planning. Compared to‌ prior work 23,‌‌ our objective is to handle both ruptured and‌ non ruptured aneuryms and‌ to extract relevant radiomics‌‌ and geometrical features that can predict surgical outcome‌ (Fig. 14).

Figure‌ 14: Ground truth‌‌ annotations of anatomical structures surrounding abdominal aortic aneurysms‌ used for our segmentation‌ algorithm.

8.1.13 A Scalable Spatio-Temporal Atlas of Neurodegenerative Brain‌ Changes

This work was‌ funded by PEPR Santé‌‌ Numerique (project Rewind).

Keywords: Longitudinal disease progression; Neural‌ cellular automata.

Participants:‌ John Kalkhof [Correspondant],‌‌ Marco Lorenzi.

In this project we developed‌ SMART (Fig. 15),‌ a flexible and interpretable‌‌ spatio-temporal brain atlas framework for modeling longitudinal disease‌ progression in MRI. The‌ core contribution is part‌‌ of a scientific work that introduces a model‌ which:

Disentangles population-level disease‌ dynamics from subject-specific anatomical‌‌ changes.
Learns region-wise ODE trajectories with subject-specific temporal‌ alignment.
Generates anatomically coherent,‌ diffeomorphic deformations using a‌‌ conditioned multi-scale Neural Cellular Automata.

Figure 15:‌ Framework for Alzheimer's progression‌ forecasting: (A) infer disease‌‌ stage and predict change over dt; (B) apply‌ region-wise signals voxelwise to‌ condition MED-NCA and estimate‌‌ a deformation field to the next timepoint; (C)‌ results show expected changes,‌ notably ventricular expansion and‌‌ hippocampal atrophy.

8.1.14‌ MRI-TRUS prostate registration

This‌‌ work was funded by the French government, through‌ the 3IA Cote d'Azur‌ Investments in the project‌‌ managed by the National Research Agency (ANR) with‌ the reference number ANR-23-IACL-0001.‌

Keywords: MRI-TRUS rigid registration;‌‌ Learning-based regression; ICP.‌

Participants: Manasi Kattel [Correspondant], Federica Facente,‌ Benjamin Billot, Hervé Delingette, Nicholas Ayache‌.

MUReg is a fully automated pipeline for‌ rigid MRI-TRUS registration in prostate cancer biopsy 39‌.
Translation is initialized by aligning prostate mask‌ centers obtained with 3D UNets. Rotation is estimated‌ using an attention-based CNN trained on segmentations, overcoming‌ MRI-TRUS domain gaps and prostate spheroidal symmetry with‌ a novel bounding box vertices registration erroe (BBVRE)‌ displacement-based loss. Alignment is refined using the iterative‌ closest points (ICP) algorithm (Figure 16).
MUReg‌ significantly outperforms state-of-the-art methods on a large clinical‌ dataset.
Work was done in collaboration with industrial‌ partner Koelis.

Figure 16: Overview of the‌ MUReg pipeline for 3D MRI-TRUS rigid registration, including‌ segmentation-based translation initialization, attention-based CNN rotation estimation from‌ masks, and final refinement using ICP.

8.1.15 Data Exfiltration and Data‌ Anonymization

This project has been supported by the‌ French government, through the National Research Agency (ANR)‌ 3IA Côte d'Azur and IA Cluster project (ANR-19-3IA-0002‌ and ANR23-IACL-0001).

Keywords: Data exfiltration by compression attack;‌ Medical image anonymization; Privacy; Security.

Participants: Huiyu‌ Li [Correspondant], Nicholas Ayache, Hervé Delingette‌.

We introduce a novel data exfiltration attack‌ (Fig. 17A), named Data Exfiltration by Compression‌ 25 to reveal the potential data leackage from‌ a healthcare data lake.
We address the medical‌ image anonymization problem with a two-stage solution (Fig.‌ 17B): latent code projection and optimization 40‌ to protect the data privacy from the outset.‌

Figure 17: A. Privacy Threats and Protection‌ in Data Lake. For the Data Exfiltration Attack,‌ the attacker first encodes the target data into‌ compression codes Z. Then, the codes and the‌ network are exported. Finally, a decoder decompresses the‌ codes to reconstruct the original data. B. Overview‌ of the proposed method, comprising two main stages:‌ (1) latent code projection, and (2) Latent code‌ optimization to balance the trade-off between utility preservation‌ and privacy protection.

8.1.16 Lung Cancer Risk‌ Prediction From a Single Low-Dose Chest Computed Tomography‌

Keywords: Lung cancer risk; Low-dose computed tomography (LDCT);‌ Deep learning.

Participants: Huiyu Li [Correspondant], Benjamin Billot, Rima‌ Guettache, Maud Collomb‌, Adeline Champrigaud,‌‌ Isabelle Calléa, Julien Dinkel, Charles Marquette‌, Hervé Delingette.‌

We employ a validated‌‌ deep learning model (Sybil) to estimate lung cancer‌ risk over a 6-year‌ horizon using three real-world‌‌ hospital datasets (Fig. 18).
We conduct a‌ comprehensive analysis of the‌ prediction results to identify‌‌ patients who may benefit from increased clinical attention‌ and follow-up.

Figure 18‌: Overview of the‌‌ Lung Cancer Risk Prediction Project.

8.1.17 Thyrosonics

This project‌ has received funding through‌‌ BoostUrCAreer from the European Union's Horizon 2020 research‌ and innovation program under‌ grant agreement 847581. It‌‌ has been co-funded by the Region Provence-Alpes-Côte d'Azur‌ and IDEX UCA/JEDI.

Keywords:‌ Thyroid Cancer; Ultrasound.‌‌

Participants: Hari Sreedhar [Correspondant], Hervé Delingette,‌ Guillaume Lajoinie, Charles‌ Raffaelli.

This project‌‌ investigates diverse aspects of thyroid ultrasound, including inter-expert‌ variability
A multi-center study‌ of thyroid ultrasound was‌‌ conducted, and submitted to the European Thyroid Journal.‌

Figure 19: Examples‌ of thyroid nodule images‌‌ generating inter-expert variability, here in the case of‌ describing echogenicity.

8.1.18 Mitigating Data Exfiltration Attacks‌ through Layer-Wise Learning Rate‌ Decay Fine-Tuning

This work‌‌ was supported by Région Sud and France 2030‌ through the I-Démo project‌ PLICIA, and by the‌‌ French National Research Agency (ANR) under the IA‌ Cluster project ANR-23-IACL-0001.

Keywords:‌ Data lake security; Data‌‌ exfiltration mitigation.

Participants: Elie Thellier [Correspondant],‌ Huiyu Li, Nicholas‌ Ayache, Hervé Delingette‌‌.

We study data exfiltration attacks in medical‌ data lake-trained models, where‌ adversaries embed latent patient‌‌ information into model parameters for later reconstruction. We‌ propose a new method‌ to mitigate those attacks‌‌ that can be applied before any model export.‌ It is based on‌ fine-tuning with a decaying‌‌ layer-wise learning rate that corrupts embedded data while‌ preserving task performance 46‌ (Fig. 20). Experiments‌‌ show strong robustness against state-of-the-art attacks.

Figure 20‌: Overview of the‌ export-time mitigation of data‌‌ exfiltration. A data lake-trained model may encode both‌ utility and private data.‌ Layer-wise learning rate decay‌‌ fine-tuning applied at export‌ disrupts memorization while preserving task performance, yielding a‌ usable model without reconstruction ability.

8.1.19 Disease Progression Modeling and Stratification for‌ detecting sub-trajectories in the natural history of pathologies‌

The work has been supported by the Michael‌ J. Fox Foundation for Parkinson's Research (MJFF), and‌ to the French government, through the 3IA Côte‌ d'Azur Investments in the Future project managed by‌ the National Research Agency (ANR) with the reference‌ number ANR-19-P3IA- 0002, by the TRAIN project ANR-22-FAI1-0003-02,‌ and by the ANR JCJC project Fed-BioMed 19-CE45-0006-01.‌

Keywords: Disease progression modeling; Expectation maximization; Disease subtyping‌.

Participants: Alessandro Viani, Emile d'Angremont,‌ Boris Gutman, Marco Lorenzi [Correspondant].

Development‌ of the Disease Progression Modeling and Stratification (DPMoSt)‌ method, designed to analyze biomarker sensitivity in subpopulations.‌
Application of DPMoSt to the Parkinson's disease on‌ both ENIGMA and PPMI datasets.
Application of DPMoSt‌ to the Alzheimer's disease on ADNI dataset. Notably,‌ this analysis revealed an association between APOE4 and‌ accelerated cognitive decline (Fig. 21).
Presentation of‌ the results at the Longitudinal Disease Tracking and‌ Modeling with Medical Images and Data workshop, part‌ of the MICCAI conference.

Figure 21: DPMoSt-estimated‌ biomarker trajectories derived from the ADNI dataset. The‌ y-axis represents biomarker severity, while the x-axis denotes‌ re-parameterized time. Dots indicate individual longitudinal measurements for‌ each subject. Solid lines show the estimated trajectories,‌ with shaded bands representing standard deviations. Colors distinguish‌ the two subtypes identified by the model. Each‌ plot title displays the probability of biomarker specificity‌ for the corresponding subtype.

8.2 Imaging & Phenomics, Biostatistics

8.2.1 Cardiac Electromechanical‌ Model Sensitivity Analysis using Causal Discovery

This work‌ has received funding from the European Union Horizon‌ 2020 Research and Innovation Program SimCardioTest (101016496), from‌ the French government through the National Research Agency‌ (ANR) projects PEPR Digital Health ChroniCardio (22-PESN-0015), from‌ ANR under the France 2030 project RHU Talent‌ (ANR-23-RHUS-0015), and from 3IA Côte d'Azur and IA‌ Cluster (ANR-19-3IA-0002 and ANR-23-IACL-0001).

Keywords: Sensitivity analysis; Causal‌ discovery; Electromechanical properties; Heart modeling.

Participants: Safaa‌ Al-Ali [Correspondant], Jairo Rodriguéz Padilla, Maxime‌ Sermesant, Irene Balelli.

In 32,‌ we propose a causal discovery-based pipeline for global sensitivity analysis of a‌ cardiac electromechanical model. The‌ method identifies and quantifies‌‌ the relationships between the model parameters and the‌ key output of interest:‌ ejection fraction and pressure‌‌ change in the left ventricular cavity. It provides‌ a precise identification of‌ the key parameters to‌‌ be focused on, and ensures stable results compared‌ to classical global sensitivity‌ analysis methods, and despite‌‌ a limited number of available simulations. Figure 22‌ shows the proposed sensitivity‌ analysis pipeline.

Figure 22‌‌: Overview of the proposed pipeline. Model parameters‌ are sampled using Latent‌ Hypercube Sampling method (LHS)‌‌ and the electromechanical model computes the biomarkers of‌ interest (the Ejection Fraction‌ and maximum pressure in‌‌ the left ventricular cavity). Prior knowledge can be‌ incorporated to estimate a‌ weighted causal graph using‌‌ the causal discovery method. Causal weights obtained within‌ the graph quantify sensitivity‌ and sign effects. Clustering‌‌ and classification are finally used to validate the‌ causal relationships and their‌ interpretation.

8.2.2 Association of mtDNA variants and phenotype in‌ mitochondrial diseases with multi-OMICS‌ approaches

This work was‌‌ funded by ANR MITOMICS.

Keywords: Machine Learning; Multi-omics;‌ Mitochondrial diseases.

Participants:‌ Eléonore Birgy [Correspondant],‌‌ Marco Lorenzi, Cécile Rouzier.

Mitomics is‌ a collaborative project between‌ INRIA Epione, CHU de‌‌ Nice, CHU d'Angers et Université de Nice et‌ de Nantes aiming to‌ better understand mitochondrial diseases‌‌ through the creation of a multiomics data collection‌ network and the development‌ of a database Mitomatcher.‌‌ Mitochondrial diseases are heterogeneous because both the nuclear‌ and mitochondrial genomes are‌ involved. The goal is‌‌ to better understand the molecular mechanisms responsible for‌ the clinicogenetic heterogenesis of‌ mitochondrial diseases through the‌‌ co-occurrence of variants, multi-omics data, and the use‌ of innovative in silico‌ tools.

Work on human‌‌ phenotype ontology (HPO, Protégé, Sparql) and creation of‌ clinic groups from simplified‌ ontology (Figure 23).‌‌
Modeling: spectral clustering approaches helps us to classify‌ patients in different groups‌ corresponding to clusters, related‌‌ with severity forms. Analysis in progress with more‌ included patient in the‌ database to study the‌‌ association with genetic variants.
Multi-OMICS: adaptation and validation‌ of OUTRIDER package for‌ analyzing transcriptomics data.

Figure‌‌ 23: WP2: Spectral clustering gives 2 clusters‌ related to age of‌ patients and clinical pattern,‌‌ related to different severity forms of the disease.‌ WP4: validation of outrider‌ to detect aberrant expression‌‌ in control group. 2.

8.2.3 Deciphering Fragile X Syndrome‌ via Multi-Omic Integration

This work was funded by‌ the Neuromod Institute.

Keywords: Fragile X syndrome; Multi-omics;‌ Translation; Bioinformatics.

Participants: Wassila Khatir [Correspondant],‌ Irene Balelli, Marco Lorenzi, Carole Gwizdek‌.

We applied a Multi-Channel Variational Autoencoder (MCVAE)‌ to integrate transcriptomic (TP) and translatomic (TL) datasets‌ from wild-type mice, aiming to capture physiological communication‌ between these omic layers. The model was then‌ tested on Fmr1-knockout mice, which lack the‌ RNA-binding protein FMRP. FMRP normally regulates translation, and‌ its absence leads to disrupted TP-TL communication, causing‌ Fragile X Syndrome (FXS), a neurodevelopmental disorder. The‌ MCVAE identified anomalies reflecting coordinated molecular dysregulation in‌ the knockout samples.

Key findings include:

FMRP target‌ enrichment confirmed that detected molecular perturbations overlap significantly‌ with published FMRP-bound transcripts.
Functional annotation highlights indirect‌ effects of Fmr1 deficiency.
Pathway analysis shows enrichment‌ in neurodevelopmental and synaptic pathways, including synaptic plasticity‌ and axon guidance, consistent with Fragile X Syndrome‌ mechanisms.
The full workflow and results are summarized‌ in Figure 24.

Figure 24: Multi-omic‌ analysis of Fragile X Syndrome using MCVAE. Top:‌ (A) transcriptomic (TP) and translatomic (TL) layers coordinated‌ by FMRP and (B) cascading effect of lack‌ of FMRP in Fragile X syndrome; bottom: workflow‌ including data collection, preprocessing, MCVAE integration, and anomaly‌ detection in knockout samples.

8.2.4 Consensus Enhances Individual‌ Causal Models: a Use Case on Lung Cancer‌

This work was carried out with the support‌ of Inria's institutional funding.

Keywords: Causal discovery; Uncertainty;‌ Cancer genomics.

Participants: Arnaud Lang [Correspondant],‌ Rodrigo Ramos, Safaa Al-Ali, Mohammad Mousavi‌, Anna Calissano, Irene Balelli.

Discovering‌ causal relationships in real-world medical data remains challenging‌ due to the strong assumptions required by classical‌ causal discovery algorithms, which aim to infer directed‌ graphs whose edges represent cause-effect relationships between features.‌ In this work, we propose a consensus causal‌ model that aggregates multiple causal discovery methods to‌ improve robustness and reliability. We apply this approach‌ to a lung cancer dataset combining patient characteristics,‌ tumor data, and genetic mutations. The resulting consensus‌ causal graph (see Fig. 25, right graph)‌ captures biologically validated causal relationships that individual algorithms fail to identify, highlighting‌ the relevance of consensus-based‌ causal discovery in complex‌‌ biomedical settings. This work has been submitted.

Figure‌ 25: (Left) Output‌ of a single algorithm‌‌ (PC) and (Right) our consensus causal graph. Light‌ orange nodes correspond to‌ patient and tumor level‌‌ variables, while dark orange nodes contain information about‌ genetic mutations affecting cellular‌ pathways. Dashed edges are‌‌ undirected edges; edge color intensity in the consensus‌ causal model corresponds to‌ the frequency of the‌‌ edge occurrence and yields a heuristic measure of‌ uncertainty regarding the causal‌ association (darker (resp. lighter)‌‌ edge = higher (resp. lower) occurrence).

8.3‌ Computational Anatomy & Geometric‌‌ Statistics

8.3.1 Log-Euclidean frameworks for smooth brain connectivity‌ trajectories

This work was‌ funded by ERC grant‌‌ #786854 (G-Statistics, European Research Council, Horizon 2020) and‌ by the French government‌ through the 3IA Côte‌‌ d'Azur Investments ANR-23-IACL-0001 (managed by ANR).

Keywords: Functional‌ connectomes; Correlation matrices; Log-Euclidean‌ geometry.

Participants: Olivier‌‌ Bisson [Correspondant], Yanis Aeschlimann, Samuel Deslauriers-Gauthier‌, Xavier Pennec.‌

Intrinsic polynomial regression of‌‌ longitudinal connectomes in ${Cor}^{+} (n)‌$ via Log-Euclidean diffeomorphisms (Off-Log‌ / Log-Scaling) 34:‌‌ perform the regression in the Euclidean image and‌ pull back to preserve‌ correlation constraints (Fig. 26‌‌). Related work: we develop a unified theory‌ of log-Euclidean Lie groups‌ on $S^{+} (‌‌ n)$ and ${Cor}^{+} (n)‌$ , and show that‌ all log-Euclidean metrics in‌‌ a fixed dimension are Riemannian-isometric via explicit isometries.‌ We further characterize quotients‌ of log-Euclidean Lie groups‌‌ 57.

Figure 26: Dynamic functional connectome‌ trajectory and its intrinsic‌ polynomial regression (Log-Euclidean pullback),‌‌ visualized after 3D PCA.

8.3.2‌‌ Eigengap sparsity for covariance parsimony

This work was‌ supported by the ERC‌ grant #786854 G-Statistics from‌‌ the European Research Council under the European Union's‌ Horizon 2020 research and‌ innovation program and by‌‌ the French government through the 3IA Côte d'Azur‌ Investments ANR-23-IACL-0001 managed by‌ the National Research Agency.‌‌

Keywords: Covariance estimation; Parsimony;‌ Eigengaps; Flag manifolds; Monotone cone; Isotonic regression.‌

Participants: Tom Szwagier [Correspondant], Guillaume Olikier,‌ Xavier Pennec.

We propose a parsimonious covariance‌ estimator leveraging the stratification of symmetric matrices by‌ the multiplicities of the eigenvalues 44. It‌ involves solving a penalized log-likelihood optimization problem via‌ a projected gradient descent on a monotone cone.‌ The algorithm, illustrated in Figure 27, turns‌ out to draw an interesting link between covariance‌ parsimony and shrinkage.

Figure 27: Illustration of‌ the projected gradient descent algorithm from the double‌ point of view of the constraint set (monotone‌ cone, top) and the eigenvalue profile (scree plot,‌ bottom). The gradient step makes eigenvalues mutually-attracted, up‌ to the point where some of them violate‌ the ordering constraints, going outside of the monotone‌ cone (gray zone). The projection onto the constraint‌ set–which is nothing but an isotonic regression–then boils‌ down to equalizing the unsorted eigenvalues. This automatically‌ induces covariance parsimony.

8.3.3 Parsimonious‌ Gaussian mixture models with piecewise-constant eigenvalue profiles

This‌ work was supported by the ERC grant #786854‌ G-Statistics from the European Research Council under the‌ European Union's Horizon 2020 research and innovation program‌ and by the French government through the 3IA‌ Côte d'Azur Investments ANR-23-IACL-0001 managed by the National‌ Research Agency.

Keywords: Gaussian mixture models; Parsimonious models;‌ Eigenvalues; EM algorithm; Model selection.

Participants: Tom‌ Szwagier [Correspondant], Pierre-Alexandre Mattei, Charles Bouveyron‌, Xavier Pennec.

We introduce a new‌ family of parsimonious Gaussian mixture models with piecewise-constant‌ covariance eigenvalue profiles 28. These extend several‌ low-rank models like the celebrated mixtures of probabilistic‌ principal component analyzers. To address the notoriously-challenging issue‌ of jointly learning the mixture parameters and hyperparameters,‌ we propose a provably-monotonous component-wise penalized expectation–maximization algorithm‌ (see Figure 28). Our models achieve superior‌ likelihood–parsimony tradeoffs on a variety of unsupervised experiments.‌

Figure 28: Parsimonious density fitting with the‌ mixtures of principal subspace analyzers. Left: evolution of‌ the mixture parameters over the iterations, with increasing‌ opacity. Right: evolution of the penalized log-likelihood (top)‌ and the number of parameters (bottom) over the‌ iterations.

8.3.4 Rethinking‌ statistical methods with flags

This work was supported by the ERC grant‌ #786854 G-Statistics from the‌ European Research Council under‌‌ the European Union's Horizon 2020 research and innovation‌ program and by the‌ French government through the‌‌ 3IA Côte d'Azur Investments ANR-23-IACL-0001 managed by the‌ National Research Agency.

Keywords:‌ Subspace learning; Grassmann manifolds;‌‌ Flag manifolds; Nested subspaces; Dimensionality reduction; Parsimony; Principal‌ component analysis; Riemannian geometry;‌ Statistics.

Participants: Tom‌‌ Szwagier [Correspondant], Xavier Pennec.

The PhD‌ thesis of Tom Szwagier‌ from Univ. Côte d'Azur‌‌ 55 (defended on November 28, 2025) aimed at‌ establishing the interest of‌ flag manifolds in statistics.‌‌ It comprises contributions on: an efficient algorithm for‌ the Riemannian logarithm on‌ flag manifolds, the principal‌‌ subspace analysis methodology for covariance matrices with close‌ eigenvalues 29 and its‌ l1-relaxation 44, a‌‌ Bayesian inference framework (in preparation), an extension to‌ Gaussian mixture models 28‌ and an extension beyond‌‌ Gaussian distributions 61.

This last contribution develops‌ the idea of replacing‌ Grassmanians (subspaces) with flags‌‌ in subspace learning methods beyond PCA. We propose‌ a simple and easily‌ implementable principle (the flag‌‌ trick) to enforce nestedness of subspaces. The‌ flag trick consists in‌ lifting Grassmannian optimization criteria‌‌ to flag manifolds—the space of nested subspaces of‌ increasing dimension—via nested projectors‌ (see Figure 29).‌‌ We apply the flag trick to several classical‌ machine learning methods and‌ show that it successfully‌‌ addresses the nestedness issue in subspace learning.

Figure‌ 29: Illustration of‌ the flag trick methodology.‌‌ A $q$ -dimensional subspace learning problem is converted‌ into a multilevel nested‌ subspace learning problem and‌‌ addressed via a steepest descent algorithm on flag‌ manifolds. The estimators associated‌ with each subspace can‌‌ then be aggregated via ensembling methods and yield‌ improved predictions.

8.4‌‌ Computational Cardiology & Image-Based Cardiac Interventions

8.4.1 Cardiac‌ Electrophysiology Model Personalization

This‌ work was funded by‌‌ BPI i-démo MediTwin.

Keywords: Cardiac electrophysiology modeling; Lattice-Boltzmann‌ method; Model personalization.‌

Participants: Nicolas Cedilnik [Correspondant]‌‌, Jairo Rodríguez, Buntheng Ly, Mihaela‌ Pop, Maxime Sermesant‌.

We improved "cardiaclbm"‌‌ (see Fig. 30), a python package for‌ monodomain, organ-scale fast (GPU-powered)‌ simulations of cardiac electrophysiology‌‌ (EP) using the lattice-Boltzmann method (LBM).
We used‌ it in ongoing works,‌ where image-based arythmia simulations‌‌ in infarcted pigs were compared to real EP‌ recordings.
We showed that‌ the LBM is fast‌‌ enough to allow iterative parameter tuning against non-invasive‌ EP data, in a‌ multimodal (imaging and electrocardiographic)‌‌ model personalization framework.

Figure 30: Screenshot of‌ a dynamic visualization tool‌ developed on top of‌‌ the cardiaclbm python package,‌ showing re-entrant depolarization wave (top left), analysis of‌ EP properties (top right, middle), and a simulated‌ ECG (bottom).

8.4.2 Differentiable Electromechanical Modeling for Patient-Specific‌ Cardiac Biomechanics

This work was funded by the‌ France 2030 program (MediTwin project).

Keywords: Cardiac biomechanics;‌ Electromechanics; Personalization; Numerical simulation; Machine learning.

Participants:‌ Gaëtan Desrues [Correspondant], Maxime Sermesant.

Development‌ of a 3D cardiac electromechanical model using the‌ SOFA simulation framework. Application to patient-specific four-chamber cardiac‌ biomechanics (Fig. 31).
Design of a modular‌ differentiable simulation framework for physics-based modeling in JAX.‌ FEM-based electromechanical formulations implemented with fully differentiable operators.‌
Integration of energy-based Hamiltonian formulations and AI-driven methods‌ for physics-informed modeling of cardiac electromechanics.
Cardiac electromechanical‌ simulation for the assessment and optimization of mitral‌ valve implant design, including annular downsizing strategies.

Figure‌ 31: Finite element representations of patient-specific four-chamber‌ cardiac anatomy and associated biomechanical submodels used for‌ electromechanical simulations.

8.4.3 Prediction of stroke based on‌ shape and simulation of the left atrium

This‌ work was funded by the ANR under the‌ France 2030 project RHU Talent (ANR-23-RHUS-0015) (ANR, IHU‌ Lyric, CHU Bordeaux, Université de Bordeaux, Inria, CHU‌ Dijon Bourgogne, inHEART, Cardiologs et Incepto).

Keywords: Stroke;‌ Shape analysis; Bloodflow simulation; Atrial fibrillation.

Participants:‌ Nicolas Drettakis [Correspondant], Maxime Sermesant, Hubert‌ Cochet.

This year was first focused on‌ adapting and upgrading the automatic segmentation, labeling and‌ feature extraction pipeline for the left atrium that‌ was developed first in J. Harrison's PhD to‌ use it on a larger database. The pipeline‌ was also adapted to be used in the‌ BeatAF project to automatically measure the diameters of‌ the pulmonary veins before and after left atrium‌ appendage ablation. We also worked on preparing the‌ use of bloodflow simulation of the left atrium‌ to extract corresponding features (see Fig. 32).‌

Figure 32: Finite element representations of patient-specific‌ four-chamber cardiac anatomy and associated biomechanical submodels used‌ for electromechanical simulations.

8.4.4 Myocardial‌ Stiffness Quantification using Ultrasound Shear Wave Elastography and‌ Reduced Modeling

This work was funded by the ANR (PEPR Digital Health‌ ChroniCardio, 3IA Côte d'Azur,‌ IA Cluster, IHU Liryc),‌‌ France 2030, and the European Union (MediTwin project).‌

Keywords: Cardiac mechanics; Reduced-order‌ models; Ultrafast ultrasound imaging;‌‌ Myocardial stiffness; Patient-specific analyses.

Participants: Camilla Ferrario‌ [Correspondant], Jairo Rodríguez‌ Padilla, Maelys Venet‌‌, Olivier Villemain, Maxime Sermesant.

We‌ developed a pipeline to‌ estimate active myocardial contractility‌‌ from shear wave elastography (SWE). A reduced spherical‌ electromechanical model predicts stiffness‌ using a rheological circuit.‌‌ Active parameters, comprising reference stiffness and cross bridge‌ cycling rates, are calibrated‌ via the CMA-ES algorithm‌‌ to minimize the mismatch between modeled and clinical‌ stiffness trajectories (Fig. 33‌) 35.

Figure‌‌ 33: Active parameter estimation. (i) Clinical input:‌ In vivo shear wave‌ elastography (SWE) yields target‌‌ myocardial stiffness $k_{c}^{SWE} (t)‌$ . The bottom panel‌ shows cohort trajectories (gray)‌‌ and their ensemble average (dashed). (ii) Reduced electromechanical‌ model: The left ventricle‌ (LV) is modeled as‌‌ a thick-walled sphere. A rheological circuit with active‌ ( $k_{c},‌ τ_{c}$ ) and‌‌ passive ( $W_{e}, W_{v},‌ μ$ ) components predicts‌ stiffness $k_{c}^{model‌‌} (t; θ)$ from kinematics. (iii)‌ Active parameter optimization: Parameters‌ $θ$ (max stiffness ${k‌‌}_{0}$ , rates ${k}_{ATP}, k_{SR‌}$ ) are identified by‌ minimizing cost $J (‌‌ θ)$ via CMA-ES. The blue curve shows‌ the optimal fit against‌ clinical data (dashed).

8.4.5 Multimodal Personalization of Cardiac Electrophysiology Models‌ combining 12-lead ECG and‌ Computed Tomography

This work‌‌ has been supported by the French government through‌ the National Research Agency‌ (ANR) 3IA Côte d'Azur,‌‌ IA Cluster project and IHU LIRYC (ANR-19-3IA-0002, ANR-23-IACL-0001‌ and ANR-10-IAHU-04) and through‌ the France 2030 and‌‌ the European Union (Next Generation EU) MediTwin project.‌ The authors are grateful‌ to the OPAL infrastructure‌‌ from Université Côte d'Azur for providing resources and‌ support.

Keywords: Computational modeling;‌ Myocardial infarction; ECG; Computed‌‌ tomography; Model personalization.

Participants: Buntheng Ly,‌ Nicolas Cedilnik, Mihaela‌ Pop, Josselin Duchateau‌‌, Frédéric Sacher, Pierre Jaïs, Hubert‌ Cochet, Maxime Sermesant‌.

We propose an‌‌ automated framework for the parameterization of cardiac electrophysiological‌ model by combining information‌ derived from CT scans‌‌ and 12-lead ECGs, with the aim of fine-tuning‌ model parameters based on‌ electrical features extracted from‌‌ recorded ECGs at sinus rhythm 41 (Fig. 34‌). The optimized parameters‌ induced virtual VT with‌‌ cycle length and pattern closer to those recorded,‌ as compared to the‌ baseline parameters.

Figure 34‌‌: VT simulation pipeline‌ with multi-modal personalization strategy.

8.4.6 Learning Cardiac Electrophysiology with Graph‌ Neural Networks for Fast Data-driven Personalized Predictions

This‌ work has received funding from the French government‌ through the National Research Agency DeepNum project (ANR-21-CE23-0017).‌

Keywords: Graph neural network; Data-driven modeling; Personalized cardiac‌ electrophysiology.

Participants: Maëlis Morier [Correspondant], Patrick‌ Gallinari, Maxime Sermesant.

We present AGATA‌51, an Autoregressive Graph Attention network for‌ fast and accurate cardiac action potential simulation. Trained‌ on FEM data (A) (see Fig. 35),‌ AGATA predicts seconds of propagation from only 25‌ ms of input. Its architecture (B) (shown in‌ Fig. 35) generalizes from simple to realistic‌ cardiac geometries, captures healthy and pathological dynamics, achieves‌ a $3 \times {10}^{- 5}$ mean absolute‌ error, and is up to 19 times faster‌ than FEM, enabling fast, accurate, patient-specific digital twin‌ modeling.

Figure 35: A. Framework for Action‌ Potential simulation, starting from an ellipsoid mesh, with‌ Firedrake FEM computations and Pytorch Geometric graph conversion.‌ $t$ is a step of the simulation. B.‌ The architecture of AGATA, combining an autoregressive approach‌ with an attention mechanism. k1, k2 are the‌ number of nodes in the layers.

8.4.7 Uncertainty-Informed Multimodal Infarct Age‌ Prediction from Imaging and Clinical Data

This work‌ has been supported by the French government through‌ France 2030 (MediTwin) and the European Union (Next‌ Generation EU), the National Research Agency (ANR) Investments‌ in the Future with 3IA Côte d'Azur and‌ IA Cluster (ANR-19-3IA-0002 and ANR-23-IACL-0001), LIRYC (ANR-10-IAHU-04) and‌ ChroniCardio - ANR-22-PESN-0015. The authors are grateful to‌ the OPAL infrastructure from Université Côte d'Azur for‌ providing resources and support.

Keywords: Multimodal deep learning;‌ Infarct age regression; Medical imaging; Intramyocardial fat and calcification; Myocardial thickness.‌

Participants: Evariste Njomgue Fotso‌ [Correspondant], Marta Nuñez-Garcia‌‌, Buntheng Ly, Hubert Cochet, Maxime‌ Sermesant.

Accurately estimating‌ the age of a‌‌ myocardial infarction (MI) is critical for prognostic assessment‌ and for guiding post-MI‌ clinical management, particularly in‌‌ applications such as arrhythmia risk stratification. In this‌ study, we address this‌ challenge by proposing a‌‌ novel multimodal regression framework for infarct age prediction‌ 36.

The proposed‌ framework integrates quantitative descriptors‌‌ of intramyocardial fat, calcification, and myocardial thickness extracted‌ from mid-wall mesh nodes,‌ followed by a quantile-range‌‌ decision fusion strategy to combine modality-specific predictions (Fig.‌ 36). Accurate infarct‌ age estimation is especially‌‌ important in clinical scenarios such as silent MI—a‌ well-established risk factor for‌ sudden cardiac death—where delayed‌‌ diagnosis complicates timely therapeutic intervention.

Moreover, in situations‌ where infarct age information‌ is incomplete or uncertain,‌‌ robust regression models can provide reliable age estimates,‌ thereby supporting more informed‌ clinical decision-making and personalized‌‌ treatment planning.

Figure 36: (Top): Data‌ processing pipeline. (Down):‌ Quantiles Range Decision Fusion:‌‌ general setting. For each modality, a predictor regressor‌ and a conditional quantile‌ regressor are trained separately.‌‌ Late fusion at the decision level is then‌ applied, selecting the best‌ infarct age prediction based‌‌ on the smallest quantile range.

8.4.8 In silico‌ Assessment of Arrhythmia Inducibility‌ Dependence on Stimulus Location‌‌ using Calibrated MR-based Infarcted Heart Models

This work‌ has received funding from‌ the European Union Horizon‌‌ 2020 Research and Innovation Program SimCardioTest (101016496), from‌ the French government through‌ the National Research Agency‌‌ (ANR) projects PEPR Digital Health ChroniCardio (22-PESN-0015), 3IA‌ Côte d'Azur and IA‌ Cluster (ANR-19-3IA-0002 and ANR-23-IACL-0001).‌‌ The authors are also grateful to the OPAL‌ infrastructure from Université Côte‌ d'Azur for providing computational‌‌ resources and associated support.

Keywords: Electrophysiology; Modeling; Simulations;‌ Finite element; Personalization.‌

Participants: Jairo Rodríguez Padilla‌‌ [Correspondant], Rafael Silva, Buntheng Ly,‌ Graham Wright, Mihaela‌ Pop, Maxime Sermesant‌‌.

The aim of this work (42‌) was to implement‌ a robust computational pipeline‌‌ (Figure 37) to build and calibrate digital‌ twins able to accurately‌ predict VT inducibility using‌‌ high-resolution preclinical MRI-EP datasets. Novel aspects of the‌ work include:

A data-driven‌ macroscopic model personalization method‌‌ using intracardiac electrograms (iECGs) from very dense contact‌ catheter-based electro-anatomical maps.
Integration‌ of an atlas of‌‌ fiber directions (instead of rule-based fibers) into the‌ 3D heart models.
A‌ fully Phyton-coded FEM implementation‌‌ for the numerical solver in FEniCSx.

Figure 37‌: Workflow of 3D‌ MR-based biventricular model construction‌‌ and the in silico testing of stimulation protocols‌ from different pacing sites‌ to study VT inducibility.‌‌ GZ: Grey zone.

8.4.9 Miniaturizing Automated External Defibrillation‌ with Frugal AI

This work has been supported‌ by the French government, through the National Research‌ Agency (ANR) 3IA Côte d'Azur and IA Cluster‌ project (ANR-19-3IA-0002 and ANR-23-IACL-0001).

Keywords: Frugal AI; Shockable‌ rhythm detection; Embedded systems; Deep learning; ECG analysis‌.

Participants: Rafael Silva [Correspondant], Caroline Stehlé‌, Maxime Sermesant.

In collaboration with Inn'Pulse‌ and 3IA Côte d'Azur, we designed a frugal‌ deep learning shock advisory algorithm for ultra-portable defibrillators,‌ combining neural architecture search, 8-bit quantization, external dataset‌ validation, and STM32 deployment tests to meet international‌ standards while minimizing computation and energy consumption 43‌, 62. The methodology is summarized in‌ Figure 38.

Figure 38: Overview of‌ the proposed methodology: from ECG data processing to‌ model optimization and its deployment on a microcontroller.‌

8.5 Multi-centric data and Federated Learning‌

8.5.1 Development of a 18FDG-PET normative uptake atlas‌ and its clinical application for abnormal metabolic activity‌ detection

This work was funded by the project‌ FEDERATED-PET.

Keywords: PET/CT scans; Machine learning; Lung cancer‌.

Participants: Lucie Chambon [Correspondant], Francesco Cremonesi‌, Marco Lorenzi, Olivier Humbert.

This‌ work proposes a normative modeling framework for whole-body‌ 18F-FDG PET scans (Fig. 39):

We constructed‌ a multi-center PET normative uptake atlas from healthy‌ control cohorts.
We introduced an interpretable measure to‌ quantify organ-level metabolic deviations.
We validated the method‌ on an external lung cancer cohort, showing robust‌ detection of pathological hyper-metabolism across organs, and improved‌ generalization compared with classical metrics based on standardized‌ uptake value (SUV).

Figure 39: Illustration of‌ the normative atlas development and usage workflow.

8.5.2 Real-world Deployment‌ of Federated Learning in‌‌ Biomedical Research Consortia with Fed-BioMed

This work was‌ funded by the European‌ Union EUCAIM project under‌‌ Grant Agreement 101100633.

Keywords: Biomedical data; Federated learning;‌ Machine learning.

Participants:‌ Francesco Cremonesi [Correspondant],‌‌ Sergen Cansiz, Lucie Chambon, Ali Tolga‌ Dincer, Yannick Bouillard‌, Jhonatan Leonardo Torres‌‌ Sanchez, Marc Vesin, Marco Lorenzi.‌

Fed-BioMed is an open-source‌ initiative aimed at enabling‌‌ real-world deployment of federated learning in biomedical research.‌
In 2025, Fed-BioMed has‌ been actively developed and‌‌ improved, in particular in terms of its deployment‌ process, integration with hospital‌ and research infrastructures, and‌‌ interoperability with medical data standards. The software is‌ now in release v6.2.‌
Fed-BioMed supports state-of-the-art low-overhead‌‌ secure aggregation protocols which have been demonstrated on‌ a real-world deployment of‌ three French hospitals from‌‌ the UniCancer consortium (Centre Antoine Lacassagne, Centre Henri‌ Becquerel, Institut Curie), see‌ Fig. 40. The‌‌ technical architecture has been developed following a co-design‌ approach involving all stakeholders‌ including data holders, data‌‌ scientists, and software developers 53. A preliminary‌ integration of Fed-BioMed with‌ the public platform for‌‌ the European Cancer Imaging Infrastructure (EUCAIM) has also‌ been achieved in 2025‌ 49.

Figure 40‌‌: a) Pixel intensity distribution grouped by clinical‌ site. b) Breakdown of‌ FL experiment wallclock runtime.‌‌ c) Distribution of Dice scores for centralized (CL)‌ and federated (FL) models,‌ combining all cross-validation folds.‌‌ d) Clockwise from top left: an example raw‌ image, the ground truth‌ segmentation, the FL model‌‌ prediction and the CL model prediction.

8.5.3 Federated domain adaptation for brain‌ vessel segmentation

Keywords: Federated‌ learning; Domain adaptation; Vessel‌‌ segmentation; Angiography.

Participants: Tuan Anh Nguyen [Correspondant]‌, Francesco Cremonesi,‌ Lucie Chambon, Marco‌‌ Lorenzi.

The application of Domain Adaptation techniques‌ in the context of‌ federated learning allows to‌‌ mitigate issues due to heterogeneous data and missing‌ lagels in a privacy-preserving‌ manner.
In the context‌‌ of this project, the scenario of Angiography-to-Angiography Translation‌ was adressed in a‌ real-world deployment involving three‌‌ institutions: INRIA, EURECOM, and GIN (Grenoble Institute of‌ Neuroscience), see Fig. 41‌.
Domain Adaptation techniques‌‌ were used to transfer annotations from publicly-available datasets‌ to real-world clinical datasets‌ where only a handful‌‌ of samples were annotated by experts.
The deployment,‌ supported by the Fed-BioMed‌ software, showed that the‌‌ federated approach produced results‌ that were closely aligned with those of the‌ centralized domain adaptation, with performance differences not exceeding‌ 4% for most metrics.

Figure 41: Using‌ federated learning, 3 clients collaboratively train a multimodal‌ data factory F (in blue). Afterwards, source clients‌ can contribute by training locally segmentation branches (in‌ orange), while target clients can asynchronously acquire the‌ information required to segment their data via domain‌ adaptation (in pink) (figure taken from Galati et‌ al, Federated Multi-centric Image Segmentation with Uneven Label‌ Distribution). On the right, an example of domain‌ adaptation results.

8.5.4 Knowledge‌ Guided Medical Report Generation for Pathology Specific Findings‌

This work was funded by PEPR Santé Numerique.‌

Keywords: Radiology report generation; Vision language models.‌

Participants: Giuseppe Orlando [Correspondant], Olivier Humbert,‌ Marco Lorenzi.

Our research is focused on‌ optimizing Vision Language Models for medical reporting. In‌ collaboration with the Centre Antoine Lacassagne, we demonstrated‌ that prompting generalist models with targeted anatomical questions‌ and merging the responses significantly improves report quality‌ (Fig. 42). Building on this, we introduce‌ a segmentation guided 3D chest CT framework using‌ quantitative features to ground text in explicit evidence.‌

Figure 42: The left panels illustrate our‌ framework: prompting generalist models (e.g., CT-CHAT) with targeted‌ questions to extract specific clinical entities from CT‌ scans for final report assembly. The right panel‌ shows F1 score comparisons, demonstrating the improvement of‌ this knowledge guided approach.

8.5.5 Fed-ComBat: A Generalized Federated Framework‌ for Batch Effect Harmonization in Collaborative Studies

This‌ project received financial support by the PEPR Santé‌ Numerique.

Keywords: Federated learning; HarmonizationComBat; Medical imaging;‌ Batch effects.

Participants: Ghiles Reguig [Correspondant],‌ Santiago Silva, Neil Oxtoby, Andre Altmann‌, Marco Lorenzi.

The aim of this‌ project is to develop robust analysis tools for‌ neuroimaging data, that can find application in multi-centric studies such as the‌ French CATI. We developed‌ a federated flexible framework‌‌ for multi-centric data harmonization based on ComBat. The‌ described implementation relies on‌ a stochastic gradient descent‌‌ optimization model which allows to leverage a large‌ family of linear and‌ nonlinear models. The method‌‌ was tested on the harmonization of a set‌ of public datasets and‌ was compared to various‌‌ methods from the literature (using linear or nonlinear‌ modelization of the biological‌ effects to keep). The‌‌ results depicted in Figure 43 show that our‌ method yields similar results‌ to the state-of-the-art while‌‌ allowing both a federated learning scheme and nonlinear‌ modelization of the biological‌ effects. The paper describing‌‌ our contribution is currently in submission and an‌ implementation in the Fed-BioMed‌ library is in development‌‌ for real-world usage.

Figure 43: Comparison of‌ age trajectories of the‌ right hippocampus. The columns‌‌ represent conditions under which the data is shown.‌ We observe that harmonization‌ using linear models (SGD‌‌ Linear, NeuroComBat and Fed-ComBat Linear) show atrophy patterns‌ that are not consistent‌ with the literature while‌‌ the nonlinear ones (SGD MLP, ComBatGAM and Fed-ComBat‌ MLP) better match the‌ different rates of atrophy‌‌ across lifespan regarding controls (CN).

9 Bilateral‌‌ contracts and grants with industry

9.1 Bilateral contracts‌ with industry

9.1.1 Spin-off‌ company inHEART

Participants: Maxime‌‌ Sermesant.

inHEART is a spin-off of the‌ Epione team and IHU‌ Liryc funded in 2017.‌‌ inHEART provides a service to generate detailed anatomical‌ and structural meshes from‌ medical images, that can‌‌ be used during ablation interventions. inHEART received 2‌ awards, one from Aquitaine‌ region and one i-LAB‌‌ from the BPI. It raised 3.2 million euros‌ in 2020. It currently‌ employs 27 people. It‌‌ is FDA and CE certified.

9.1.2 Start-up Inn'Pulse‌

Participants: Rafael Silva,‌ Caroline Stehle, Maxime‌‌ Sermesant.

Inn'Pulse is a start-up developing an‌ ultra-portable automatic cardiac defibrillator.‌ We are designing AI‌‌ algorithms for better signal processing in collaboration with‌ the 3IA TechPool.

9.1.3‌ Koelis

Participants: Nicholas Ayache‌‌, Hervé Delingette, Benjamin Billot.

The‌ company Koelis participates in‌ the thesis work of‌‌ Manasi Kattel on the multimodal registration of MR‌ and ultrasound images. The‌ objective is to improve‌‌ the accuracy of the targeted biopsies inside the‌ prostate.

10 Partnerships and‌ cooperations

10.1 International initiatives‌‌

10.1.1 International collaborations

Computer Science and Artificial Intelligence‌ Laboratory, Massachusetts Institute of‌ Technology. Benjamin Billot is‌‌ currently collaborating with Prof. Polina Golland, through the‌ common supervision of Ramya‌ Muthukrishnan, on developing equivariant‌‌ networks for motion tracking in fetal MRI time‌ series.
Harvard Medical School‌ and Massachusetts General Hospital.‌‌ Benjamin Billot is a collaborator of Dr. Juan‌ Eugenio Iglesias to develop‌ new tools for neuro-imaging‌‌ based on domain randomization‌ strategies.
Hawkes Institute, University College London (UCL), London,‌ UK. Benjamin Billot collaborates with Prof. Daniel Alexander‌ and Dr. Henry Tregidgo on developing new domain‌ randomization and simulation strategies for domain-agnostic segmentation of‌ brain MRI scans across a wide range of‌ populations.
University College London (UCL), London, UK. Irene‌ Balelli collaborates with the Department of Statistical Science‌ and the UCL Causality group (Dr. Anna Calissano‌ and Dr. Karla Diaz Ordaz). The collaboration consists‌ in exploring new methdologies for causal learning based‌ on graph theories for high dimensional data.
King's‌ College London (KCL) and Guy's Hospital, London, UK.‌ Irene Balelli is currently collaborating with Pr. Mohammad‌ Mousavi (King's Quantum group) and Dr. Rocio Martinez-Nuñez‌ on causal learning for respiratory epidemiology and quantum‌ causal discovery.
McGill University, Canada. Marco Lorenzi collaborates‌ with Prof. J.-B. Poline for the development of‌ federated learning architectures and standards for reproducible collaborative‌ analysis in multicentric neuroimaging studies.
University College London‌ (UCL), London, UK. Marco Lorenzi is a collaborator‌ of the COMputational Biology in Imaging and geNEtics‌ (COMBINE) group within the Centre for Medical Image‌ Computing (CMIC) of UCL. His collaboration is on‌ the topic of spatio-temporal analysis of medical images‌ and imaging-genetics, with special focus on brain imaging‌ analysis and biomarker development.
Laboratory of Neuroimaging of‌ Aging (LANVIE), Faculty of Medicine, Geneva University Hospitals.‌ Marco Lorenzi collaborates with the LANVIE laboratory led‌ by Prof. Giovanni B. Frisoni. The collaboration consists‌ in developing and translating novel approaches for disease‌ progression modeling in neurodegenerative disorders, such as Alzheimer's‌ disease.
Illinois Institute of Technology (IIT, IL, USA).‌ Marco Lorenzi is currently a collaborator of IIT‌ for the investigation of the progression of Parkison's‌ disease using disease progression modeling.
Laboratory of Physics‌ of Fluids, University of Twente, Netherlands. Hervé Delingette‌ is collaborating with Assistant Professor Guillaume Lajoinie, on‌ the topics of Deep Learning for ultrasound imaging‌ in the framework of the BoostUrCareer Cofund program‌ and the thesis of Hari Sreedhar.

10.1.2 Participation‌ in other International Programs

CausalGene

Funding:
EPSRC-funded project‌ via the CHAI (Causality in Healthcare AI) Hub.‌
Title:
Glucocorticoid effects on human airway cells through‌ population of causal graphs.
Duration:
From November 1,‌ 2025 to June 30, 2026
Partners:
- University College‌ London (UCL), UK
- Inria, France
- King's College London‌ (KCL), UK
- Guy's Hospital, UK
- Asthma+Lung UK,‌ UK
Inria contact:
Irene Balelli
Coordinator:
UCL
Summary:‌
1% of the population worldwide receives long-term oral‌ type of steroids called glucocorticoids. 40% of those‌ treatments are taken by patients with respiratory diseases‌ including asthma, which affects around 300 million people‌ worldwide. However, how glucocorticoids work at the molecular‌ level is not fully understood, causing many undesired‌ effects. CausalGene aims to unveil how the gene‌ expression of lungs cells changes upon administration of‌ glucocorticoids. To understand the causal effects of glucocorticoids‌ on cells reprogramming, CausalGene combines methods from causal‌ discovery and graph theory to reveal novel molecular‌ mechanisms governing glucocorticoid cellular activity and to start designing better anti-inflammatory drugs.‌

10.2 International research visitors‌

10.2.1 Visits of international‌‌ scientists

Jean-Baptiste Poline

Status:
Professor
Institution of origin:‌
McGill University
Dates:
06/10/2025‌ - 17/10/02026
Context of‌‌ the visit:
Prof. J.-B. Poline visited Epione with‌ a grant IVADO-Inria for‌ building a collaboration with‌‌ Marco Lorenzi on reproducible collaborative learning methods and‌ software in neuroimaging studies.‌
Mobility program/type of mobility:‌‌
Research stay

Hervé Lombaert

Status:
Professor
Institution of‌ origin:
Polytechnique Montréal
Dates:‌
December 2025
Context of‌‌ the visit:
Hervé Lombaert visited Epione during 3‌ weeks in December to‌ discuss potential collaborations, especially‌‌ about X-Ray image processing.
Mobility program/type of mobility:‌
Research stay

Alessandra Corda‌

Status:
PhD student
Institution‌‌ of origin:
Politecnico di Milano
Dates:
01/11/2025 -‌ 01/02/02026
Context of the‌ visit:
adaptation of electrophysiological‌‌ models to experimental data
Mobility program/type of mobility:‌
Research stay

10.2.2 Visits‌ to international teams

Benjamin‌‌ Billot received fundings from Inria-London to visit Prof.‌ Daniel Alexander and Dr.‌ Henry Tregidgo for a‌‌ 1-week visit at the Hawkes institute (UCL) in‌ order to discuss a‌ future project for contrast-agnostic‌‌ brain MRI segmentation.
Marco Lorenzi was invited to‌ join the French delegation‌ to attend the “France-Japan‌‌ Bilateral Seminar on Health Data,” co-hosted by the‌ Science and Technology Department‌ of the French Embassy‌‌ in Japan, PEPR Digital Health (France 2030 Investment‌ Plan), and the Keio‌ University School of Medicine.‌‌
Marco Lorenzi was invited to the Computer Science‌ Department of Ruhr Universität‌ Bochum (RUB) within the‌‌ framework of the Franco-German project TRAIN.
Arnaud Lang‌ received funding from the‌ Inria-London program for a‌‌ 1 week visit to Pr. Calissano at University‌ College London in the‌ framework of his M2‌‌ internship under the supervision of Irene Balelli .‌

10.3 European initiatives

10.3.1‌ Horizon Europe

EUCAIM

EUCAIM‌‌ project on cordis.europa.eu

Title:
EUropean Federation for CAncer‌ Images
Duration:
From January‌ 1, 2023 to December‌‌ 31, 2026
Partners:
- BBMRI-ERIC, Austria
- European Institute for‌ Biomedical Imaging Research, Austria‌
- Medical University of Innsbruck,‌‌ Austria
- Charité, Germany
- ELIXIR, Germany
- German Cancer Research‌ Center, Germany
- Inria, France‌
- Technical University of Munich,‌‌ Germany
- Aristotle University of Thessaloniki, Greece
- Hellenic Cancer‌ Society, Greece
- National and‌ Kapodistrian University of Athens,‌‌ Greece
- Gemelli University Hospital, Italy
- Italian National Research‌ Council, Italy
- University of‌ Pisa, Italy
- Andalusian Health‌‌ Service, Spain
- Instituto de Salud Carlos III, Spain‌
- University of Barcelona, Spain‌
- Karolinska Institute, Sweden
- Linköping‌‌ University, Sweden
- Umeå University, Sweden
- Pohjois-Savon Hyvinvointialue, Finland‌
- Oslo University Hospital, Norway‌
Inria contact:
Marco Lorenzi‌‌
Coordinator:
Marco Lorenzi
Summary:
The EUCAIM project is‌ a cornerstone of the‌ European Cancer Imaging Initiative‌‌ under Europe’s Beating Cancer Plan. This initiative is‌ a significant contributor to‌ the European Health Data‌‌ Space and aims to establish a pan-European digital‌ and federated infrastructure of‌ FAIR (Findable, Accessible, Interoperable,‌‌ and Reusable) cancer images.

DTRIP4H

DTRIP4H project on‌ cordis.europa.eu

Title:
Enabling Decentralized‌ Digital Twin Era in‌‌ existing Research Infrastructures for Predictive, Preventive, Personalized, and‌ Participatory Health
Duration:
From‌ January 1, 2025 to‌‌ December 31, 2028
Partners:‌
- UAB Teraglobus, Lithuania
- Inria, France
- Instituto Pedro Nunes‌ associacao para a inovacao e desenvolvimento em ciencia‌ e technologia, Portugal
- Metropolia Ammattikorkeakoulu Oy, Finland
- Ludwig-Maximilians-Universitaet‌ Muenchen, Germany
- Helsingin Yliopisto, Finland
- Digitaltwin Technology GMBH,‌ Germany
- Masarykova Univerzita, Czechia
- Centre for research and‌ technology Hellas Certh, Greece
- Oulun Yliopisto, Finland
- Artificial‌ intelligence expert SRL, Romania
- Klinikum der Ludwig-Maximilians-Universitaet Muenchen,‌ Germany
- Nec Italia SPA, Italy
- Chino SRL (Chino.io),‌ Italy
- Demcon Data Driven Solutions B.V., Netherlands
- Europrean‌ Health Management Association, Belgium
- Linac-PET Scan Opco ltd,‌ Cyprus
- Nec Labortories Europe GMBH, Germany
- Lapland University‌ of Applied Sciences, Finland
- Alma Mater Studiorum -‌ Universita di Bologna, Italy
- Near Real Oy, Finland‌
- Demcon Sync Biosystems, Netherlands
- Protobios, Estonia
- Universidad del‌ pais Vasco/ Euskal Herriko Unibertsitatae, Spain
- Tallinn University‌ of Technology, Estonia
- Oulu University of Applied Sciecnces,‌ Finland
Inria contact:
Marco Lorenzi
Coordinator:
Marco Lorenzi‌
Summary:
In the face of a rapidly advancing‌ digital healthcare terrain, the DTRIP4H project emerges as‌ a momentous effort to revolutionize predictive, preventive, personalized,‌ and participatory health paradigms within the EU. Amid‌ significant incidence of chronic conditions and cancer, there‌ is a pressing need for a proactive shift‌ in health strategies. Yet, the full potential of‌ European research infrastructures (RIs) is curtailed by investment‌ deficits, fragmentation, and the intricacies of data management.‌ Digital Twin (DT) technology introduces a new age‌ of precision by enabling sophisticated simulations and analyses‌ of intricate biological processes. In DTRIP4H, we start‌ a new initiative in Europe “decentralized health digital‌ twin ecosystem consisting of RIs”. Using DTs, we‌ aim to resolve critical challenges around data harmonization,‌ equitable access, and stringent privacy safeguards. Incorporating technologies‌ such as federated learning, Generative AI, and Virtual‌ Reality (VR), the project aspires to create a‌ decentralized digital twin environment (DDTE). This will empower‌ both internal and external RI users, such as‌ researchers, innovators, and SMEs, to craft DT applications‌ that address specific scientific challenges, utilizing a blend‌ of real-world and synthetic data in compliance with‌ regulatory frameworks, i.e. GDPR. We will develop 7‌ innovative proof of concept thematic health-related Use cases‌ fulfilling the needs of scientists, SMEs, and industrial‌ end users, particularly in health topics related to‌ cancer treatment, drug development, human environmental exposome, precision‌ treatment for schizophrenia and personalized medicine through Artificial‌ Intelligence (AI), AR/VR empowered DTs utilizing DDTE, while‌ adhering to FAIR data principles. DTRIP4H adopts a‌ human-centric methodology to elevate research efficacy, narrow the‌ skills gap, and align with the objectives of‌ the European Research Area (ERA) and the Sustainable‌ Development Goals (SDGs) by 2030.

10.3.2 H2020 projects‌

SimCardioTest

SimCardioTest project on cordis.europa.eu

Title:
Simulation of‌ Cardiac Devices and Drugs for in-silico Testing and‌ Certification
Duration:
From January 1, 2021 to June‌ 30, 2025
Partners:
- Universidad Pompeu Fabra, Spain
- Inria,‌ France
- Virtual Physiological human Institute for integrative biomedical‌ research, Belgium
- Sorin CRM SAS, France
- ExactCure, France‌
- Simula research laboratory, Norway
- IniSilicoTrials Technologies, Netherlands
- Université‌ De Bordeaux, France
- Boston Scientific, United States
- Universitat Politecnica de Valencia, Spain‌
- IniSilicoTrials Technologies, Italy
Inria‌ contact:
Maxime Sermesant
Coordinator:‌‌
Maxime Sermesant
Summary:

Despite massive investment in healthcare,‌ huge research and development‌ cost increase and regulatory‌‌ pathway complexity hamper tremendously commercialization of new devices‌ and medicines, putting patient‌ populations at risk of‌‌ not receiving adequate therapy. At the same time,‌ outside healthcare, computer modeling‌ and simulation (CM&S) is‌‌ precisely recognized to increase speed and agility while‌ reducing costs of development.‌ CM&S can create scientific‌‌ evidence based on controlled investigations including variability, uncertainty‌ quantification, and satisfying demands‌ for safety, efficacy and‌‌ improved access.

Cardiac modeling has dramatically gained maturity‌ over the last decades,‌ with personalization to clinical‌‌ data enabling validation. We selected a number of‌ cardiac devices and medicines‌ where CM&S is mature‌‌ enough and that represent the most common cardiac‌ pathologies, to demonstrate a‌ standardized and rigorous approach‌‌ for in-silico clinical trials.

SimCardioTest will bring a‌ disruptive innovation by creating‌ an integrated and secure‌‌ platform standardizing and bridging model simulations, in-silico trials,‌ and certification support. This‌ environment will go beyond‌‌ the state-of-the-art in computational multi-physics and multi-scale personalized‌ cardiac models. Diseased conditions‌ and gender/age differences will‌‌ be considered to overcome clinical trials limitations such‌ as under-representation of groups‌ (e.g. women, children, low‌‌ socio-economic status). Advanced big data, visual analytics and‌ artificial intelligence tools will‌ extract the most relevant‌‌ information.

It is critical that Europe demonstrates its‌ capacity to leverage in-silico‌ technology in order to‌‌ be competitive in healthcare innovation. SimCardioTest exploitation aims‌ at delivering a major‌ economic impact on the‌‌ European pharmaceutical and cardiac devices industry. It will‌ accelerate development, certification and‌ commercialization, and will produce‌‌ a strong societal impact contributing to personalized healthcare.‌

inEurHeart

inEurHeart website

Title:‌
inEurHeart: AI, Digital Twin‌‌ & Clinical Trial for a Disruption in Catheter‌ Ablation
Duration:
2022-2025
Partners:‌
- Inria, France
- Rotterdam University,‌‌ Netherlands
- Inserm, France
- CHU Bordeaux, France
- inHEART, France‌
- Université de Bordeaux, France‌
Inria contact:
Maxime Sermesant‌‌
Coordinator:
Inria
Summary:
inEurHeart is an innovation project‌ in Artificial Intelligence, Digital‌ Twin & a Clinical‌‌ Trial for a Disruption in Catheter Ablation for‌ Ventricular Tachycardia, making ablation‌ therapy accessible to most‌‌ patients. This project is a collaborative project between‌ 5 organizations in France‌ and Netherlands funded by‌‌ EIT Health - the European Institute of Innovation‌ and Technology, co-funded by‌ the European Union. This‌‌ project will exemplify how the academic-industrial relationships can‌ be fostered and can‌ lead to drastic changes‌‌ in clinical practice. EIT Health provides a unique‌ opportunity to transfer Artificial‌ Intelligence tools to enable‌‌ the scale-up phase, and to validate the technology‌ through a randomized clinical‌ trial.

10.4 National initiatives‌‌

10.4.1 PEPR Digital Health ChroniCardio

Participants: Jairo Rodriguez‌, Mihaela Pop,‌ Maxime Sermesant.

Duration:‌‌
2023 - 2027
Partners:
- Inria
- Hospices Civils de‌ Lyon
- Assistance Publique -‌ Hôpitaux de Marseille
- Creatis,‌‌ Lyon
Inria contact:
Maxime Sermesant
Coordinator:
Inria
Summary:‌
ChroniCardio is a new‌ 4-year multi-institution project funded‌‌ by the French Research‌ Priority Programme on Digital Health to accelerate the‌ integration of multi-scale data (clinical, imaging, genetic, ECG,‌ etc.) and the development of advanced modeling tools‌ to predict the long-term evolution of non-ischemic dilated‌ and hypertrophic cardiomyopathies. This includes the risk of‌ arrhythmia, heart failure and sudden cardiac death. Our‌ consortium brings together research scientists and engineers from‌ Inria (Sophia Antipolis, Lyon, Rennes, Bordeaux) and INSA‌ / Lyon /University / CNRS (Lyon), and clinicians‌ from Lyon and Marseille University Hospitals. The project‌ is coordinated by Maxime Sermesant , from Inria‌ Epione team.

10.4.2 PEPR Digital Health Rewind

Participants:‌ Marco Lorenzi.

Duration:
2023 - 2027
Partners:‌
- Inria
- CNRS
- INSERM
- Université Grenoble Alpes
Inria contact:‌
Stephanie Allassonniére
Coordinator:
Inria
Summary:
The project Rewind‌ will focus on the development of new mathematical‌ and statistical approaches for the analysis of multimodal‌ multiscale longitudinal data. These models will be designed,‌ implemented as prototypes and then transferred to an‌ easy-used-well-documented platform where researchers from diverse communities, in‌ particular physicians, will be able to analyze their‌ own data set.

10.4.3 PEPR Digital Health Secure,‌ safe and fair machine learning for healthcare

Participants:‌ Marco Lorenzi.

Duration:
2023 - 2027
Partners:‌
- Inria
- Lamsade (CNRS, Dauphine-PSL)
- CEA
Inria contact:
Aurelien‌ Bellet
Coordinator:
Inria, Dauphine-PSL
Summary:
The goal of‌ this project is to overcome the challenges that‌ prevent the effective use of personalized health data.‌ To achieve this, we will develop new machine‌ learning algorithms that are designed to handle the‌ unique characteristics of multi-scale and heterogeneous individual health‌ data, while providing formal privacy guarantees robustness against‌ adversarial attacks and changes in data dynamics, and‌ fairness for under-represented populations. By addressing these barriers,‌ we hope to unlock the full potential of‌ personalized health data for a wide range of‌ applications.

10.4.4 PEPR Digital Health Stratify Aging

Participants:‌ Marco Lorenzi.

Duration:
2023 - 2027
Partners:‌
- CEA
- Inria
- CNRS
- INSERM
Inria contact:
Marco Lorenzi‌ , Ghiles Reguig
Coordinator:
CEA
Summary:
The goal‌ of StratifyAging is to focus on using high-quality,‌ curated data from clinical research to advance the‌ field of patient stratification through the use of‌ hypothesis-driven approaches or AI algorithms. By harmonizing and‌ aggregating data from various studies, it will be‌ possible to reach a large enough sample size‌ to effectively stratify patients. This process will also‌ lead to the development of standard protocols that‌ can be applied in routine care. As a‌ result, a virtuous cycle may be created, in‌ which standardized data from routine care is collected‌ and analyzed through the Health Data Hub and‌ used to perform population monitoring and develop normative‌ charts and decision support tools.

10.4.5 PEPR Digital‌ Health Rewind

Participants: Marco Lorenzi.

Duration:
2023‌ - 2027
Partners:
- Inria
- CNRS
- INSERM
Inria contact:‌
Marco Lorenzi , John Kalkhof , Giuseppe Orlando‌
Coordinator:
Inria
Summary:
Rewind focuses on the development‌ of new mathematical and statistical approaches for the‌ analysis of multimodal multiscale longitudinal data. These models will be designed, implemented‌ as prototypes and then‌ transferred to an easy-used-well-documented‌‌ platform where researchers from diverse communities, in particular‌ physicians, will be able‌ to analyze their own‌‌ data set.

10.4.6 MediTwin

Participants: Maxime Sermesant,‌ Hervé Delingette, Xavier‌ Pennec.

Duration:
2024‌‌ - 2028
Partners:
- Dassault Systemes
- Inria
- IHUs
- Start-ups‌
Inria contact:
Maxime Sermesant‌
Coordinator:
Dassault Systemes
Summary:‌‌
The MEDITWIN project will offer personalized virtual twins‌ of organs, metabolism and‌ cancer, for better diagnosis‌‌ and treatment. In particular, MEDITWIN will enable doctors‌ to simulate future scenarios‌ for a patient. MEDITWIN‌‌ will enable the industrialization, clinical validation and standardization‌ of these innovations, so‌ that these technologies can‌‌ be deployed in a standardized way and benefit‌ as many people as‌ possible. The best standards‌‌ of care will be incorporated into virtualized experiences‌ made accessible worldwide, setting‌ a new benchmark for‌‌ quality in healthcare and providing a decisive learning‌ ground for progress in‌ medical science. The benefits‌‌ of virtual twins will be assessed for medical‌ teams, patients, and the‌ healthcare system, notably in‌‌ terms of improving the efficiency of care, quality‌ of multidisciplinary decision-making, and‌ effectiveness and safety of‌‌ medical practices and interventions.

10.4.7 DAICAP

Participants: Hervé‌ Delingette.

Duration:
2020‌ - 2025
Partners:
- AP-HP‌‌
- Inria
- Incepto
- CHU Bordeaux, CHU Lille, CHU Strasbourg,‌ Hopitaux Civils de Lyon‌
Inria contact:
Hervé Delingette‌‌
Coordinator:
AP-HP
Summary:
The DAICAP project aims at‌ creating a large multi-centric‌ study (8 clinical centers‌‌ on 5 partner university hospitals) combining multiparametric MR‌ images of the prostate‌ and histology for the‌‌ detection and characterization of prostate cancer. Inria participates‌ to the data collection,‌ quality control of restrospective‌‌ and prospective data from 1250 patients. It also‌ performs the training and‌ evaluation of AI algorithms‌‌ for prostate lesion detection. The infrastructure of the‌ Health Data Hub is‌ used to centralize the‌‌ data and to fine-tune the AI models. The‌ DAICAP project was selected‌ by the Health Data‌‌ Hub, the Grand Défi « Amélioration des diagnostics‌ médicaux par l'Intelligence Artificielle‌ », and Bpifrance in‌‌ July 2020.

10.4.8 AICOO

Participants: Hervé Delingette.‌

Duration:
2024 - 2028‌
Partners:
- Incepto
- Inria
- AP-HP‌‌
- France Imagerie Territoires
- EDL
- Easydoct
Inria contact:
Hervé‌ Delingette
Coordinator:
Incepto
Summary:‌
The AICOO project has‌‌ been selected among the winners of the French‌ national « Innovation in‌ medical imaging » call‌‌ for projects. It aims to transform patient care‌ at the early stage‌ of cancer detection by‌‌ developing an oncology coordination platform. The platform includes‌ AI solutions for the‌ early detection of Prostate‌‌ Cancer and the extraction of advanced biomarkers. In‌ this project, Inria develops‌ advanced machine learning solutions‌‌ for the automatic characterization of prostate lesion malignancy‌ using multiparametric Magnetic Resonance‌ Imaging.

10.4.9 RHU ReBONE‌‌

Participants: Hervé Delingette, Alix de Langlais,‌ Benjamin Billot.

Duration:‌
2024 - 2029
Partners:‌‌
- Partenaires académiques : Université Côte d'Azur, Inserm, Inria,‌ CNRS, Institut Pasteur,Université de‌ Bretagne Occidentale, Université Paris‌‌ Cité, Université Aix-Marseille, Mines‌ de Paris
- Partenaires industriels : Abys Medical, Newclip‌ Technics, Addidream, Aguila Expertise, Digital Medical Hub
- Partenaires‌ Cliniques : CHU Nice, AP-HP
Inria contact:
Hervé‌ Delingette
Coordinator:
CHU Nice
Summary:
The ReBone project‌ has been selected among the winners of the‌ French national « Recherche Hospitalo-Universitaire en santé »‌ (RHU) call for projects. It aims to develop‌ novel personalized, automated, collaborative and validated preoperative planning‌ tools to simulate, prepare and then perform a‌ secure and patient-specific surgical intervention in osteoarticular surgery.‌ In this project, Inria works on the collection‌ and quality control of medical image datasets associated‌ with 3 use cases. We also develop new‌ automated methods to delineate fractured bony structures in‌ CT images and participate to their validations in‌ terms of clinical and industrial use.

10.4.10 RHU‌ TALENT

Participants: Maxime Sermesant, Irene Balelli,‌ Marco Lorenzi.

Duration:
2024 - 2029
Partners:‌
- Partenaires académiques : U Bordeaux, IHU Liryc, Inria‌
- Partenaires industriels : Cardiologs, inHEART, Incepto, AMPS
- Partenaires‌ cliniques : CHU Bordeaux, CHU Dijon
Inria contact:‌
Maxime Sermesant
Coordinator:
Université de Bordeaux
Summary:
The‌ consortium of academic centers and private companies brought‌ together in the TALENT project aims to revolutionize‌ the prediction of stroke risk by developing digital‌ tools capable of detecting this increased risk. The‌ work will focus on the analysis of widely‌ available data, including chest CT scans and/or electrocardiograms,‌ and simpler clinical data such as age or‌ the presence of diabetes. Inria is in charge‌ of image and shape analysis, causal discovery and‌ multimodal learning.

10.4.11 IHU RespirERA

Participants: Hervé Delingette‌, Nicholas Ayache, Benjamin Billot.

Duration:‌
2024 - 2034
Partners:
- Founders of the institute:‌ Inria, Inserm, CHU Nice, Université Côte d'Azur
Inria‌ contact:
Hervé Delingette
Coordinator:
CHU Nice
Summary:
The‌ RespirERA institute has been funded following the third‌ wave of the French national « Institut Hospitalo-Universitaire‌ » (IHU) call for proposals. The institute aims‌ to improve the care in the field of‌ respiratory diseases. The objectives are to reduce the‌ incidence of lung diseases linked to pollution and‌ age and the impact of the exposome (all‌ exposures to environmental factors), extend the life expectancy‌ of patients, delay dependency and progression to respiratory‌ failure and avoid hospitalizations. Inria is coordinating a‌ workpackage in this new institute focusing on AI‌ solutions for lung cancer screening, and biomarker extraction‌ from heterogenous data for the diagnosis of respiratory‌ diseases. Hervé Delingette and Nicholas Ayache are the‌ members of the executive team.

10.4.12 Other national‌ initiatives

Consulting for Industry

Nicholas Ayache has joined‌ the Scientific Advisory Board of Caranx Medical in‌ Oct 2021.
Maxime Sermesant is a scientific advisor‌ for the company inHEART (Bordeaux).

Institute 3IA Côte‌ d'Azur

The 3IA Côte d'Azur is one of‌ the four "Interdisciplinary Institutes of Artificial Intelligence" that‌ were created in France in 2019. Its ambition‌ is to create an innovative ecosystem that is‌ influential at the local, national and international levels, and a focal point‌ of excellence for research,‌ education and the world‌‌ of AI.
Epione is heavily involved in this‌ institute since 5 permanent‌ researchers (Nicholas Ayache‌‌ , Hervé Delingette , Marco Lorenzi , Maxime‌ Sermesant and Xavier Pennec‌ ) are chair holders‌‌ in this institute, and Nicholas Ayache serves as‌ scientific director. The 5‌ Epione chairs were renewed‌‌ in 2023 by an international jury. Hervé Delingette‌ and Nicholas Ayache are‌ members of its scientific‌‌ committee.

Funded projects

Marco Lorenzi is PI of‌ the project Fed-Ops (2025-2029),‌ with IBV, CAL and‌‌ EURECOM. The project is funded by the ANR,‌ and aims at operationalizing‌ federated learning methods and‌‌ software for real-world applications of medical image analysis‌ in multicentric studies.
Marco‌ Lorenzi is co-PI of‌‌ the project FEDERATED-PET (2022-2026), with Prof. Olivier Humbert‌ (CAL, Nice). The project‌ is funded by the‌‌ Institut National du Cancer (INCa), and aims at‌ developing the first French‌ federated learning infrastructure in‌‌ a network of hospitals from the Unicancer consortium.‌
Marco Lorenzi is principal‌ investigator of the project‌‌ TRAIN, funded by the ANR, and co-PI of‌ the project StratifyAging of‌ the PEPR Santé Numerique.‌‌ He also participates to the Horizon Europe Project‌ EUCAIM.

Collaboration with national‌ hospitals

Epione has a‌‌ longstanding collaboration with the IHU-Bordeaux (Pr M. Haïssaguere‌ and Pr P. Jaïs)‌ on cardiac imaging and‌‌ modeling.
Epione also maintains a close partnership with‌ the Brain Institute at‌ Pitié Salpétrière (Dr. O.‌‌ Colliot and Pr. B. Stankoff) on neuroimaging and‌ multiple sclerosis. This year,‌ this collaboration led to‌‌ a common publication 27 and a patent 64‌.
The IHU RespirERA‌ was selected in May‌‌ 2023 among the 12 new institutes in France.‌ This IHU is based‌ in Nice, and focuses‌‌ on respiratory diseases. Inria is one of its‌ 4 founding institutions together‌ with the University Hospital‌‌ of Nice, the Université Côte d'Azur and INSERM.‌ Hervé Delingette and Nicholas‌ Ayache are the members‌‌ of the executive team and are leading a‌ workpackage focusing on AI‌ algorithms for data analysis.‌‌
Several research projects of Epione are part of‌ the joint laboratory Bernouilli‌ between Inria and Assistance‌‌ Publique des Hôpitaux de Paris (AP-HP), in particular,‌ the DAICAP, and PAIMRI‌ projects with Pr Raphaele‌‌ Renard-Penna (Hospital La Pitié Salpêtrière), on prostate cancer‌ detection and characterization.
We‌ also have long term‌‌ collaborations with the CHU Nice, the Centre Antoine‌ Lacassagne of Nice, and‌ the Hospital Lenval of‌‌ Nice.

11 Dissemination

11.1 Promoting scientific activities

11.1.1‌ Scientific events: organization

General‌ chair, scientific chair

Maxime‌‌ Sermesant was the general chair of the IABM‌ (Intelligence Artificielle en Imagerie‌ BioMedicale) conference (300 people)‌‌ organized in Nice (IABM 2025).

Member‌ of the organizing committees‌

Irene Balelli was member‌‌ of the organizing committee of the Complex days‌ (Nice, February), and of‌ the spring school GeMSS/Statlearn‌‌ (Sophia Antipolis, April).
Maxime Sermesant was a co-organizer‌ of the STACOM MICCAI‌ (Medical Image Computing and‌‌ Computer Assisted Intervention, Daejeon,‌ Korea, SE) workshop with 100 participants. He also‌ co-organized the InnovaHeart workshop in Paris (50 people)‌ and the "Imaging and Arrhythmia" workshop in Monaco‌ (50 people). He is co-president of the "One‌ Health" conference series organized by ANRT.

11.1.2‌ Scientific events: selection

Chair of conference program committees‌

Irene Balelli was Program Chair for ECAI 2025‌ (European Conference on Artificial Intelligence, Bologna, Italy). She‌ is also part of the scientific committee of‌ IABM 2026 (Colloque Français en Intelligence Artificielle en‌ Imagerie Biomédicale, Lyon).
Benjamin Billot was Area Chair‌ for MIDL 2025 (Medical Imaging with Deep Learning,‌ Salt Lake City, USA).
Xavier Pennec was a‌ member of the scientific committee of GSI 2025‌ (Geometric Science of Information, Saint-Malo).
Marco Lorenzi was‌ Area Chair for NeurIPS 2025 (Neural Information Processing‌ Systems, San Diego, USA).
Maxime Sermesant was part‌ of the scientific committee of the 2025 SophIA‌ Summit, the 2025 FIMH conference (Functional Imaging and‌ Modeling of the Heart) and the STACOM 2025‌ (Statistical Atlases and Computational Modeling of the Heart)‌ workshop.

Reviewer

Benjamin Billot was a reviewer for‌ MICCAI 2025 (Medical Image Computing and Computer Assisted‌ Intervention, Daejeon, Korea) for which he received the‌ Outstanding reviewer award. He also reviewed workshop proposals‌ for EurIPS 2025 (European Information Processing Systems, Copenhagen,‌ Denmark).
Marco Lorenzi was a reviewer for the‌ conferences MICCAI 2025, ICML 2025 (International Conference on‌ Machine Learning, Vancouver, Canada), CVPR 2025 (Conference on‌ Computer Vision and Pattern Recognition, Nashville, USA), AISTATS‌ 2025 (Artificial Intelligence and Statistics, Thailand).
Hervé Delingette‌ was reviewer for MICCAI 2025, for the MICCAI‌ workshops MLMAI 2025, UNSURE 2025, and DECAF 2025.‌
Francesco Cremonesi was a reviewer for MICCAI 2025.‌
Bernhard Föllmer was a reviewer for MICCAI 2025.‌
Huiyu Li was a reviewer for MIDL 2025.‌
Jairo Rodríguez Padilla was a reviewer for the‌ STACOM workshop of MICCAI 2025 (Statistical Atlases and‌ Computational Modeling of the Heart), as well as‌ for FIMH 2025 (Functional Imaging and Modeling of‌ the Heart, Dallas, USA).
Rafael Silva was a‌ reviewer for ICLR 2025 (International Conference on Learning‌ Representations, Singapore).
Tom Szwagier was a reviewer for‌ GSI 2025.
Maxime Sermesant was a reviewer for‌ the FIMH conference and the STACOM workshop.

11.1.3‌ Journal

Member of editorial boards

Nicholas Ayache is‌ the co-founder and the Co-Editor in Chief with‌ J. Duncan of Medical Image Analysis journal (Elsevier).‌
Nicholas Ayache is a member of the advisory‌ board of the Computer Assisted Surgery journal (Taylor‌ & Francis).
Hervé Delingette is a member of‌ the editorial board of Medical Image Analysis (Elsevier).‌
Marco Lorenzi is member of the editorial board‌ of Medical Image Analysis (Elsevier).
Xavier Pennec is‌ a member of the editorial boards of Medical‌ Image Analysis (Elsevier), the International Journal of Computer‌ Vision (Springer), and of the Journal of Mathematical‌ Imaging and Vision (Springer).
Bernhard Föllmer is associate‌ editor for the Journal of Cardiovascular Computed Tomography‌ (Elsevier).

Reviewer - reviewing activities

Benjamin Billot was reviewer for the following‌ journals: Science, Nature Communications,‌ Medical Image Analysis, IEEE‌‌ transactions on Medical Imaging, NeuroImgae, and Imaging Neuroscience.‌
Irene Balelli was a‌ reviewer for the following‌‌ journals: Vaccine, Medical Image Analysis, Computers in Biology‌ and Medicine, Neuroimage, SMAI‌ J. of Computational Mathematics.‌‌
Francesco Cremonesi was a reviewer for the following‌ journals: Medical Image Analysis,‌ Intelligence-Based Medicine, Computerized Medical‌‌ Imaging and Graphics, and Computers in Biology and‌ Medicine.
Bernhard Föllmer was‌ a reviewer for the‌‌ following Journals: Medical Image Analysis, IEEE Transactions on‌ Medical Imaging, International Journal‌ of Imaging Systems and‌‌ Technology, International Journal of Cardiovascular Imaging, Nature Reports,‌ Quantitative Imaging in Medicine‌ and Surgery.
Maëlis Morier‌‌ was a reviewer for the following journals: SoftwareX‌ and Medical Image Analysis.‌
Rafael Silva was a‌‌ reviewer for Nature Scientific Reports.
Tom Szwagier was‌ a reviewer for the‌ International Journal of Computer‌‌ Vision.
Maxime Sermesant was a reviewer for Medical‌ Image Analysis.

11.1.4 Invited‌ talks

Nicholas Ayache gave‌‌ a series of invited plenary talks at the‌ following events and locations:‌
- “AI, Science, and Society”‌‌ Scientific Conference (Palaiseau, February). This event was part‌ of the French Government's‌ AI Action Summit program‌‌.
- The International Academicians Hong Kong Forum (Hong‌ Kong, March).
- The Hong‌ Kong University of Science‌‌ and Technology (Hong Kong, March).
- The Chinese University‌ of Hong Kong (Hong‌ Kong, March).
- First Conference‌‌ on Data Science 4 Health and Biology (DS4HB)‌, Politecnico di Milano‌ (Milan, Italy, April)
- The‌‌ French Academy of Sciences (Paris, April).
- The French‌ Academy of Medicine (Paris,‌ June).
Benjamin Billot participated‌‌ to the following events:
- invited speaker at the‌ Statlearn'25 Spring School (Sophia-Antipolis,‌ March) and at the‌‌ 3IA Côte d'Azur (Sophia-Antipolis, May).
- pannelist in the‌ webinar "Best practices for‌ MICCAI reviews and rebuttals"‌‌ (online, February).
Xavier Pennec participated to the following‌ events:
- keynote speaker at‌ the 2025 Workshop on‌‌ Geometry, Topology, and Machine Learning (Leipsiz, Germany, November).‌
- keynote courses at the‌ Schrödinger institute (ESI) for‌‌ the Program on Infinite-dimensional Geometry: Theory and Applications‌ (Vienna, Austria, February).
- was‌ invited speaker at: the‌‌ French Académie des Sciences (May, Paris); AfterShape 2025‌ workshop (Saclay, June); Mathematical‌ Imaging and Surface Processing‌‌ workshop (Oberwolfach, Germany, February); Geometry for statistics and‌ AI workshop (Lesbos, Greece,‌ Oberwolfach, May); Séminaire Données‌‌ et Aléatoire Théorie & Applications, (Grenoble, June).
Marco‌ Lorenzi participated to the‌ following events:
- keynote talk‌‌ at the Annual Meeting of the PEPR Santé‌ Numerique (Rennes, October).
- invitation‌ by the French Embassy‌‌ in Japan to present to the France-Japan bilateral‌ seminar on health data‌ hosted by the Keio‌‌ University School of Medicine (Tokyo, Japan, June).
- invited‌ lecturer at: RUB University‌ (Germany, June), University of‌‌ Queensland (Australia, October), and to Geneva University Hospitals‌ (Switzerland, June).
- panelist in‌ the round table “AI‌‌ in healthcare” of the AI4People Summit 2025 held‌ by AI4People Institute and‌ the European Parliament.
Herve‌‌ Delingette gave invited presentations during the following events:‌
- Sophia Summit 2025 (Sophia‌ Antipolis, Novermber)
- The 5th‌‌ Joint meeting on Lung‌ Cancer (Nice, October)
- IABM 2025 meeting (Nice, March).‌
Maxime Sermesant gave invited talks at:
- the King's‌ College London Doctoral School
- the Health Data Hub‌ - Citadel meeting in Montreal
- the Imaging Workshop,‌ IHU Liryc.

11.1.5 Research administration

Nicholas Ayache has‌ been the scientific director of the 3IA Cote-d'Azur‌ for 6 years since its creation (Sept 2019-‌ Oct 2025) and Chair of its Scientific Council.‌ He has been a member of the scientific‌ council of the Mécénat Santé program of the‌ AXA group (2024-2025). He is a member of‌ the scientific advisory board of the start-up companies‌ inHeart (digital heart) and Caranx Medical (Medical Robotics).‌
Nicholas Ayache is a member of the French‌ Academy of Sciences, and participates to the scientific‌ activities of two of its sections (Computer Science‌ and Applications of Sciences). He is also a‌ member of the French Academy of Surgery.
Irene‌ Balelli is a member of the scientific advisory‌ board of the GIS (scientific interest group) FC3R‌ since July 2023, and of the Scientific committee‌ of the Academy 2 (Complex Systems) since November‌ 2023.
Irene Balelli is in charge of the‌ pedagogical orgization of the AI for Health track‌ of the Data Science & AI Master, Université‌ Côte d'Azur, France.
Irene Balelli is member of‌ the NICE committee since October 2025.
Xavier Pennec‌ is co-director of the Ecole doctorale STIC of‌ Université Côte d'Azur. He is a member of‌ the committee of EDSTIC, of the Doctoral follow-up‌ Committee (CSD) at Inria Sophia Antipolis. He was‌ also a member of the executive committee of‌ the Academy 4 (Living systems Complexity and diversity)‌ of the IDEX JEDI at University Côte d'Azur‌ up to October 2025.
Xavier Pennec was a‌ member of the grant panel for the Collaborative‌ Research in Computational Neuroscience (CRCNS) call 2025.
Marco‌ Lorenzi became member of the European Laboratory for‌ Learning and Intelligent Systems (ELLIS).
Marco Lorenzi is‌ member of the Member of the Comité de‌ Centre of the Centre Inria d'Université Côte d'Azur.‌ He is External Advisory Board of the HealthData@EU‌ Pilot project.
Marco Lorenzi was a member of‌ the grant panel of the call for DataIA‌ Fellowships 2025.
Hervé Delingette is one of the‌ 2 scientific directors of the IdEx program UCA‌ JEDI under the direction of the IdEX vice-president‌ of the Université Côte d'Azur. He is an‌ administrator and a member of the scientific committee‌ of the Groupement de Coopération Sanitaires (GCS) CARES‌ involving the Université Côte d'Azur and the 3‌ local hospitals (CHU Nice, Centre Antoine Lacassagne, Fondation‌ Lenval). He is also a member of the‌ research and innovation committee organized by the employer‌ union UPE06.
Hervé Delingette is the contact person‌ at the Inria center of Université Côte d'Azur‌ for research data management. Hervé Delingette is the‌ Inria representative at the executive committee of the‌ DATAZUR structure, helping Université Côte d'Azur researchers handle‌ their research data.

11.2 Teaching - Supervision - Juries - Educational and‌ pedagogical outreach

11.2.1 Teaching‌

Master: Irene Balelli ,‌‌ Research awareness, 6h EURECOM, Sophia Antipolis.
Licence: Irene‌ Balelli , Advanced statistical‌ modeling, 22.5h ETD, Univ.‌‌ Côte d'Azur, France.
Licence: Irene Balelli , Statistical‌ modeling for complex data‌ and Big Data, 37.5h‌‌ ETD, Univ. Côte d'Azur, France.
Master: Hervé Delingette‌ and Xavier Pennec ,‌ Medical Image Analysis based‌‌ on generative, geometric and biophysical models, 21h course‌ (28.5 ETD), Master 2‌ MVA, ENS Saclay, France.‌‌
Master: Hervé Delingette and Xavier Pennec , Medical‌ Image Processing, 24h course,‌ Master Data-Science and Artificial‌‌ Intelligence, Université Côte d'Azur, France.
Master: Hervé Delingette‌ , AI in Digital‌ Pathology, 3h course, Master‌‌ of Science Biobanks and Complex Data Management, Univ.‌ Côte d'Azur, France.
Hervé‌ Delingette , 2h course,‌‌ Diplôme d'Etudes Supérieures Inter Universitaires - Réutilisation de‌ données pour la recherche‌ en santé (DESIU REDS)‌‌
Master: Hervé Delingette , AI in Oncology 2h‌ course, Master Cancérologie et‌ Recherche Translationnelle, Univ. Côte‌‌ d'Azur, France.
Master: Marco Lorenzi and V. Alessandro,‌ Bayesian Learning, 30h course,‌ Master Data Science, Univ.‌‌ Côte d'Azur, France.
Master: Marco Lorenzi , Model‌ Selection and Resampling Methods,‌ 30h course, Master Data‌‌ Science, Univ. Côte d'Azur, France.
Marco Lorenzi and‌ Francesco Cremonesi presented a‌ 6 hour workshop with‌‌ title "Analyse des données de santé sensibles par‌ l'apprentissage fédéré” for Inria‌ Academy, Valbonne, France
Marco‌‌ Lorenzi , Francesco Cremonesi , and Lucie Chambon‌ , Federated Learning and‌ the Fed-BioMed software AI4Health‌‌ Summer School, 1h ETD, Paris.
Master: Francesco Cremonesi‌ , Federated Learning, 8h‌ ETD, École d'ingénieur ISIS,‌‌ Castres, France
Master: Francesco Cremonesi , Federated Learning,‌ 6h ETD, École Centrale‌ Marseille, Marseille, France
Lucie‌‌ Chambon , Francesco Cremonesi and John Kalkhof ,‌ Federated Learning and the‌ Fed-BioMed software, AI for‌‌ maternal health: machine learning, federated learning, and ethical‌ innovation for obstetric ultrasound,‌ 6h ETD, University of‌‌ Embu, Embu, Kenya.
Master: Lucie Chambon and Francesco‌ Cremonesi , Federated Learning,‌ 16h ETD, École d'ingénieur‌‌ ISIS, Castres, France
Master: Lucie Chambon and Ali‌ Tolga Dincer , 6h‌ ETD, Master DSAI, Université‌‌ Cote d'Azur
License: Olivier Bisson , L2 Math‌ (Calculus II), 16h ETD,‌ and L1 Math (Introduction‌‌ à l'analyse), 36h ETD, Univ. Côte d'Azur, France.‌
Project Management: Gaëtan Desrues‌ , Mathématiques Appliquées et‌‌ Modélisation (Ingénierie Numérique), 20h, École d'ingénieur Polytech Nice‌ Sophia, France
License: Nicolas‌ Drettakis , Introduction à‌‌ l'informatique, 18h ETD, Université Nice Cote d'Azur, Nice,‌ France.
Bachelor: Giulia Foroni‌ and Olivier Humbert ,‌‌ Séminaire IA et santé SANURN, 6h ETD, School‌ of Medicine, Université Cote‌ d'Azur, Nice, France.
Master:‌‌ Jairo Rodríguez Padilla , Cardiac Digital Twin, 15h‌ ETD, École d'ingénieur ISEN‌ Yncréa Ouest, Caen, France.‌‌
Master: Rafael Silva , Électronique Analogique, 54h TP,‌ Polytech Nice Sophia, Sophia‌ Antipolis, France
University Diploma:‌‌ Rafael Silva , DU Intelligence Artificielle et Santé,‌ 8h TD, Faculté de‌ Médecine, Nice, France

11.2.2‌‌ Supervision: defended PhDs

Lisa Guzzi , Automatic segmentation‌ of the vascular system‌ to enhance AI-based decision‌‌ support system for peripheral‌ artery disease 54, Université Côte d'Azur, 3IA‌ fellowship. Started in 2022. Directed by Hervé Delingette‌ , Juliette Raffort-Lareyre
Tom Szwagier , Rethinking statistical‌ methods with Flags spaces 55, Université Côte‌ d'Azur. Started in 2022. Directed by Xavier Pennec‌ .

11.2.3 Supervision: ongoing PhDs

Amel Bakhouche :‌ Analysis of European National Health data and vascular‌ registries to better understand the outcomes of patients‌ with vascular diseases. CHU-Inria PhD started in February‌ 2025. Co-directed by Hervé Delingette , Irene Balelli‌ and Juliette Raffort-Lareyre (CHU Nice).
Olivier Bisson :‌ Géométrie, Stratification et application des matrices de corrélation‌ structurées. Directed by Xavier Pennec . 3IA PhD‌ started in October 2023.
Florencia Boccarato : Effective‌ AI based Characterization of Prostate Cancer from Multiparametric‌ MRI. Directed by Hervé Delingette and Raphaele Renard-Penna‌ (PUPH Sorbonne Université, AP-HP), funded by the AICOO‌ project.
Fahym Bounazou : Early AI-based detection of‌ prostate cancer from multiparametric MRI. Directed by Hervé‌ Delingette and Raphaele Renard-Penna (PUPH Sorbonne Université, AP-HP),‌ funded by the AICOO project.
Alix de Langlais‌ : Automatic generation of three-dimensional models of extremity‌ fractures of proximal humerus for preoperative planning and‌ intraoperative assistance in mixed reality, Inria-INSERM funded PhD‌ started in June 2024. Co-directed by Hervé Delingette‌ , and Marc-Olivier Gauci (Orthopedics Surgeon, CHU Nice,‌ IBV).
Nicolas Drettakis : CT scan-based prediction of‌ stroke risk from the shape of the left‌ atrium. Directed by Maxime Sermesant .
Ezem Sura‌ Ekmekci : Temporal Boundary Distillation Module for Surgical‌ Gesture Segmentation. Directed by Nicholas Ayache and François‌ Brémond , and co-supervised by Hervé Delingette and‌ Pierre Berthet-Rayne .
Federica Facente : Learning Statistical‌ and Biomedical Models for multimodal image analysis -‌ application to Image Guided Surgical Robotics, 3IA PhD‌ started in September 2023. Co-directed by Nicholas Ayache‌ , Pierre Berthet-Rayne (CTO and co-founder of Caranx-Medical,‌ 3IA affiliate chair holder) and Hervé Delingette .‌
Camilla Ferrario : Electromechanical modeling of non-ischemic cardomyopathies,‌ funded by PEPR Digital Health ChroniCardio, started in‌ September 2024.
Giulia Foroni : Development of survival‌ analysis methods to model spatio-temporal changes in PET/CT‌ images of lung cancer patients treated with immunotherapy.‌ 3IA funding. Co-directed by Marco Lorenzi and Olivier‌ Humbert (IBV, CAL).
Sébastien Goffart : Development of‌ predictive models in patients with peripheral artery disease,‌ Université Côte d'Azur. ANR grant handled by CHU‌ Nice. Started in 2023. Co-directed by Hervé Delingette‌ , and Juliette Raffort-Lareyre.
Manasi Kattel : Deep‌ learning methods for the analysis and registration of‌ US images, Université Côte d'Azur. 3IA Côte d'Azur‌ fellowship. Started in December 2023. Co-directed by Nicholas‌ Ayache and Hervé Delingette .
Wassila Khatir :‌ Integromics analysis: a new approach to study the‌ pathophysiology of X-Fragile Syndrome (FXS), Université Côte d'Azur.‌ Neuromod fellowship. Started in March 2024. Co-directed by‌ Irene Balelli , Marco Lorenzi and Carole Gwizdek‌ (IPMC).
Arnaud Lang : Multimodal prediction for Cardioembolic‌ stroke: etiology and risk. RHU TALENT funded PhD‌ started in December 2025. Co-directed by Irene Balelli , Marco Lorenzi and‌ Maxime Sermesant .
Maëlis‌ Morier : Deep Learning‌‌ Meets Numerical Modeling, AI and Biophysics for Computational‌ Cardiology, started in 2023.‌ Co-supervised by Maxime Sermesant‌‌ and Patrick Gallinari (Sorbonne Universités).
Huyen Trang Nguyen‌ : Robust Biomarker Extraction‌ in PET-CT Imaging Data‌‌ for Immunotherapy in Lung Cancer. Franco-German ANR Train,‌ co-directed with Olivier Humbert‌ (Centre Antoine Lacassagne).
Evariste‌‌ Njomgue Fotso : Multimodal learning for sudden cardiac‌ death risk prediction. Funded‌ by MediTwin, started in‌‌ 2023.
Giuseppe Orlando : Vision-Language analysis for multimodal‌ analysis of health records‌ and PET/CT imaging data‌‌ in lung cancer. Project funced by PEPR Santé‌ Numerique. Co-supervision of Marco‌ Lorenzi and Olivier Humbert‌‌ (IBV, CAL).
Rafael Silva : Artificial Intelligence for‌ Cardiac Monitoring: Portable Multimodal‌ Cardiac Function Analysis. Started‌‌ in 2023. Co-directed by Maxime Sermesant and Pamela‌ Moceri (CHU Nice).
Adrien‌ Tchuem Tchuente : Generative‌‌ AI applied to multi-modal and multi-scale cardiac imaging.‌ Directed by Maxime Sermesant‌ .
Elie Thellier :‌‌ Generative AI for the anonymization of medical images.‌ Directed by Hervé Delingette‌ and Nicholas Ayache ,‌‌ funded by project PLICIA.
Tony Zaayter : Geometric‌ statistics on stratified quotient‌ spaces: topologically constrained multi-atlases‌‌ for brain diffeomorphometry. Directed by Xavier Pennec and‌ co-supervised by Mathieu Carrière‌ . Funded by Meditwin,‌‌ started in Nov 2025.

11.2.4 Juries

Irene Balelli‌ was Member of the‌ admissibility jury for the‌‌ 2025 recruitment campaign for CRCN/ISFP at Inria Center‌ at Université Côte d'Azur.‌
Xavier Pennec was a‌‌ member of the jury and reviewer of the‌ HDR of Benjamin Charlier‌ (Univ. Montpellier), a member‌‌ of the jury and reviewer of the PhD‌ of Alexey Lazarev (Univ.‌ Toulouse) and a member‌‌ of the PhD jury of of Gaël Le‌ Ruz (Univ. Paris Sorbonne).‌ He was also a‌‌ jury member of the PhD of Tom Szwagier‌ as PhD advisor.
Marco‌ Lorenzi was reviewer and‌‌ member of the HDR of Kassem Kallas (University‌ of Western Brittany), a‌ member of the jury‌‌ and reviewer of the PhD of Matthis Manthe‌ (Université de Lyon), of‌ Jon Middleton (University of‌‌ Copenhagen), and reviewer of the PhD of Martin‌ Saint-Jalmes (University of Melbourne).‌
Hervé Delingette was reviewer‌‌ of the thesis of Emma Sarfati (Institut polytechnique‌ de Paris, Telecom Paris),‌ of Robin Cremese (Université‌‌ Paris sciences et lettres, Institut Pasteur), of Ali‌ Keshavarzi (Institut Polytechnique de‌ Paris, Telecom Paris). He‌‌ chaired the PhD defense of Faisal Jayousi (Université‌ Côte d'Azur, Inria Morpheme),‌ and of Yanis Aeschlimann‌‌ (Université Côte d'Azur, Inria Cronos). He was a‌ member of the PhD‌ defense of Lisa Guzzi‌‌ (Université Côte d'Azur) as its PhD co-director, and‌ co-supervisor. He was a‌ member of the recruitment‌‌ committee for the 2025 Inria "Chaire de Professeur‌ Junior" on Trustworthy AI‌ for Personalized Medicine.‌‌

11.3 Popularization

11.3.1 Specific official responsibilities in science‌ outreach structures

Irene Balelli‌ is part of the‌‌ EssentiElles Santé network, for women in healthcare science.‌

11.3.2 Participation in Live‌ events

Hervé Delingette gave‌‌ an invited presentation at‌ the Terra Numerica institute on the topics of‌ "AI for Medicine" on February 26th in Sophia‌ Antipolis.

12 Scientific production

12.1 Major publications

1‌ articleN.Nicolas Cedilnik, M.Mihaela Pop‌, J.Josselin Duchateau, F.Frédéric Sacher‌, P.Pierre Jaïs, H.Hubert Cochet‌ and M.Maxime Sermesant. Efficient Patient-Specific Simulations‌ of Ventricular Tachycardia Based on Computed Tomography-Defined Wall‌ Thickness Heterogeneity.JACC: Clinical ElectrophysiologySeptember 2023‌HAL DOI
2 articleH.Hind Dadoun,‌ H.Hervé Delingette, A.-L.Anne-Laure Rousseau,‌ E.Eric de Kerviler and N.Nicholas Ayache‌. Deep Clustering for Abdominal Organ Classification in‌ US imaging.Journal of Medical Imaging10‌32023, 034502HAL DOI
3 article‌Y.Yann Fraboni, R.Richard Vidal,‌ L.Laetitia Kameni and M.Marco Lorenzi.‌ A General Theory for Federated Optimization with Asynchronous‌ and Heterogeneous Clients Updates.Journal of Machine‌ Learning Research24March 2023, 1-43HAL‌
4 miscY.Yann Fraboni, R.Richard‌ Vidal, L.Laetitia Kameni and M.Marco‌ Lorenzi. Clustered Sampling: Low-Variance and Improved Representativity‌ for Clients Selection in Federated Learning.May‌ 2021HAL
5 articleN.Nicolas Guigui,‌ N.Nina Miolane and X.Xavier Pennec.‌ Introduction to Riemannian Geometry and Geometric Statistics: from‌ basic theory to implementation with Geomstats.Foundations‌ and Trends in Machine Learning163February‌ 2023, 329-493HALDOI
6 articleD.‌Dimitri Hamzaoui, S.Sarah Montagne, R.‌Raphaële Renard-Penna, N.Nicholas Ayache and H.‌Hervé Delingette. Morphologically-Aware Consensus Computation via Heuristics-based‌ IterATive Optimization (MACCHIatO).Journal of Machine Learning‌ for Biomedical Imaging2UNSURE 2022 Special Issue‌September 2023, 361-389HAL DOI
7 inproceedings‌J.Josquin Harrison, J.James Benn and‌ M.Maxime Sermesant. Improving Neural Network Surface‌ Processing with Principal Curvatures.NeurIPS ProceedingsNeurips‌ 2024 - 38th Annual Conference on Neural Information‌ Processing Systems2024Vancouver, CanadaDecember 2024HAL‌
8 articleV.Victoriya Kashtanova, M.Mihaela‌ Pop, I.Ibrahim Ayed, P.Patrick‌ Gallinari and M.Maxime Sermesant. Simultaneous data‌ assimilation and cardiac electrophysiology model correction using differentiable‌ physics and deep learning.Interface Focus13‌6December 2023HALDOI
9 articleJ.‌Julian Krebs, H.Hervé Delingette, N.‌Nicholas Ayache and T.Tommaso Mansi. Learning‌ a Generative Motion Model from Image Sequences based‌ on a Latent Motion Matrix.IEEE Transactions‌ on Medical ImagingFebruary 2021HAL DOI
10‌ articleM.Maxime Sermesant, H.Hervé Delingette‌, H.Hubert Cochet, P.Pierre Jaïs‌ and N.Nicholas Ayache. Applications of artificial‌ intelligence in cardiovascular imaging.Nature Reviews Cardiology‌March 2021HAL DOI
11 articleP.Paul‌ Tourniaire, M.Marius Ilie, P.Paul‌ Hofman, N.Nicholas Ayache and H.Hervé‌ Delingette. MS-CLAM: Mixed Supervision for the classification and localization of tumors‌ in Whole Slide Images‌.Medical Image Analysis‌‌852023, 102763HAL DOI
12 article‌Z.Zihao Wang,‌ T.Thomas Demarcy,‌‌ C.Clair Vandersteen, D.Dan Gnansia,‌ C.Charles Raffaelli,‌ N.Nicolas Guevara and‌‌ H.Hervé Delingette. Bayesian Logistic Shape Model‌ Inference: application to cochlear‌ image segmentation.Medical‌‌ Image AnalysisOctober 2021HAL
13 articleY.‌Yingyu Yang, M.‌Marie Rocher, P.‌‌Pamela Moceri and M.Maxime Sermesant. Echocardiography‌ Analysis with Deep Learning‌ using Priors: Multi-centric Evaluation‌‌ of Generalisation.Journal of Machine Learning for‌ Biomedical Imaging2November‌ 2024November 2024,‌‌ 2293-2325HAL DOI

12.2 Publications of the year‌

International journals

14 article‌V.Vincenzo Carbone,‌‌ I.Irene Balelli, Y.Yves Coudière,‌ O.Oscar Camara,‌ B.Beatriz Trenor,‌‌ M.Michèle Barbier and M.Maxime Sermesant.‌ A cloud-based platform for‌ seamless digital testing of‌‌ cardiac medical devices and drugs.The Project‌ Repository Journal241‌August 2025, 144-148‌‌HAL DOI
15 articleV.Vincenzo Carbone,‌ I.Irene Balelli,‌ Y.Yves Coudière,‌‌ O.Oscar Camara, B.Beatriz Trenor,‌ M.Maxime Sermesant and‌ M.Michele Barbier.‌‌ A cloud-based platform for seamless digital testing of‌ cardiac medical devices and‌ drugs.The Project‌‌ Repository JournalAugust 2025HAL DOI
16 article‌A.Andrea Chierici,‌ L.Lisa Guzzi,‌‌ S.Sebastien Goffart, N.Nizar Kamoun,‌ M.Manuel Gargiulo,‌ P.Patrick Chevallier,‌‌ A.Antonio Iannelli, R.Rodolphe Anty,‌ H.Hervé Delingette,‌ F.Fabien Lareyre and‌‌ J.Juliette Raffort. Fully Automatic Artificial Intelligence‌ Liver Anatomy Segmentation in‌ the Management of Colorectal‌‌ Liver Metastases.Cureus Journal of Medical Science‌June 2025HAL DOI‌
17 articleA.Andrea‌‌ Chierici, F.Fabien Lareyre, A.Antonio‌ Iannelli, B.Benjamin‌ Salucki, S.Sébastien‌‌ Goffart, L.Lisa Guzzi, E.Elise‌ Poggi, H.Hervé‌ Delingette and J.Juliette‌‌ Raffort. Applications of artificial intelligence in liver‌ cancer: A scoping review‌.Artificial Intelligence in‌‌ Medicine169November 2025, 103244HAL DOI‌
18 articleM.Marco‌ Corneli, E.Elena‌‌ Erosheva, X.Xunlei Qian and M.Marco‌ Lorenzi. A Bayesian‌ approach for clustering and‌‌ exact finite-sample model selection in longitudinal data mixtures‌.Computational Statistics40‌2025, 509–545HAL‌‌DOI
19 articleS.Sébastien Frey, F.‌Federica Facente, W.‌Wen Wei, E.‌‌ S.Ezem Sura Ekmekci, E.Eric Séjor‌, P.Patrick Baqué‌, M.Matthieu Durand‌‌, H.Hervé Delingette, F.Francois Bremond‌, P.Pierre Berthet-Rayne‌ and N.Nicholas Ayache‌‌. Optimizing Intraoperative AI: Evaluation of YOLOv8 for‌ Real-Time Recognition of Robotic‌ and Laparoscopic Instruments.‌‌Journal of Robotic Surgery19131March 2025‌HAL DOI back to‌ text
20 articleS.‌‌Sebastien Goffart, H.‌Hervé Delingette, A.Andrea Chierici, L.‌Lisa Guzzi, B.Bahaa Nasr, F.‌Fabien Lareyre and J.Juliette Raffort. Artificial‌ Intelligence Techniques for Prognostic and Diagnostic Assessments in‌ Peripheral Artery Disease: A Scoping Review.Angiology‌2025. In press. HAL DOI back to‌ text
21 articleS.Sébastien Goffart, O.‌Odette Hart, F.Fabien Lareyre, L.‌Lisa Guzzi, K. K.Kak Khee Yeung‌, H.Hervé Delingette, M.Manar Khashram‌ and J.Juliette Raffort. Deep Learning Strategies‌ for Predicting Amputation Free Survival in Patients with‌ Peripheral Artery Disease.European Journal of Vascular‌ and Endovascular SurgeryOctober 2025HAL DOI back‌ to text
22 articleJ.Jia Guo,‌ F.Fabien Lareyre, S.Sébastien Goffart,‌ A.Andrea Chierici, H.Hervé Delingette and‌ J.Juliette Raffort. Automatic Segmentation of Intraluminal‌ Thrombus in Abdominal Aortic Aneurysms Based on CT‌ Images: A Comprehensive Review of Deep Learning-Based Methods‌.Journal of Clinical Medicine1423November‌ 2025, 8497HALDOI
23 articleJ.‌Jia Guo, F.Fabien Lareyre, R.‌Regent Lee, M.Martin Teraa, H.‌Hervé Delingette and J.Juliette Raffort. Artificial‌ Intelligence and Machine Learning for Risk Prediction of‌ Abdominal Aortic Aneurysm Growth and Rupture.Angiology‌October 2025HAL DOIback to text
24‌ articleF.Fabien Lareyre, L.Lisa Guzzi‌, B.Bahaa Nasr, A.Ahmed Alouane‌, S.Sébastien Goffart, A.Andréa Chierici‌, H.Hervé Delingette and J.Juliette Raffort‌. Imaging Characterisation of Peripheral Artery Disease: A‌ Scoping Review on Current Classifications and New Insights‌ Brought by Artificial Intelligence.EJVES Vascular Forum‌642025, 87-95HAL DOI
25 article‌H.Huiyu Li, N.Nicholas Ayache and‌ H.Hervé Delingette. Data Exfiltration by Compression‌ Attack: Definition and Evaluation on Medical Image Data‌.Journal of Machine Learning for Biomedical Imaging‌3December 2025December 2025, 728-756HAL‌DOI back to text
26 articleS.Sébastien‌ Molière, D.Dimitri Hamzaoui, G.Guillaume‌ Ploussard, R.Romain Mathieu, G.Gaëlle‌ Fiard, M.Michael Baboudjian, b.benjamin‌ granger, M.Morgan Roupret, H.Hervé‌ Delingette and R.Raphaële Renard-Penna. A Systematic‌ Review of the Diagnostic Accuracy of Deep Learning‌ Models for the Automatic Detection, Localization, and Characterization‌ of Clinically Significant Prostate Cancer on Magnetic Resonance‌ Imaging.European Urology Oncology84August‌ 2025, 1182-1202HALDOI
27 articleT.‌T. Soulier, N.Ninon Burgos, R.‌Ravi Hassanaly, M.M. Pitombeira, M.‌Maëlys Solal, H.Hugues Roy, M.‌M. Hamzaoui, A.A. Yazdan-Panah, D.‌D. de Paula Faria, C.C. Louapre‌, B.B. Bodini, M.M. Bottlaender‌, N.Nicholas Ayache, O.Olivier Colliot‌ and B.B. Stankoff. Artificial intelligence in presymptomatic neurological diseases: Bridging‌ normal variation and prodromal‌ signatures.Revue Neurologique‌‌1819November 2025, 944-950HAL DOI‌back to text
28‌ articleT.Tom Szwagier‌‌, P.-A.Pierre-Alexandre Mattei, C.Charles Bouveyron‌ and X.Xavier Pennec‌. Parsimonious Gaussian mixture‌‌ models with piecewise-constant eigenvalue profiles.Statistics and‌ Computing2025. In‌ press. HAL back to‌‌ text back to text
29 articleT.Tom‌ Szwagier and X.Xavier‌ Pennec. The curse‌‌ of isotropy: from principal components to principal subspaces‌.Statistical Science2025‌. In press. HAL‌‌DOI back to text
30 articleZ.Zihao‌ Wang, C.Clair‌ Vandersteen, C.Charles‌‌ Raffaelli, N.Nicolas Guevara and H.Hervé‌ Delingette. Multi-Energy Quasi-Symplectic‌ Langevin Inference for Latent‌‌ Disentangled Learning.IEEE Transactions on Image Processing‌342025, 7037-7049‌HAL DOI
31 article‌‌W.Wilhelm Wimmer and H.Hervé Delingette.‌ Second order kinematic surface‌ fitting in anatomical structures‌‌.Medical Image AnalysisFebruary 2025, 103488‌HAL DOI

International peer-reviewed‌ conferences

32 inproceedingsS.‌‌Safaa Al-Ali, J.Jairo Rodríguez-Padilla, M.‌Maxime Sermesant and I.‌Irene Balelli. Cardiac‌‌ Electromechanical Model Sensitivity Analysis using Causal Discovery.‌Lecture notes in computer‌ scienceFIMH 2025 -‌‌ 13th Functional Imaging and Modeling of the Heart‌ International ConferenceFunctional Imaging‌ and Modeling of the‌‌ Heart. FIMH 2025 : part 1LNCS-15672Dallas,‌ United States2025HAL‌DOI back to text‌‌
33 inproceedingsB.Benjamin Billot, R.Ramya‌ Muthukrishnan, E.Esra‌ Abaci Turk, P.‌‌ E.P. Ellen Grant, N.Nicholas Ayache‌, H.Hervé Delingette‌ and P.Polina Golland‌‌. Spatial Regularisation for Improved Accuracy and Interpretability‌ in Keypoint-Based Registration.‌MICCAI 2025 - 28th‌‌ International Conference on Medical Image Computing and Computer‌ Assisted Intervention15973Daejeon,‌ South KoreaSpringer Nature‌‌September 2026, 583-593HAL DOI back to‌ text
34 inproceedingsO.‌Olivier Bisson, Y.‌‌Yanis Aeschlimann, S.Samuel Deslauriers-Gauthier and X.‌Xavier Pennec. Log-Euclidean‌ Frameworks for Smooth Brain‌‌ Connectivity Trajectories.GSI'25 - International Conference on‌ Geometric Science of Information‌Saint-Malo (France), France2025‌‌HAL back to textback to text
35‌ inproceedingsC.Camilla Ferrario‌, J.Jairo Rodríguez-Padilla‌‌, M.Maelys Venet, O.Olivier Villemain‌ and M.Maxime Sermesant‌. Myocardial Stiffness Quantification‌‌ using Ultrasound Shear Wave Elastography and Reduced Modeling‌ for Subject-specific Simulations.‌Lecture Notes in Computer‌‌ ScienceFIMH 2025 - 13th Functional Imaging and‌ Modeling of the Heart‌15673Functional Imaging and‌‌ Modeling of the Heart. FIMH 2025LNCS-15673Dallas‌ (TX), United StatesJuly‌ 2025, 41-54HAL‌‌DOI back to text
36 inproceedingsE. N.‌Evariste Njomgue Fotso,‌ M.Marta Nuñez-Garcia,‌‌ B.Buntheng Ly, H.Hubert Cochet and‌ M.Maxime Sermesant.‌ Uncertainty-Informed Multimodal Infarct Age‌‌ Prediction from Imaging and Clinical Data.Lecture‌ Notes in Computer Science‌FIMH 2025 - Functional‌‌ Imaging and Modeling of‌ the HeartFunctional Imaging and Modeling of the‌ Heart. FIMH 2025LNCS-15673Dallas (TX), United States‌June 2025HAL DOIback to text
37‌ inproceedingsL.Lisa Guzzi, M. A.Maria‌ A Zuluaga, F.Fabien Lareyre, G.‌ D.Gilles Di Lorenzo, S.Sébastien Goffart‌, A.Andrea Chierici, J.Juliette Raffort‌ and H.Hervé Delingette. Automatic Segmentation of‌ Lower-Limb Arteries on CTA for Pre-surgical Planning of‌ Peripheral Artery Disease.MICCAI-AMAI2025 - 4th Workshop‌ on Applications of Medical Artifical IntelligenceDaejeon, South‌ KoreaSeptember 2025HALback to text
38‌ inproceedingsL.Lisa Guzzi, M. A.Maria‌ A Zuluaga, R.Riccardo Taiello, F.‌Fabien Lareyre, G. D.Gilles Di Lorenzo‌, S.Sébastien Goffart, A.Andrea Chierici‌, J.Juliette Raffort and H.Hervé Delingette‌. Regional Hausdorff Distance Losses for Medical Image‌ Segmentation.MLMI 2025 - 16th International Workshop‌ on Machine Learning in Medical Imaging (In conjunction‌ with MICCAI 2025)Daejeon, South KoreaSeptember 2025‌HAL back to text
39 inproceedingsM.Manasi‌ Kattel, B.Benjamin Billot, F.Federica‌ Facente, H.Hervé Delingette and N.Nicholas‌ Ayache. Robust rigid MRI-TRUS registration in prostate‌ cancer using attention-CNN and ICP.6th International‌ Workshop of Advances in Simplifying Medical UltraSound (ASMUS)‌Daejeon, South KoreaSeptember 2025HAL back to‌ text
40 inproceedingsH.Huiyu Li, N.‌Nicholas Ayache and H.Hervé Delingette. Generative‌ medical image anonymization based on latent code projection‌ and optimization.IEEE XploreIEEE International Symposium‌ on Biomedical Imaging (ISBI 2025)Houston (Texas), United‌ StatesApril 2025HALback to text
41‌ inproceedingsB.Buntheng Ly, N.Nicolas Cedilnik‌, M.Mihaela Pop, J.Josselin Duchateau‌, F.Frédéric Sacher, P.Pierre Jaïs‌, H.Hubert Cochet and M.Maxime Sermesant‌. Multimodal Personalisation of Cardiac Electrophysiology Models combining‌ 12-lead ECG and Computed Tomography.Lecture Notes‌ in Computer ScienceFIMH 2025 - Functional Imaging‌ and Modeling of the HeartFunctional Imaging and‌ Modeling of the Heart. FIMH 2025LNCS-15672Dallas‌ (TX), United StatesJune 2025HAL DOI back‌ to text back to text
42 inproceedingsJ.‌ R.Jairo Rodríguez Padilla, R.Rafael Silva‌, B.Buntheng Ly, G.Graham Wright‌, M.Mihaela Pop and M.Maxime Sermesant‌. In silico Assessment of Arrhythmia Inducibility Dependence‌ on Stimulus Location using Calibrated MR-based Infarcted Heart‌ Models.Lecture Notes in Computer ScienceFIMH‌ 2025 - 13th Functional Imaging and Modeling of‌ the Heart International ConferenceFunctional Imaging and Modeling‌ of the Heart. FIMH 2025LNCS-15672Dallas, United‌ StatesJune 2025HALDOI back to text‌
43 inproceedingsR.Rafael Silva, C.Caroline‌ Stehlé, G.Guillaume Pétriat, P.-H.Pierre-Henri‌ Cadet, T.Thierry Tibi and M.Maxime‌ Sermesant. Frugal AI for Automated Cardiac Defibrillation:‌ Balancing Performance & Hardware Constraints.Lecture notes in computer scienceFIMH‌ 2025 - 13th International‌ Conference on Functional Imaging‌‌ and Modeling of the HeartFunctional Imaging and‌ Modeling of the Heart‌ : part 1LNCS-15672‌‌Dallas (TX), United StatesJune 2025HAL DOI‌back to text back‌ to text
44 inproceedings‌‌T.Tom Szwagier, G.Guillaume Olikier and‌ X.Xavier Pennec.‌ Eigengap Sparsity for Covariance‌‌ Parsimony.Proc. of the 7th International Conference‌ on Geometric Science of‌ Information - GSI'25GSI'25‌‌ - 7th International Conference on Geometric Science of‌ InformationLecture Notes in‌ Computer ScienceSaint-Malo (35400),‌‌ France2025HAL back to text back to‌ text back to text‌
45 inproceedingsR.Riccardo‌‌ Taiello, C.Clémentine Gritti, M.Melek‌ Önen and M.Marco‌ Lorenzi. Buffalo: A‌‌ Practical Secure Aggregation Protocol for Buffered Asynchronous Federated‌ Learning.CODASPY 2025‌ - 15th ACM Conference‌‌ on Data and Application Security and PrivacyPittsburgh,‌ United StatesACM2025‌, 1-12HAL DOI‌‌
46 inproceedingsE.Elie Thellier, H.Huiyu‌ Li, N.Nicholas‌ Ayache and H.Hervé‌‌ Delingette. Mitigating Data Exfiltration Attacks through Layer-Wise‌ Learning Rate Decay Fine-Tuning‌.Bridging Regulatory Science‌‌ and Medical Imaging Evaluation; and Distributed, Collaborative, and‌ Federated Learning6th MICCAI‌ Workshop on “Distributed, Collaborative‌‌ and Federated Learning”Lecture Notes in Computer Science‌Bridging Regulatory Science and‌ Medical Imaging Evaluation; and‌‌ Distributed, Collaborative, and Federated Learning16135Daejeon, South‌ KoreaSeptember 2025HAL‌back to text
47‌‌ inproceedingsM.Martin van Waerebeke, M.Marco‌ Lorenzi, G.Giovanni‌ Neglia and K.Kevin‌‌ Scaman. When to Forget? Complexity Trade-offs in‌ Machine Unlearning.International‌ Conference in Machine Learning‌‌ (ICML 2025)Vancouver (BC), CanadaJuly 2025HAL‌
48 inproceedingsZ.Zihao‌ Wang, Y.Yuzhou‌‌ Chen and H.Herve Delingette. Stochastic Flow‌ Inference for Medical Image‌ Digital Twins.CHASE‌‌ '25: ACM/IEEE International Conference on Connected Health: Applications,‌ Systems and Engineering Technologies‌Yeshiva University Museum New‌‌ York NY USA, United StatesACMJune 2025‌, 353-357HAL DOI‌

Conferences without proceedings

49‌‌ inproceedingsF.Francesco Cremonesi, L.Lucie Chambon‌, N.Nelson Mokkadem‌, H. T.Huyen‌‌ Thi Trang Nguyen, O.Oliver Humbert and‌ M.Marco Lorenzi.‌ Knowledge-based semantic enrichment of‌‌ medical imaging data for automatic phenotyping and pattern‌ discovery in metastatic lung‌ cancer.2025 SNMMI‌‌ Annual Meeting - Society for Nuclear Medicine and‌ Molecular ImagingNew orleans,‌ LA, United StatesJune‌‌ 2025HAL back to text
50 inproceedingsF.‌Federica Facente, B.‌Benjamin Billot, V.‌‌Vivek Gopalakrishnan, M.Manasi Kattel, W.‌Wen Wei, P.‌Polina Golland, H.‌‌Hervé Delingette, N.Nicholas Ayache and P.‌Pierre Berthet-Rayne. Multi-stage‌ CNN for fast registration‌‌ of 3D preoperative CTs to 2D intraoperative X-rays‌.MICCAI 2025 -‌ COLlaborative Intelligence and Autonomy‌‌ in Image-guided Surgery (COLAS)Daejon, South KoreaSeptember‌ 2025HAL back to‌ text
51 inproceedingsM.‌‌Maëlis Morier and J.‌Jairo Rodríguez-Padilla. Learning Cardiac Electrophysiology with Graph‌ Neural Networks for Fast Data-driven Personalised Predictions.‌Functional Imaging and Modeling of the Heart: 13th‌ International Conference, FIMH 2025, Dallas, TX, USA, June‌ 1–5, 2025, Proceedings, Part II (Lecture Notes in‌ Computer Science, 15673)FIMH 2025 - 13th Functional‌ Imaging and Modeling of the Heart International Conference‌15672 (Part I) and 15673 (Part II)Dallas‌ (TX), United StatesJuly 2025, LNCS 15672‌ and LNCS 15673HALback to text back‌ to text

Scientific books

52 bookM.Marco‌ Lorenzi and M. A.Maria A. Zuluaga.‌ Trustworthy AI in Medical Imaging.Elsevier2025‌HAL DOI back to text

Scientific book chapters‌

53 inbookF.Francesco Cremonesi, M.Marc‌ Vesin, S.Sergen Cansiz, Y.Yannick‌ Bouillard, I.Irene Balelli, L.Lucia‌ Innocenti, R.Riccardo Taiello, S.Santiago‌ Silva, S.-S.Samy-Safwan Ayed, M.Melek‌ Önen, F.Fanny Orlhac, C.Christophe‌ Nioche, B.Bastien Houis, R.Romain‌ Modzelewski, N.Nathan Lapel, R.Renaud‌ Schiappa, O.Olivier Humbert and M.Marco‌ Lorenzi. Fed-BioMed: Open, Transparent and Trusted Federated‌ Learning for Real-world Healthcare Applications.832Federated‌ Learning SystemsStudies in Computational IntelligenceSpringer Nature‌ SwitzerlandApril 2025, 19-41HAL DOI back‌ to text

Doctoral dissertations and habilitation theses

54‌ thesisL.Lisa Guzzi. Automatic segmentation of‌ the vascular system to enhance AI-based decision support‌ system for peripheral artery disease.Université Côte‌ d'AzurDecember 2025HALback to text
55‌ thesisT.Tom Szwagier. Rethinking statistical methods‌ with flags.Université Côte d'AzurNovember 2025‌HAL back to textback to text

Reports‌ & preprints

56 miscY.Yanis Aeschlimann,‌ A.Anna Calissano, T.Théodore Papadopoulo and‌ S.Samuel Deslauriers-Gauthier. Brain network alignment using‌ structural and functional connectivity with anatomical constraints.‌January 2025HAL
57 miscO.Olivier Bisson‌ and X.Xavier Pennec. Log-Euclidean Lie Groups‌.August 2025HALback to text
58‌ miscE.Elodie Maignant, X.Xavier Pennec‌, A.Alain Trouvé and A.Anna Calissano‌. Barycentric subspace analysis of network-valued data.‌July 2025HAL
59 miscG.Guillaume Olikier‌, P.Petar Mlinarić, P.-A. -.P.‌ -A. Absil and A.André Uschmajew. The‌ tangent cone to the real determinantal variety: various‌ expressions and a proof.2025HAL
60‌ miscS.Santiago Silva, G.Ghiles Reguig‌, N. P.Neil P Oxtoby, A.‌Andre Altmann and M.Marco Lorenzi. Fed-ComBat:‌ A Generalized Federated Framework for Batch Effect Harmonization‌ in Collaborative Studies.January 2026HAL
61‌ miscT.Tom Szwagier and X.Xavier Pennec‌. Nested Subspace Learning with Flags.2025‌HAL back to text

Other scientific publications

62‌ inproceedingsR.Rafael Silva, C.Caroline Stehlé‌, M.Maxime Sermesant, G.Guillaume Pétriat, P.-H.Pierre-Henri Cadet‌ and T.Thierry Tibi‌. Frugal AI for‌‌ Automated Cardiac Defibrillation: Balancing Performance & Hardware Constraints‌.FIMH 2025 -‌ 13th International Conference on‌‌ Functional Imaging and Modeling of the HeartFunctional‌ Imaging and Modeling of‌ the HeartLNCS-15672Dallas‌‌ (TX), United StatesJune 2025HAL back to‌ text

Scientific popularization

63‌ inproceedingsC.Camilla Ferrario‌‌. Myocardial Stiffness Quantification Using Ultrasound Shear Wave‌ Elastography and Reduced Modeling‌ for Subject-Specific Simulations.‌‌Journées annuelles du PEPR Santé NumériqueLille (France),‌ FranceOctober 2025HAL‌

Patents

64 patentT.‌‌Théodore Soulier, M.Mariem Hamzaoui, O.‌Olivier Colliot, N.‌Nicholas Ayache and B.‌‌Bruno Stankoff. Devices for generation of synthetic‌ 3D representations of myelin‌ content (MyeliGAN).WO‌‌ 2025/068000 A1FranceApril 2025HAL back to‌ text back to text‌

EPIONE - 2025

EPIONE - 2025

2025​​﻿﻿Activity reportProject-TeamEPIONE​​​‌

Keywords﻿﻿﻿‌

Computer Science and Digital﻿‌​‌ Science

Other﻿​​﻿ Research Topics and Application​​​‌ Domains

1﻿‌​‌ Team members, visitors, external﻿​​﻿ collaborators

Research Scientists

Post-Doctoral Fellows

PhD﻿‌​‌ Students

Technical Staff​​​‌

Interns​‌﻿﻿ and Apprentices

Administrative​​​‌ Assistant

Visiting Scientists​‌﻿﻿

External Collaborators

2 Overall objectives​‌﻿﻿

2.1 Description

2.2 Organization﻿​​﻿

3 Research program﻿‌​‌

3.1 Introduction

3.2​​​‌ Biomedical Image Analysis &﻿﻿﻿‌ Machine Learning

3.3 Imaging and​​﻿﻿ Phenomics, Biostatistics

3.4 Computational Anatomy and﻿‌​‌ Geometric Statistics

3.5 Computational Physiology and﻿​﻿﻿ Image-Guided Therapy

3.6﻿​​﻿ Computational Cardiology and Image-Based​​​‌ Cardiac Interventions

4﻿‌​‌ Application domains

5​​​‌ Social and environmental responsibility﻿​﻿﻿

5.1 Footprint of research​‌﻿﻿ activities

6​‌﻿﻿ Highlights of the year​​﻿﻿

6.1 Awards

6.2​​﻿﻿ Promotions

6.3 Others​​﻿﻿

7﻿﻿﻿‌ Latest software developments, platforms,﻿‌​‌ open data

7.1 Latest﻿​​﻿ software developments

7.1.1 MedINRIA​​​‌

7.1.2 Music﻿‌​‌

7.1.3 geomstats​​​‌

7.1.4 Fed-BioMed

8 New results﻿﻿﻿‌

8.1 Medical Image Analysis﻿‌​‌ & Machine Learning

8.1.1﻿​​﻿ Prostate Cancer Detection and​​​‌ Characterization from multiparametric MRI﻿﻿﻿‌

8.1.2 Analysis﻿​​﻿ of European National Health​​​‌ data to study the﻿﻿﻿‌ outcomes of patients with﻿‌​‌ vascular diseases

8.1.3 Spatial regularization​​​‌ for improved accuracy and﻿​﻿﻿ interpretability in keypoint-based registration​‌﻿﻿

8.1.4 Resource-efficient Automatic Refinement​​​‌ of Segmentations via Weak﻿﻿﻿‌ Supervision from Light Feedback﻿‌​‌

8.1.5﻿‌​‌ Segmentation of Fractured Bones﻿​​﻿ from CT

8.1.6 TBDM: Temporal﻿​​﻿ Boundary Distillation Module for​​​‌ Surgical Gesture Segmentation

8.1.7 Multi-stage CNN​​﻿﻿ for fast registration of​​​‌ 3D preoperative CTs to﻿​﻿﻿ 2D intraoperative X-rays

8.1.8﻿‌​‌ Exploring Variability in Medical﻿​​﻿ Image Segmentation: New Metrics​​​‌ and Frameworks

8.1.9 AI-based precision oncology​​​‌ to monitor response of﻿﻿﻿‌ metastatic cancer to immunotherapy﻿‌​‌ using PET/CT imaging

8.1.10​‌﻿﻿ Development of predictive models​​﻿﻿ in patients with Peripheral​​​‌ Artery Disease

8.1.11 Automatic​​​‌ Segmentation of Lower-Limb Arteries﻿​﻿﻿ on CTA for Pre-surgical​‌﻿﻿ Planning of Peripheral Artery​​﻿﻿ Disease.

8.1.12 Segmentation of Abdominal​​​‌ Aortic Aneurysm from CT﻿﻿﻿‌ Angiography

8.1.13 A Scalable Spatio-Temporal﻿​​﻿ Atlas of Neurodegenerative Brain​​​‌ Changes

8.1.14﻿﻿﻿‌ MRI-TRUS prostate registration

8.1.15﻿​﻿﻿ Data Exfiltration and Data​‌﻿﻿ Anonymization

8.1.16 Lung Cancer Risk​‌﻿﻿ Prediction From a Single​​﻿﻿ Low-Dose Chest Computed Tomography​​​‌

8.1.17 Thyrosonics

8.1.18﻿​​﻿ Mitigating Data Exfiltration Attacks​​​‌ through Layer-Wise Learning Rate﻿﻿﻿‌ Decay Fine-Tuning

8.1.19 Disease Progression﻿​﻿﻿ Modeling and Stratification for​‌﻿﻿ detecting sub-trajectories in the​​﻿﻿ natural history of pathologies​​​‌

8.2 Imaging & Phenomics,​​﻿﻿ Biostatistics

8.2.1 Cardiac Electromechanical​​​‌ Model Sensitivity Analysis using﻿​﻿﻿ Causal Discovery

8.2.2 Association of mtDNA﻿​​﻿ variants and phenotype in​​​‌ mitochondrial diseases with multi-OMICS﻿﻿﻿‌ approaches

8.2.3​​﻿﻿ Deciphering Fragile X Syndrome​​​‌ via Multi-Omic Integration

8.2.4 Consensus Enhances Individual​‌﻿﻿ Causal Models: a Use​​﻿﻿ Case on Lung Cancer​​​‌

8.3﻿﻿﻿‌ Computational Anatomy & Geometric﻿‌​‌ Statistics

8.3.1 Log-Euclidean frameworks﻿​​﻿ for smooth brain connectivity​​​‌ trajectories

8.3.2﻿‌​‌ Eigengap sparsity for covariance﻿​​﻿ parsimony

8.3.3 Parsimonious​‌﻿﻿ Gaussian mixture models with​​﻿﻿ piecewise-constant eigenvalue profiles

8.3.4 Rethinking​‌﻿﻿ statistical methods with flags​​﻿﻿

8.4﻿‌​‌ Computational Cardiology & Image-Based﻿​​﻿ Cardiac Interventions

8.4.1 Cardiac​​​‌ Electrophysiology Model Personalization

8.4.2 Differentiable​​﻿﻿ Electromechanical Modeling for Patient-Specific​​​‌ Cardiac Biomechanics

8.4.3 Prediction​​﻿﻿ of stroke based on​​​‌ shape and simulation of﻿​﻿﻿ the left atrium

8.4.4 Myocardial​​​‌ Stiffness Quantification using Ultrasound﻿​﻿﻿ Shear Wave Elastography and​‌﻿﻿ Reduced Modeling

8.4.5 Multimodal Personalization﻿​​﻿ of Cardiac Electrophysiology Models​​​‌ combining 12-lead ECG and﻿﻿﻿‌ Computed Tomography

8.4.6 Learning​​﻿﻿ Cardiac Electrophysiology with Graph​​​‌ Neural Networks for Fast﻿​﻿﻿ Data-driven Personalized Predictions

8.4.7​​﻿﻿ Uncertainty-Informed Multimodal Infarct Age​​​‌ Prediction from Imaging and﻿​﻿﻿ Clinical Data

8.4.8 In silico​​​‌ Assessment of Arrhythmia Inducibility﻿﻿﻿‌ Dependence on Stimulus Location﻿‌​‌ using Calibrated MR-based Infarcted﻿​​﻿ Heart Models

8.4.9﻿​﻿﻿ Miniaturizing Automated External Defibrillation​‌﻿﻿ with Frugal AI

8.5 Multi-centric﻿​﻿﻿ data and Federated Learning​‌﻿﻿

2025Activity reportProject-TeamEPIONE‌

Keywords‌

Computer Science and Digital‌‌ Science

Other Research Topics and Application‌ Domains

1‌‌ Team members, visitors, external collaborators

PhD‌‌ Students

Technical Staff‌

Interns‌ and Apprentices

Administrative‌ Assistant

Visiting Scientists‌

2 Overall objectives‌

2.2 Organization

3 Research program‌‌

3.2‌ Biomedical Image Analysis &‌ Machine Learning

3.3 Imaging and Phenomics, Biostatistics

3.4 Computational Anatomy and‌‌ Geometric Statistics

3.5 Computational Physiology and Image-Guided Therapy

3.6 Computational Cardiology and Image-Based‌ Cardiac Interventions

4‌‌ Application domains

5‌ Social and environmental responsibility

5.1 Footprint of research‌ activities

6‌ Highlights of the year

6.2 Promotions

6.3 Others

7‌ Latest software developments, platforms,‌‌ open data

7.1 Latest software developments

7.1.1 MedINRIA‌

7.1.2 Music‌‌

7.1.3 geomstats‌

8 New results‌

8.1 Medical Image Analysis‌‌ & Machine Learning

8.1.1 Prostate Cancer Detection and‌ Characterization from multiparametric MRI‌

8.1.2 Analysis of European National Health‌ data to study the‌ outcomes of patients with‌‌ vascular diseases

8.1.3 Spatial regularization‌ for improved accuracy and interpretability in keypoint-based registration‌

8.1.4 Resource-efficient Automatic Refinement‌ of Segmentations via Weak‌ Supervision from Light Feedback‌‌

8.1.5‌‌ Segmentation of Fractured Bones from CT

8.1.6 TBDM: Temporal Boundary Distillation Module for‌ Surgical Gesture Segmentation

8.1.7 Multi-stage CNN for fast registration of‌ 3D preoperative CTs to 2D intraoperative X-rays

8.1.8‌‌ Exploring Variability in Medical Image Segmentation: New Metrics‌ and Frameworks

8.1.9 AI-based precision oncology‌ to monitor response of‌ metastatic cancer to immunotherapy‌‌ using PET/CT imaging

8.1.10‌ Development of predictive models in patients with Peripheral‌ Artery Disease

8.1.11 Automatic‌ Segmentation of Lower-Limb Arteries on CTA for Pre-surgical‌ Planning of Peripheral Artery Disease.

8.1.12 Segmentation of Abdominal‌ Aortic Aneurysm from CT‌ Angiography

8.1.13 A Scalable Spatio-Temporal Atlas of Neurodegenerative Brain‌ Changes

8.1.14‌ MRI-TRUS prostate registration

8.1.15 Data Exfiltration and Data‌ Anonymization

8.1.16 Lung Cancer Risk‌ Prediction From a Single Low-Dose Chest Computed Tomography‌

8.1.18 Mitigating Data Exfiltration Attacks‌ through Layer-Wise Learning Rate‌ Decay Fine-Tuning

8.1.19 Disease Progression Modeling and Stratification for‌ detecting sub-trajectories in the natural history of pathologies‌

8.2 Imaging & Phenomics, Biostatistics

8.2.1 Cardiac Electromechanical‌ Model Sensitivity Analysis using Causal Discovery

8.2.2 Association of mtDNA variants and phenotype in‌ mitochondrial diseases with multi-OMICS‌ approaches

8.2.3 Deciphering Fragile X Syndrome‌ via Multi-Omic Integration

8.2.4 Consensus Enhances Individual‌ Causal Models: a Use Case on Lung Cancer‌

8.3‌ Computational Anatomy & Geometric‌‌ Statistics

8.3.1 Log-Euclidean frameworks for smooth brain connectivity‌ trajectories

8.3.2‌‌ Eigengap sparsity for covariance parsimony

8.3.3 Parsimonious‌ Gaussian mixture models with piecewise-constant eigenvalue profiles

8.3.4 Rethinking‌ statistical methods with flags

8.4‌‌ Computational Cardiology & Image-Based Cardiac Interventions

8.4.1 Cardiac‌ Electrophysiology Model Personalization

8.4.2 Differentiable Electromechanical Modeling for Patient-Specific‌ Cardiac Biomechanics

8.4.3 Prediction of stroke based on‌ shape and simulation of the left atrium

8.4.4 Myocardial‌ Stiffness Quantification using Ultrasound Shear Wave Elastography and‌ Reduced Modeling

8.4.5 Multimodal Personalization of Cardiac Electrophysiology Models‌ combining 12-lead ECG and‌ Computed Tomography

8.4.6 Learning Cardiac Electrophysiology with Graph‌ Neural Networks for Fast Data-driven Personalized Predictions

8.4.7 Uncertainty-Informed Multimodal Infarct Age‌ Prediction from Imaging and Clinical Data

8.4.8 In silico‌ Assessment of Arrhythmia Inducibility‌ Dependence on Stimulus Location‌‌ using Calibrated MR-based Infarcted Heart Models

8.4.9 Miniaturizing Automated External Defibrillation‌ with Frugal AI

8.5 Multi-centric data and Federated Learning‌

8.5.1 Development of a 18FDG-PET normative uptake atlas‌ and its clinical application for abnormal metabolic activity‌ detection

8.5.2 Real-world Deployment‌ of Federated Learning in‌‌ Biomedical Research Consortia with Fed-BioMed

8.5.3 Federated domain adaptation for brain‌ vessel segmentation

8.5.4 Knowledge‌ Guided Medical Report Generation for Pathology Specific Findings‌

8.5.5 Fed-ComBat: A Generalized Federated Framework‌ for Batch Effect Harmonization in Collaborative Studies

9 Bilateral‌‌ contracts and grants with industry

9.1 Bilateral contracts‌ with industry

9.1.1 Spin-off‌ company inHEART

9.1.2 Start-up Inn'Pulse‌

9.1.3‌ Koelis