2025Activity report​​​‌Project-TeamMIMESIS

3.1.1 Immersed​ boundary methods

Since several​‌ years, we investigate in​​ the team finite element​​​‌ methods that fall under​ the class of unfitted​‌ (also known as immersed​​ boundary) methods. Because such​​​‌ methods do not require​ a discretization that strictly​‌ conforms to the domain​​ boundary, they are particularly​​​‌ suited for the development​ of digital twins, as​‌ they facilitate the automatic​​ generation of patient-specific simulations​​​‌ on complex geometries. We​ particularly focus on the​‌ development, the numerical analysis​​ and mathematical foundations for​​​‌ a level-set based method​ that we have called​‌ $\phi$-FEM. The main​​ advantage of $\phi$-FEM​​​‌ is that it uses​ the classical Finite Element​‌ tools on unfitted meshes.​​

Our objective is to​​​‌ develop methods of this​ type for shape optimization​‌ and for the simulation​​ of physical phenomena in​​​‌ evolving domains. The $\mathrm{\phi }$-FEM framework is particularly​‌ well suited to these​​ settings. Indeed, since the​​ computational domain may change​​​‌ either during the optimization‌ iterations or as a‌​‌ function of time, the​​ use of the level-set​​​‌ function $\phi$ allows the‌ geometry to be represented‌​‌ implicitly, thereby avoiding costly​​ and potentially unstable remeshing​​​‌ at each iteration or‌ time step.

3.1.2 Scientific‌​‌ machine learning

Speeding up​​ the simulation of non-linear​​​‌ elastic solids is a‌ field studied by many‌​‌ reasearchers. A variety of​​ solutions has been proposed,​​​‌ including domain decomposition, GPU‌ computing, or dimensionality reduction,‌​‌ to cite just a​​ few. We investigate approaches​​​‌ based on deep neural‌ networks, both in supervised‌​‌ and unsupervised learning scenarios,​​ for which the choice​​​‌ of input/output couples, network‌ architecture and physical laws‌​‌ are critical for accuracy​​ and speed.

This line​​​‌ of work has the‌ dual objective of avoiding‌​‌ black box approaches (by​​ keeping the underlying physics​​​‌ law or known numerical‌ schemes in the prediction)‌​‌ and generalizing as much​​ as possible the learned​​​‌ solution to avoid repeated‌ training costs when changing‌​‌ the problem. We currently​​ focus on enabling dynamic​​​‌ simulations, as well as‌ developing deep neural network‌​‌ architectures that provide more​​ genericity in the solution​​​‌ space (by being somewhat‌ invariant to the meshing‌​‌ of the domain for​​ instance).

3.1.3 Non-smooth mechanics:​​​‌ contacts and multiphysic

Many‌ clinical interventions involve complex‌​‌ mechanical interactions between medical​​ devices and anatomical structures,​​​‌ with surgery and interventional‌ radiology being primary examples.‌​‌ Accurately simulating these interactions,​​ particularly contacts with friction,​​​‌ is challenging due to‌ numerical instability, slow computation,‌​‌ and potential inaccuracies. Our​​ research focuses on modeling​​​‌ these contact phenomena and‌ improving the robustness and‌​‌ efficiency of simulations, which​​ is especially important for​​​‌ applications such as realistic‌ haptic feedback in medical‌​‌ tool manipulation.

The team’s​​ objectives are to enhance​​​‌ computational performance and stability‌ in contact problems through‌​‌ innovative approaches, including hybrid​​ numerical solvers and asynchronous​​​‌ methods. By developing these‌ techniques, we aim to‌​‌ provide reliable, high-frequency force​​ updates and precise simulations​​​‌ that can support interactive‌ systems, surgical planning, and‌​‌ the development of advanced​​ haptics interfaces for medical​​​‌ interventions.

3.2.1‌ Optimal control

A central‌​‌ theme of our research​​ is optimal control in​​​‌ computational modeling, with applications‌ ranging from patient-specific simulations‌​‌ to interactive and real-time​​ systems. In particular, we​​​‌ focus on patient-specific computational‌ models and differentiable simulation,‌​‌ where the goal is​​ to determine model parameters​​​‌ or control inputs that‌ optimize performance or match‌​‌ observed data, all while​​ respecting the underlying physical​​​‌ laws described by partial‌ differential equations.

Our objectives‌​‌ are to develop and​​ apply advanced numerical methods​​​‌ for these optimal control‌ problems. For patient-specific models,‌​‌ this involves identifying parameters​​​‌ from clinical or intraoperative​ observations using cost functionals​‌ that quantify the mismatch​​ between model outputs and​​​‌ data. For differentiable simulations,​ we leverage automatic differentiation​‌ to efficiently compute gradients​​ without deriving problem-specific adjoint​​​‌ equations. In both cases,​ we emphasize robust and​‌ scalable methods, including adjoint-based​​ techniques, regularization strategies, and​​​‌ stable discretization schemes, to​ enable applications such as​‌ surgical planning, biomechanics, and​​ near real-time optimization in​​​‌ complex PDE-based systems.

3.2.2​ Differentiable simulation

The MIMESIS​‌ team investigates advanced methodologies​​ in computational modeling, focusing​​​‌ in particular on the​ paradigm of differentiable simulation.​‌ This approach aims to​​ make the entire computational​​​‌ pipeline—from numerical discretization to​ physical simulation—explicitly differentiable with​‌ respect to model parameters,​​ extending beyond classical optimal​​​‌ control techniques.

Our objectives​ are to leverage this​‌ framework to compute accurate​​ and efficient gradients using​​​‌ automatic differentiation, without relying​ on problem-specific adjoint equations.​‌ This enables seamless integration​​ with modern optimization and​​​‌ learning-based methods, supporting tasks​ such as parameter identification,​‌ uncertainty quantification, and near​​ real-time optimization. We focus​​​‌ on carefully balancing numerical​ accuracy, computational cost, and​‌ gradient stability in complex​​ PDE-based simulations to deliver​​​‌ practical and reliable tools​ for interactive and optimization-driven​‌ applications.

3.2.3 Control in​​ medical robotics

The MIMESIS​​​‌ team is actively involved​ in the control of​‌ medical robotic systems, with​​ a particular focus on​​​‌ needle-based interventions and endovascular​ procedures. Current surgical robots​‌ are mostly teleoperated, with​​ little autonomy, and their​​​‌ performance is strongly affected​ by the complex interactions​‌ between deformable instruments and​​ soft biological tissues. To​​​‌ address this, we have​ developed numerical methods for​‌ real-time inverse simulation and​​ control synthesis, ranging from​​​‌ finite element–based optimization to​ deep reinforcement learning approaches.​‌ These methods allow us​​ to compensate for tissue​​​‌ deformations and improve robot​ accuracy in controlled experimental​‌ settings.

Our objectives are​​ to extend these approaches​​​‌ towards robust, clinically viable​ autonomous control. This includes​‌ handling limited intraoperative imaging,​​ large tissue deformations, and​​​‌ uncertainty in anatomical models.​ We aim to design​‌ control strategies that maintain​​ stability and reliability in​​​‌ realistic clinical conditions, combining​ advanced numerical simulations, optimization​‌ techniques, and learning-based methods​​ to enable fast, adaptive,​​​‌ and safe robotic interventions.​

5.1.1 SimRender

• Name:​
  Simulation Rendering in 3D​‌
• Keywords:
  Modelization and numerical​​ simulations, Data visualization, 3D​​​‌
• Scientific Description:

  SimRender is​ a Python package for​‌ creating interactive 3D rendering​​ of numerical simulations in​​​‌ a very few lines​ of code. The main​‌ feature is that users​​ can launch an interactive​​​‌ 3D rendering window without​ blocking the execution of​‌ the python process running​​ a numerical simulation. This​​​‌ is very useful for​ better understanding how optimization​‌ processes evolve during deep​​ learning and deep reinforcement​​​‌ learning algorithms and to​ better detect any errors.​‌ The project is compatible​​ with any numerical simulation​​​‌ written in Python and​ provides dedicated useful features​‌ for SOFA numerical simulations.​​

  SimRender provides a simple​​​‌ API for creating and​ updating 3D objects from​‌ simulated data and displaying​​ them with very few​​​‌ lines of code. A​ viewer can be launched​‌ automatically in a python​​ sub-process so as not​​​‌ to block the execution​ of the main python​‌ process. A fast method​​ for sharing 3D data​​​‌ between python sub-processes has​ been implemented using Shared​‌ Nympy Arrays. The viewer​​ is written using Vedo​​​‌ to create various customizable​ 3D objects (meshes, point​‌ clouds, vectors). Using SimRender​​ with any SOFA scene​​​‌ is made even easier​ by the automatic rendering​‌ of several components and​​ the automatic update of​​​‌ data during time steps.​

• Functional Description:
  SimRender provides​‌ a simple API for​​ creating and updating 3D​​​‌ objects from simulated data​ and displaying them with​‌ very few lines of​​ code. A viewer can​​​‌ be launched automatically in​ a Python sub-process so​‌ as not to block​​ the execution of the​​​‌ main python process. A​ fast method for sharing​‌ 3D data between Python​​ sub-processes has been implemented​​​‌ using Shared Numpy Arrays.​ The viewer is written​‌ using Vedo to create​​ various customizable 3D objects​​​‌ (meshes, point clouds, vectors).​ Using SimRender with any​‌ SOFA scene is made​​ even easier by the​​​‌ automatic rendering of several​ components and the automatic​‌ update of data during​​ time steps.
• URL:
  https://­github.­com/­RobinEnjalbert/­SimRender​​​‌
• Contact:
  Robin Enjalbert

5.1.2​ SOFA

• Name:
  Simulation Open​‌ Framework Architecture
• Keywords:
  Real​​ time, Multi-physics simulation, Medical​​​‌ applications
• Functional Description:
  SOFA​ is an Open Source​‌ framework primarily targeted at​​ real-time simulation, with an​​​‌ emphasis on medical simulation.​ It is mostly intended​‌ for the research community​​ to help develop new​​​‌ algorithms, but can also​ be used as an​‌ efficient prototyping tool. Based​​ on an advanced software​​​‌ architecture, it allows the​ creation of complex and​‌ evolving simulations by combining​​ new algorithms with algorithms​​ already included in SOFA,​​​‌ the modification of most‌ parameters of the simulation‌​‌ (deformable behavior, surface representation,​​ solver, constraints, collision algorithm​​​‌ etc.) by simply editing‌ an XML file, the‌​‌ building of complex models​​ from simpler ones using​​​‌ a scene-graph description, the‌ efficient simulation of the‌​‌ dynamics of interacting objects​​ using abstract equation solvers,​​​‌ the reuse and easy‌ comparison of a variety‌​‌ of available methods.
• News​​ of the Year:
  The​​​‌ new version v20.06 has‌ been released including new‌​‌ elements on SoftRobots +​​ ModelOrderReduction integration, in addition​​​‌ to an improved architecture‌ and lots of cleans‌​‌ and bugfixes.
• URL:
  http://­www.­sofa-framework.­org​​
• Publication:
  hal-00681539
• Contact:
  Hugo​​​‌ Talbot
• Participants:
  Christian Duriez,‌ François Faure, Hervé Delingette,‌​‌ Stephane Cotin, Hugo Talbot,​​ Maud Marchal
• Partners:
  IGG,​​​‌ CRIStAL

5.1.3 PhiFEM

• Name:‌
  Finite Element Method On‌​‌ Unfitted Mesh Using Level-set​​ description of the Geometry​​​‌
• Keywords:
  Finite element modelling,‌ Numerical analysis, Biomechanics
• Scientific‌​‌ Description:
  The phi-Finite Element​​ Method (phiFEM) is an​​​‌ unfitted finite element methodology‌ for the numerical approximation‌​‌ of partial differential equations​​ posed on domains implicitly​​​‌ defined by level-set functions.‌ The method is built‌​‌ on a background finite​​ element mesh independent of​​​‌ the physical geometry, on‌ which standard finite element‌​‌ spaces are defined and​​ suitably extended to the​​​‌ computational domain. The geometry‌ is incorporated into the‌​‌ variational formulation through the​​ level-set representation, allowing the​​​‌ accurate treatment of curved‌ boundaries without mesh conformity.‌​‌ Boundary conditions are enforced​​ weakly using consistent Nitsche-type​​​‌ formulations combined with stabilization‌ terms that ensure coercivity,‌​‌ optimal convergence rates, and​​ robustness with respect to​​​‌ small cut elements. PhiFEM‌ admits high-order accuracy and‌​‌ is supported by rigorous​​ stability and error analyses,​​​‌ making it well suited‌ for problems involving complex,‌​‌ evolving, or parameter-dependent geometries.​​
• Functional Description:
  The phi-Finite​​​‌ Element Method (phiFEM) provides‌ a computational framework for‌​‌ solving partial differential equations​​ on geometries that are​​​‌ implicitly defined and may‌ vary from one simulation‌​‌ to another. The method​​ takes as input a​​​‌ background mesh and a‌ level-set function describing the‌​‌ physical domain, and automatically​​ constructs the discrete problem​​​‌ without requiring geometry-fitted meshing.‌ It enables the user‌​‌ to define governing equations,​​ material parameters, and boundary​​​‌ conditions independently of the‌ underlying mesh, while ensuring‌​‌ consistent enforcement of interface​​ and boundary effects through​​​‌ stabilized variational formulations. The‌ approach supports high-order discretizations,‌​‌ handles complex topologies in​​ a robust manner, and​​​‌ allows rapid geometry updates‌ with minimal preprocessing. As‌​‌ a result, phiFEM facilitates​​ efficient parametric studies, optimization​​​‌ loops, and patient-specific simulations‌ where geometry plays a‌​‌ central role.
• URL:
  http://­github.­com/­phifem​​
• Publications:
  hal-04473794, hal-02521111​​​‌, hal-03372733, hal-02521042‌, hal-03588715, hal-03685445‌​‌, hal-04731164
• Contact:
  Michel​​ Duprez
• Participants:
  Alexei Lozinski,​​​‌ Vanessa Lleras, Killian Vuillemot,‌ Raphael Bulle, Stephane Cotin,‌​‌ Michel Duprez

6.1.1 Digital twins​‌ of organs

Participants: Michel​​ Duprez, Frederique Lecourtier​​​‌, Stéphane Cotin.​

Physics-based patient-specific biomechanical models,​‌ particularly those using FEM,​​ simulate organ behaviors accurately​​​‌ but are computationally intensive,​ especially for hyper-elastic tissues.​‌ To address this, we​​ introduced U-Mesh, a data-driven​​​‌ approach using a U-Net​ architecture, achieving real-time inference​‌ but reliant on precise​​ stiffness knowledge at training.​​​‌ In 51, 52​, we introduced HyperU-Mesh,​‌ an extension that integrates​​ a Hypernetwork to condition​​​‌ the U-Mesh based on​ stiffness prior distributions. By​‌ training with FEM-simulated data​​ that varies stiffness under​​​‌ a predefined distribution, HyperU-Mesh​ ensures accuracy across variable​‌ stiffness without retraining. Experimental​​ results highlight its effectiveness​​​‌ across different scenarios, showing​ comparable accuracy to FEM​‌ while significantly improving speed.​​ In 53 we also​​​‌ proposed a method to​ predict the deformation of​‌ a liver model using​​ the centerlines of its​​​‌ vascular system as input​ of a deep neural​‌ network. Information about the​​ vascular network geometry can​​​‌ be obtained, intraoperatively, using​ ultrasound imaging. The work​‌ published in 49,​​ 37 shows how we​​​‌ can differentiate the main​ branches of this vascular​‌ tree. More recently, in​​ 35, we introduces​​​‌ Deform Any Liver (DAL),​ a realtime surrogate model​‌ designed to accurately predict​​ any liver deformation, given​​​‌ a set of applied​ forces.

Moreover, our current​‌ results have focused on​​ the development, the numerical​​​‌ analysis and mathematical foundations​ of a new method​‌ called $\Phi$-FEM (see​​ 50). The main​​​‌ advantage of our method​ is that it uses​‌ the classical FE tools​​ on unfitted meshes. We​​​‌ have also highlighted that​ the method significantly improves​‌ the convergence when compared​​ to a similar, fitted,​​​‌ discretization of the domain.​ Recently, we have also​‌ shown how to use​​ $\Phi$-FEM simulations to​​​‌ train FNO neural networks​ (see 25, 42​‌) and developed a​​ finite-difference scheme inspired by​​​‌ the $\Phi$-FEM paradigm​ (see 24, 42​‌). Both approaches significantly​​ reduce computation time. We​​​‌ have also develop a​ python library 48 to​‌ solve PDE with FEniCS.​​ Our current activity on​​​‌ $\Phi$-FEM consists in​ an Open Source implementation​‌ in a SOFA plugin.​​ This is done in​​​‌ relationship with another development​ activity that integrates automatic​‌ differentiation tools from the​​ project FEniCS (www.fenicsproject.org) to​​​‌ quickly and efficiently add​ new constitutive models in​‌ our code base, through​​ simple Python scripting. We​​​‌ are convinced of the​ impact of this work​‌ for both our research​​ activity and the field​​​‌ in general.

In the​ $\Phi$-FEM framework, the​‌ finite element space is​​ enriched using the level-set​​​‌ function that describes the​ boundary of the computational​‌ domain. Building on this​​ idea, we proposed in​​​‌ 43 a novel approach​ in which the finite​‌ element space is enriched​​ by the prediction of​​​‌ a physics-informed neural network​ (PINN). This enrichment can​‌ be incorporated into the​​ variational formulation either additively​​ or multiplicatively. On the​​​‌ one hand, this hybrid‌ strategy improves and certifies‌​‌ the predictions of neural​​ networks, yielding fast and​​​‌ accurate numerical solutions. On‌ the other hand, we‌​‌ establish a priori error​​ estimates showing that the​​​‌ proposed enriched methods outperform‌ classical finite element approaches‌​‌ by a factor that​​ depends solely on the​​​‌ quality of the prior‌ provided by the PINN.‌​‌ We validate our approach​​ through numerical experiments on​​​‌ parametric problems in one-,‌ two-, and three-dimensional geometries.‌​‌ These results demonstrate that,​​ for a prescribed accuracy,​​​‌ significantly coarser meshes can‌ be employed compared to‌​‌ standard FEM, leading to​​ substantial reductions in computational​​​‌ cost, especially in parametric‌ settings.

6.1.2 A posteriori‌​‌ error estimation in numerical​​ biomecanics

Participants: Michel Duprez​​​‌, Raphaël Bulle.‌

The Finite Element Method‌​‌ (FEM) is a well-established​​ approach for computing approximate​​​‌ solutions to deterministic engineering‌ problems governed by partial‌​‌ differential equations. FEM produces​​ discrete approximations of the​​​‌ solution, whose accuracy is‌ affected by discretization errors‌​‌ that can be rigorously​​ quantified using a posteriori​​​‌ error estimates. However, the‌ practical relevance of such‌​‌ error estimates in biomechanics,​​ and in particular for​​​‌ soft tissues whose mechanical‌ response is governed by‌​‌ large strains and strong​​ nonlinearities, has been only​​​‌ marginally addressed. Yet, reliable‌ error quantification is crucial‌​‌ for the certification and​​ trustworthiness of numerical simulations​​​‌ in biomechanical and biomedical‌ applications. Within the team,‌​‌ we first investigated goal-oriented​​ a posteriori error estimators​​​‌ based on the Dual‌ Weighted Residual (DWR) technique.‌​‌ These estimators allow one​​ to specifically control and​​​‌ minimize the error on‌ a user-defined quantity of‌​‌ interest or on a​​ targeted region of the​​​‌ computational domain, which is‌ particularly relevant for patient-specific‌​‌ simulations. In 21,​​ 22, we validated​​​‌ this methodology against experimental‌ measurements obtained from silicone‌​‌ samples and demonstrated its​​ applicability to realistic, patient-specific​​​‌ computations of pressure ulcers‌ on a human heel.‌​‌ In addition, we compared​​ the magnitude of the​​​‌ discretization error with the‌ modeling error induced by‌​‌ constitutive law choices, highlighting​​ the respective impact of​​​‌ numerical and modeling uncertainties.‌ Beyond classical FEM, we‌​‌ also developed and validated​​ in 44 a residual-based​​​‌ a posteriori error estimator‌ for a $\phi$-FEM‌​‌ scheme, representing, to our​​ knowledge, the first a​​​‌ posteriori error estimator specifically‌ designed for this class‌​‌ of immersed boundary methods.​​ Finally, in 45,​​​‌ we proposed a multimesh‌ adaptive refinement strategy for‌​‌ the numerical approximation of​​ fractional Laplacian equations, leading​​​‌ to significant improvements in‌ accuracy and computational efficiency.‌​‌

6.1.3 Recovering surgically-induced deformations​​ of the liver

Participants:​​​‌ François Lecomte, Juan‌ Verde, Stéphane Cotin‌​‌.

In 29,​​ we propose a method​​​‌ for estimating, in real‌ time, a 3D displacement‌​‌ field from a single​​ fluoroscopic image. Our approach​​​‌ uses a fully convolutional‌ network architecture to solve‌​‌ the associated inverse problem.​​ Supervised learning is performed​​​‌ on synthetic data, using‌ Digitally Reconstructed Radiographs as‌​‌ input and displacement fields​​ as output. We use​​​‌ randomized Gaussian kernels to‌ produce a synthetic training‌​‌ dataset with displacement fields​​​‌ that are smooth and​ diffeomorphic. In contrast to​‌ other 2D-3D registration methods,​​ our novel data generation​​​‌ approach does not rely​ on a statistical motion​‌ model. This enables our​​ model to accurately predict​​​‌ deformations unrelated to breathing​ or other predetermined motion​‌ patterns. As an example​​ of clinical application, we​​​‌ show that our model​ is able to predict​‌ deformations related to percutaneous​​ needle insertions accurately, potentially​​​‌ removing the need for​ contrast agent injection.

6.1.4​‌ Dynamic cutting simulation using​​ elastic snapping for mesh​​​‌ quality optimization

Participants: Hadrien​ Courtecuisse.

The team​‌ aims to develop a​​ novel cutting method based​​​‌ on a vertex-snapping strategy​ that fits the boundary​‌ surface to the cutting​​ path while avoiding the​​​‌ creation of new mesh​ elements. The cutting path​‌ is generated from a​​ point cloud using polynomial​​​‌ fitting, enabling the handling​ of unscheduled cuts and​‌ potential perturbations during the​​ procedure. To support progressive​​​‌ cutting, we develop efficient​ geometric operations capable of​‌ managing the associated topological​​ changes in real time.​​​‌ A key challenge lies​ in simultaneously preserving mesh​‌ quality and accurately aligning​​ the cut surface with​​​‌ the prescribed cutting path.​ In 34, we​‌ address this issue by​​ introducing an innovative approach​​​‌ that reformulates the geometric​ problem as a quasi-static​‌ elastic problem. This is​​ achieved by solving a​​​‌ constrained elastic system within​ an auxiliary simulation, allowing​‌ the mesh to naturally​​ optimize its quality as​​​‌ the system reaches equilibrium.​ In addition, we propose​‌ adaptations of a GPU-based​​ matrix-free solver, enabling efficient​​​‌ updates of precomputed data​ stored in GPU memory​‌ and thereby ensuring real-time​​ performance.

6.1.5 A computational​​​‌ model of altered neuronal​ activity in altered gravity​‌

Participants: Camille Gontier.​​

Electrophysiological experiments have shown​​​‌ that neuronal activity changes​ upon exposure to altered​‌ gravity. More specifically, neurons'​​ firing rates increase during​​​‌ microgravity and decrease during​ centrifugal-induced hypergravity. Different biophysical​‌ explanations have been proposed​​ for this phenomenon: however,​​​‌ they have not been​ backed by quantitative analyses​‌ nor simulations. More generally,​​ classical computational models of​​​‌ neurons and networks do​ not account for the​‌ effect of altered gravity,​​ which limits the possibility​​​‌ to perform in-silico experiments​ and simulations. In a​‌ preprint 47, we​​ propose computational implementations for​​​‌ different effects of altered​ gravity on cellular functions,​‌ and modify existing models​​ to account for the​​​‌ effect of micro- and​ hyper-gravity. Firstly, in line​‌ with previous experiments, we​​ suggest that microgravity could​​​‌ be modeled as an​ increase of the voltage-dependent​‌ channel transition rates, which​​ is assumed to be​​​‌ the result of a​ higher membrane fluidity and​‌ can be readily implemented​​ into the Hodgkin-Huxley model.​​​‌ Using in-silico simulations of​ single neurons, we show​‌ that this model of​​ the influence of gravity​​​‌ on neuronal activity allows​ to reproduce the observed​‌ increased firing and burst​​ rates. Secondly, we explore​​​‌ the role of mechano-gated​ (MG) ion channels on​‌ population activity. We show​​ that recordings can be​​​‌ fitted by a network​ of connected excitatory neurons,​‌ whose activity is balanced​​ by firing rate adaptation.​​ Adding a small depolarizing​​​‌ current to account for‌ the activation of MG‌​‌ channels also reproduces the​​ observed increased firing and​​​‌ burst rates. Overall, our‌ results fill an important‌​‌ gap in the literature,​​ by providing a computational​​​‌ link between altered gravity‌ and neuronal activity. Starting‌​‌ from historical observations of​​ the effects of gravity​​​‌ on cellular functions, we‌ derived gravity-sensitive models of‌​‌ neurons and networks, whose​​ predictions could be refined​​​‌ using future experiments.

6.2.1 Influence of neural‌ network heterogeneity on neurostimulation‌​‌ impact

Participants: Axel Hutt​​.

Brain stimulation is​​​‌ a modern therapy in‌ clinical practice. The various‌​‌ types of stimulation affect​​ the brain's internal structure​​​‌ and functioning, which results‌ to learning processes and,‌​‌ in case of mental​​ disorders, to improving the​​​‌ patient's health condition. To‌ better understand the conditions‌​‌ under which neurostimulation may​​ modulate the brain dynamics,​​​‌ we have studied the‌ impact of heterogeneity in‌​‌ neural systems on the​​ stimulation. In fact, heterogenous​​​‌ neuron morphology in neural‌ systems exhibiting spiking neural‌​‌ activity strongly affects the​​ system dynamics heavily 33​​​‌. Since neural network‌ heterogeneity may develop over‌​‌ time (on a time​​ scale of days or​​​‌ weeks) and may be‌ different in each brain‌​‌ area, this neural system​​ diversity affects the impact​​​‌ of neurostimulation.

6.2.2 Auditory‌ beat stimulation in humans‌​‌ affects subjects' sustained attention​​

Participants: Axel Hutt,​​​‌ Camille Gontier.

Recent‌ research on binaural beat‌​‌ stimulation has raised the​​ question whether it can​​​‌ improve sustained attention. Neurotypicals‌ and subjects with attention‌​‌ deficits of single gender​​ performed a visual attention​​​‌ task under auditory noise,‌ monoaural and binaural beat‌​‌ stimulation, while recording electroencephalographic​​ activity (EEG). We found​​​‌ 20 that subjects with‌ attention deficits perform with‌​‌ longer reaction times than​​ neurotypical subjects. To explore​​​‌ EEG activity, two periods‌ of interest were distinguished:‌​‌ before a correct detection​​ and before a miss,​​​‌ supposed to reflect respectively‌ moments of engagement versus‌​‌ disengagement of attention. Under​​​‌ noise stimulation, neurotypicals have​ larger frontal ERP-components P300​‌ and α-spectral power and​​ lower parietal θ/β spectral​​​‌ power ratio in correct​ trials than in missed​‌ trials, whereas subjects with​​ attention deficits show the​​​‌ inverse relation. Moreover, neurotypicals​ exhibit a negative relation​‌ of frontal δ-power and​​ θ/β ratio in a​​​‌ time window of 6​ s before targets, whereas​‌ subjects with attention deficits​​ show positively related δ-​​​‌ and α-power in this​ time window. Binaural beats​‌ diversify these results. Neurotypical​​ subjects respond with a​​​‌ longer reac-tion time compared​ to noise stimulation, while​‌ attention-deficit subjects respond equally.​​ Moreover, frontal P300 and​​​‌ α-power and parietal θ/β​ ratio resemble corresponding results​‌ under noise stimulation, whereas​​ brain activity in subjects​​​‌ with attention deficits is​ rather heterogeneous. In addition,​‌ in attention-deficit subjects frontal​​ and parietal δ- and​​​‌ α-power are positively related​ in a 6 s​‌ time window before targets.​​ In sum, under noise​​​‌ stimulation we found behavioral​ and electrophysiological biomarkers, which​‌ were inverse in neurotypicals​​ and subjects with attention​​​‌ deficits. Binaural beats break​ up these relations in​‌ both subject groups and​​ they have not been​​​‌ found to be beneficial,​ neither in behavior nor​‌ in electrophysiological biomarkers.

6.3.1 Optimization in latent​ space for real-time intraoperative​‌ characterization of digital twins​​

Participants: Stéphane Cotin.​​​‌

Physics-based Digital Twins, particularly​ those using the finite​‌ element method to solve​​ the underlying partial differential​​​‌ equation, accurately simulate organ​ behaviors but are computationally​‌ intensive, especially for hyper-elastic​​ tissues. Recently, approaches have​​​‌ leveraged neural-network-based surrogate models​ to accelerate computation time.​‌ However, these models are​​ limited by the accurate​​​‌ knowledge of patient-specific characteristics,​ such as material properties​‌ and boundary conditions, at​​ training time. This paper​​​‌ introduces a novel methodology​ for patient-specific characteristics estimation​‌ from live observations during​​ medical interventions. To retain​​​‌ the benefits of neural​ network-based surrogate models, we​‌ propose in 46 a​​ hypernetwork architecture that conditions​​​‌ the surrogate models on​ patient-specific characteristics, thus maintaining​‌ accuracy over a predefined​​ distribution of these characteristics.​​​‌ Using the trained network,​ we perform a gradient-based​‌ optimization process to determine​​ the patient characteristics given​​​‌ an intraoperative observation. We​ demonstrate the flexibility and​‌ efficiency of our approach​​ through experiments with varying​​​‌ geometries, complex physics laws,​ and various patient characteristics.​‌

6.3.2 Neural controllers for​​ autonomous medical robots

Participants:​​​‌ Stéphane Cotin, François​ Lecomte, Michel Duprez​‌, Ahmed Moustafa.​​

The primary therapeutic solution​​​‌ for cardiovascular diseases is​ endovascular interventions, thanks to​‌ their minimal invasiveness and​​ low costs. However, these​​​‌ procedures are limited by​ their complexity and by​‌ the need of acquiring​​ fluoroscopic images to visualize​​​‌ the internal structures of​ the patients. The acquisition​‌ of these images requires​​ using X-rays, which are​​​‌ dangerous for the health​ of both the patient​‌ and the clinician. Furthermore,​​ to visualize the vessel​​​‌ structures, it is necessary​ to inject a contrast​‌ agent, which is harmful​​ for the patient's kidneys.​​​‌ To address these limitations,​ the only partial solution​‌ that exists today is​​ the use of endovascular​​ robots, which allow the​​​‌ caregiver to perform the‌ intervention far away from‌​‌ the operative field. However,​​ these robots are only​​​‌ master-follower devices, which are‌ not able to provide‌​‌ additional support to the​​ clinician. To address this​​​‌ limitation, in 55,‌ we have proposed a‌​‌ zero-shot learning strategy for​​ three-dimensional autonomous endovascular navigation.​​​‌ Using a very small‌ training set of branching‌​‌ patterns, our reinforcement learning​​ algorithm is able to​​​‌ learn a control that‌ can then be applied‌​‌ to unseen vascular anatomies​​ without retraining, even when​​​‌ the anatomy is moving.‌ To retrieve the movement‌​‌ of the anatomy, starting​​ from fluoroscopic images, we​​​‌ proposed a method to‌ estimate the motion of‌​‌ the anatomy from single​​ view fluoroscopy images. This​​​‌ allows to obtain a‌ system able to automatically‌​‌ navigate across a moving​​ vascular anatomy under fluoroscopic​​​‌ imaging, even without injecting‌ a contrast agent. We‌​‌ validated our method in​​ a simulated environment on​​​‌ various synthetic static anatomies‌ on two realistic scenarios:‌​‌ a simulated beating heart​​ and a liver subjected​​​‌ to breathing motion. Our‌ approach leads to an‌​‌ average success rate of​​ 95% in reaching random​​​‌ targets within these anatomies.‌ This work formed the‌​‌ basis of Valentina Scarponi’s​​ PhD thesis, which was​​​‌ successfully completed last year.‌ It is now being‌​‌ extended and further developed​​ within the framework of​​​‌ a PhD thesis which‌ has started Ahmed Moustafa’s‌​‌ PhD thesis, which started​​ in October 2025.

6.3.3​​​‌ Contact models and haptics‌

Participants: Claire Martin,‌​‌ Thuc Long Ha,​​ Hadrien Courtecuisse.

Needle-based​​​‌ procedures such as biopsies‌ or radio-frequency ablation (RFA)‌​‌ of tumors are often​​ considered to diagnose and​​​‌ treat liver cancer for‌ their low invasiveness but‌​‌ raise difficulties for practitioners​​ related to needle placement​​​‌ and visibility of internal‌ anatomical structures. In 54‌​‌, 41 we developed​​ a contact model specific​​​‌ to needle-tissue interactions to‌ improve the realism of‌​‌ the resulting haptic rendering.​​ We present a novel​​​‌ method to update the‌ compliant coupling at high‌​‌ rates of a complete​​ contact system involving the​​​‌ mechanics of a large‌ object and the complete‌​‌ model of a flexible​​ needle. These updates allow​​​‌ to adapt the contact‌ directions to the needle‌​‌ deformations in the haptic​​ thread, with the aim​​​‌ of improving the resulting‌ haptic feedback. Updates of‌​‌ contact directions and the​​ related mechanical system according​​​‌ to high-rate deformations decrease‌ force feedback artifacts associated‌​‌ with low-rate mechanics while​​ maintaining high-rate performances for​​​‌ the haptic loop. In‌ 26, 40,‌​‌ 36, we presents​​ a fluoroscopy image‐based registration​​​‌ method along with a‌ comprehensive protocol for robotic‌​‌ needle insertion in radiofrequency​​ ablation (RFA) to treat​​​‌ liver cancer. The proposed‌ method uses real‐time fluoroscopic‌​‌ images acquired from a​​ C‐ARM system and integrates​​​‌ an inverse finite element‌ (FE) simulation to compute‌​‌ robotic commands for accurate​​ and adaptive needle steering.​​​‌ The registration procedure is‌ fully automated and involves‌​‌ the injection of multiple​​ radiopaque markers into the​​​‌ liver, enabling precise anatomical‌ registration and targeted tumor‌​‌ localization. A key challenge​​​‌ addressed in this work​ is the integration of​‌ this image‐based registration with​​ the inverse biomechanical simulation​​​‌ used to guide the​ robot during insertion. We​‌ describe how registration constraints​​ can be mapped onto​​​‌ the surface of the​ biomechanical model to ensure​‌ consistent alignment between image​​ data and robotic actuation.​​​‌ Designed to be adaptable​ to varying levels of​‌ radiologist expertise and applicable​​ across a wide range​​​‌ of tumor locations, this​ method provides a robust​‌ and versatile solution for​​ improving the accuracy and​​​‌ safety of minimally invasive​ liver cancer treatments.

6.3.4​‌ Advances in percutaneous intervention​​ devices and simulation

Participants:​​​‌ Juan Verde.

Recent​ work from the MIMESIS​‌ team has focused on​​ improving percutaneous interventions through​​​‌ innovative devices, simulation tools,​ and planning methods. Isambert​‌ et al. 28 introduced​​ a single-insertion deployment strategy​​​‌ for multimodal fiducial markers,​ combining a flexible ARC​‌ needle and a custom​​ deployment tool, enabling precise​​​‌ placement of multiple markers​ via a single entry​‌ while maintaining visibility across​​ CT, fluoroscopy, ultrasound, and​​​‌ MRI. Morin et al.​ 38 further investigated the​‌ ARC passive-steerable needle, demonstrating​​ its ability to perform​​​‌ non-linear trajectories in phantom​ models, allowing multiple ablations​‌ with a single insertion​​ and providing valuable data​​​‌ for enhancing simulation models​ and preoperative planning. Complementing​‌ these device-focused studies, Mehtali​​ et al. 32 developed​​​‌ HEAT, a multi-resolution simulation​ framework for thermal ablation​‌ therapy, which accelerates thermal​​ propagation calculations while preserving​​​‌ accuracy, enabling interactive planning​ with multiple needles. Together,​‌ these contributions advance safer,​​ more efficient, and more​​​‌ precise guidance and planning​ for minimally invasive therapies.​‌ We were involved in​​ the planning and implementation​​​‌ of clinical studies.

6.3.5​ Optimal control to limit​‌ epidemia

Participants: Michel Duprez​​.

There is currently​​​‌ no highly effective vaccine​ against many mosquito-borne diseases​‌ such as malaria, dengue,​​ lymphatic filariasis, Zika, chikungunya,​​​‌ yellow fever, and Japanese​ encephalitis, which continue to​‌ represent a major public​​ health burden worldwide. One​​​‌ effective way to limit​ the spread of these​‌ diseases is to target​​ their vector, the mosquito,​​​‌ rather than the pathogen​ itself. The sterile insect​‌ technique (SIT) is a​​ biological control method that​​​‌ can be used either​ to eradicate a wild​‌ mosquito population or to​​ reduce it below a​​​‌ critical threshold, thereby decreasing​ both human nuisance and​‌ epidemiological risk. The core​​ principle of SIT consists​​​‌ in releasing large numbers​ of sterilized male mosquitoes​‌ into the environment. Wild​​ females then mate indiscriminately​​​‌ with both sterilized and​ fertile wild males. Since​‌ eggs resulting from matings​​ with sterilized males are​​​‌ non-viable, repeated releases lead​ to a progressive decline​‌ of the overall mosquito​​ population over time. The​​​‌ success of this strategy​ depends on a large​‌ number of ecological, biological,​​ and operational parameters, including​​​‌ mosquito life-cycle dynamics, mating​ competitiveness, spatial heterogeneity, and​‌ environmental conditions. One of​​ the main biological and​​​‌ logistical challenges is to​ determine where and when​‌ sterilized males should be​​ released in order to​​​‌ maximize the efficiency of​ the control strategy while​‌ minimizing costs. In a​​ series of articles, we​​ have investigated the mathematical​​​‌ properties of optimal release‌ strategies using dynamical systems‌​‌ and optimal control theory.​​ More recently, in 23​​​‌, we studied the‌ impact of mosquito migration‌​‌ and spatial dispersal on​​ the effectiveness of SIT-based​​​‌ control strategies.

6.3.6 Behavioral‌ feedback improves sustained attention‌​‌ in human subjects

Participants:​​ Axel Hutt, Negin​​​‌ Majzoubi.

Attention deficit‌ disorder with or without‌​‌ hyperactivity (ADHD) is a​​ common neurodevelopmental disorder. Patients​​​‌ with ADHD exhibit symptoms‌ of hyperactivity, impulsivity and/or‌​‌ inattention, which significantly impact​​ their family, professional and​​​‌ social lives, with obvious‌ repercussions at the societal‌​‌ level. Studies have shown​​ that providing feedback has​​​‌ a positive effect on‌ performance levels. Our hypotheses‌​‌ are that performance feedback​​ can also improve attention​​​‌ deficits in adults with‌ ADHD. Behaviourally, feedback could‌​‌ improve performance, particularly by​​ increasing the number of​​​‌ correct responses and reducing‌ response time. The objective‌​‌ of this thesis project​​ is to create a​​​‌ new digital method, a‌ non-pharmacological treatment based on‌​‌ performance feedback. It is​​ complemented by EEG observations​​​‌ and statistical analysis utilizing‌ machine learning techniques. First‌​‌ analysis steps on behavioral​​ data indicate that performance​​​‌ feedback improves sustained attention.‌ The subsequent analysis aims‌​‌ to correlate these results​​ in behavioral data to​​​‌ EEG signal features.

6.3.7‌ Modulation of temporal prediction‌​‌ by transcranial magnetic stimulation​​ in the context of​​​‌ Schizophrenia

Participants: Axel Hutt‌, Telma Nette.‌​‌

To know when an​​ event will occur, be​​​‌ ready to perceive it‌ and react to it,‌​‌ we use what is​​ called temporal prediction. It​​​‌ allows us to anticipate‌ the arrival of an‌​‌ event and optimise our​​ behaviour. However, the ability​​​‌ to predict the arrival‌ of an event is‌​‌ impaired in people with​​ schizophrenia. This deterioration is​​​‌ visible on a scale‌ of seconds but also‌​‌ on a scale of​​ milliseconds. This scale is​​​‌ suggestive of a role‌ for the cerebellum There‌​‌ is currently no treatment​​ for such a highly​​​‌ debilitating disorder. One clinical‌ approach involves the use‌​‌ of transcranial magnetic stimulation​​ (TMS) applied to the​​​‌ cerebellum. It is targeted‌ because of its role‌​‌ in millisecond-scale prediction, its​​ involvement in a cerebellar-thalamo-cortical​​​‌ network that allows for‌ the adjustment of predictions,‌​‌ and because of what​​ is known about cerebellar​​​‌ dysfunction in schizophrenia. The‌ impact and mechanisms of‌​‌ action of TMS are​​ poorly understood, so it​​​‌ is necessary to verify‌ that the chosen stimulation‌​‌ does not worsen the​​ condition of people with​​​‌ schizophrenia before applying these‌ methods to patients. This‌​‌ study is therefore being​​ conducted first on healthy​​​‌ volunteers to verify whether‌ a TMS session targeting‌​‌ the cerebellum can effectively​​ modify the temporal processes​​​‌ likely to play a‌ role in the pathophysiology‌​‌ of schizophrenia. To answer​​ this question, we recruited​​​‌ human subjects who underwent‌ TMS sessions and observed‌​‌ EEG activity during prediction​​ tasks. First analysis of​​​‌ the EEG revealed an‌ impact of TMS on‌​‌ the evoked potential component​​ CNV.

7.1.1 LN Robotics​​​‌

Participants: Stéphane Cotin,​ Michel Duprez, François​‌ Lecomte.

MIMESIS participates​​ in a bilateral collaboration​​​‌ with LN Robotics,​ a company originating from​‌ the robotic research team​​ at Asan Medical Center​​​‌ (Seoul, South Korea) and​ established in 2019 to​‌ develop endovascular robots specialized​​ in cardiac interventions. The​​​‌ collaboration began in 2023​ with the loan of​‌ one of their robot​​ prototypes and was formalized​​​‌ in October 2024 with​ a two-year research contract​‌ covering autonomous robotic navigation​​ and 3D shape recovery​​​‌ from fluoroscopic images.

Within​ this project, MIMESIS contributes​‌ its expertise in computational​​ modeling, fluoroscopic image processing,​​​‌ and reinforcement learning–based control​ strategies for endovascular robots.​‌ The work focuses on​​ patient-specific navigation and anatomical​​​‌ motion estimation from single-view​ fluoroscopy, enabling autonomous operation​‌ with reduced reliance on​​ contrast agents.

This collaboration​​​‌ has already led to​ two research papers and​‌ involved one PhD student.​​ Currently, it supports one​​​‌ new PhD student (​Ahmed Moustafa) and​‌ one postdoctoral researcher (​​François Lecomte), providing​​​‌ a pathway for transferring​ advanced research methods from​‌ MIMESIS to industrial robotic​​ platforms and advancing autonomous​​​‌ endovascular interventions with improved​ safety and precision.

8.1.1​​​‌ Visits of international scientists​

ANR – SPECULAR

Participants:​​​‌ Hadrien Courtecuisse.

• Title:​
  Simulation of needle insertion​‌ with virtual reality and​​ haptics.
• Duration:
  2021 –​​​‌ 2025
• Coordinator:
  Stephane Cotin​ (MIMESIS)
• Partners:
  • Inria Antenna​‌ in Strasbourg, MIMESIS team​​ (France)
  • Inria Research Center​​​‌ at Université de Lille,​ DEFROST team (France)
  • InfinityTech​‌ 3D (France)
• Inria contact:​​
  stephane.cotin@inria.fr
• Summary:

  The objective​​​‌ of this project is​ to develop a complete​‌ virtual reality training system​​ for radiofrequency ablation. The​​​‌ research program includes the​ real-time simulation of the​‌ needle-organ interactions, realistic and​​ immersive rendering of the​​​‌ operating room, medical image​ generation and haptic feedback.The​‌ results of this project​​ will accelerate the training​​​‌ of these procedures and​ could change the standard​‌ of care which remains​​ a surgery in many​​​‌ cases.

  MIMESIS will play​ a central role in​‌ the project by contributing​​ its expertise in physically-based​​​‌ simulation, interactive modeling, and​ medical applications. In particular,​‌ the team will develop​​ advanced models for real-time​​​‌ needle-tissue interaction, ensuring both​ numerical robustness and high-fidelity​‌ biomechanical behavior. MIMESIS will​​ also contribute to the​​​‌ integration of these models​ within an immersive virtual​‌ reality framework, with a​​ strong focus on realistic​​​‌ haptic feedback and user​ interaction.

  In addition, the​‌ team will leverage its​​ experience with the SOFA​​​‌ platform to design modular​ and extensible simulation components,​‌ facilitating technology transfer and​​ future developments. Overall, MIMESIS​​​‌ will provide the scientific​ and technological foundations required​‌ to achieve a clinically​​ relevant and interactive training​​​‌ system.

ANR – S-KELOID​

Participants: Michel Duprez.​‌

• Title:
  Understanding Keloid Disorders:​​ A multi-scale in vitro/in​​ vivo/in silico approach towards​​​‌ digital twins of skin‌ organoids on the chip.‌​‌
• Duration:
  2021 – 2025​​
• Coordinator:
  Raluca Eftimie and​​​‌ Stéphane Bordas (Univ. Luxembourg)‌
• Partners:
  • Laboratoire de Mathématiques‌​‌ de Besançon (France)
  • CHU​​ de Besançon (France)
  • FEMTO-ST,​​​‌ Besançon (France)
  • Institut Mathématiques‌ de Bourgogne, Dijon (France)‌​‌
  • Université du Luxembourg (Luxembourg)​​
  • Inria Antenna in Strasbourg,​​​‌ MIMESIS team (France)
• Inria‌ contact:
  michel.duprez@inria.fr
• Summary:

  Mathematical‌​‌ and numerical modeling approaches​​ allow us to integrate​​​‌ pathological processes that occur‌ across different scales: single‌​‌ cell, cell assembly and​​ tissue. The S-Keloid project​​​‌ aims to investigate the‌ role of mechanical and‌​‌ inflammatory environmental factors on​​ cells associated with keloid​​​‌ disorders, which are the‌ formation of pathological scars.‌​‌ From applying experimental tests​​ at tissue-scale and using​​​‌ a multiscale approach, the‌ mechanical stress fields will‌​‌ be integrated into the​​ 3D mathematical model. Parameter​​​‌ identification, optimization and their‌ use across multiple scales‌​‌ will ensure the realism​​ of the models and​​​‌ the quantitative and qualitative‌ predictions of the keloid‌​‌ disorder.

  MIMESIS will contribute​​ to the S-KELOID project​​​‌ through its expertise in‌ computational biomechanics and numerical‌​‌ modeling of soft tissues.​​

ANR – PhiFEM

Participants:​​​‌ Stéphane Cotin, Michel‌ Duprez.

• Title:
  $\mathrm{\phi ‌‌}$-FEM : development of​​ a Finite Element Method​​​‌ for the design of‌ real-time digital twins in‌​‌ surgery
• Duration:
  2022 –​​ 2026
• Coordinator:
  Michel Duprez​​​‌ (Inria)
• Partners:
  • Inria Antenna‌ in Strasbourg, MIMESIS team‌​‌ (France)
  • Laboratoire de Mathématiques​​ de Besançon (France)
  • Institut​​​‌ de Mathématiques Alexander Grothendieck,‌ Montpellier (France)
  • Institut de‌​‌ Recherche en Mathématiques Appliquées,​​ Strasbourg (France)
  • Université du​​​‌ Luxembourg (Luxembourg)
• Inria contact:‌
  michel.duprez@inria.fr
• Summary:

  $\phi$-FEM‌​‌ is a recently proposed​​ finite element method for​​​‌ the efficient numerical solution‌ of partial differential equations‌​‌ posed in domains of​​ complex shapes, using simple​​​‌ regular meshes. The main‌ goal of this project‌​‌ is to further develop​​ $\phi$-FEM, turning it​​​‌ into a tool for‌ efficient, patient-specific and real-time‌​‌ simulations of human organs.​​ To reach this objective,​​​‌ we shall adapt $\mathrm{\phi ‌}$-FEM to the equations‌​‌ appropriate to biomechanics, provide​​ an efficient implementation for​​​‌ it allowing for the‌ use of actual organ‌​‌ geometries, and finally combine​​ it with convolutional neural​​​‌ networks to make it‌ real time after training.‌​‌ The ultimate, long-term, goal​​ is thus to contribute​​​‌ to the construction of‌ digital twins of organs‌​‌ able to guide the​​ surgical act in real​​​‌ time using information acquired‌ before the operation and‌​‌ to reduce the costs​​ of a medical doctors'​​​‌ training by working on‌ visual organs. The innovation‌​‌ of $\phi$-FEM lies​​ in its ability to​​​‌ combine the ease of‌ implementation of classical immersed‌​‌ boundary methods with the​​ accuracy of more recent​​​‌ CutFEM/XFEM approaches. It incorporates,‌ by its very construction,‌​‌ the popular description of​​ geometry by Level Set​​​‌ functions, which can represent‌ the real geometry with‌​‌ whatever accuracy desired which​​ makes this approach numerically​​​‌ less expensive than classical‌ finite element methods. The‌​‌ $\phi$-FEM paradigm will​​ also be used to​​​‌ develop efficient registration algorithms.‌ Our results will be‌​‌ integrated into the open-source​​​‌ SOFA platform developed in​ the MIMESIS team to​‌ facilitate dissemination.

  MIMESIS will​​ contribute to the $\mathrm{\phi ‌}$-FEM project through its​ expertise in computational biomechanics​‌ and real-time numerical simulation.​​ The team will support​​​‌ the adaptation of $\mathrm{\phi }$-FEM to biomechanical models​‌ of soft tissues, enabling​​ efficient simulations on complex,​​​‌ patient-specific geometries.

  In particular,​ MIMESIS will focus on​‌ bridging advanced numerical methods​​ with clinically relevant applications,​​​‌ contributing to the development​ of fast and accurate​‌ digital twins for surgical​​ guidance.

BPI – MediTwin​​​‌

Participants: Stéphane Cotin,​ Michel Duprez, Pablo​‌ Alvarez.

• Title:
  MediTwin​​
• Duration:
  2023 – 2028​​​‌
• Coordinator:
  Dassault Systèmes
• Partners:​
  • Inria
  • Dassault Systèmes
  • IHU​‌ Imagine
  • IHU FOReSIGHT
  • Institut​​ du Cerveau (ICM)
  • IHU​​​‌ LIRYC
  • ICAN (Institut de​ Cardiométabolisme et de la​‌ Nutrition)
  • IHU-B PRISM
  • IHU​​ STRASBOURG
  • CHU Nantes
  • CODOC​​​‌
  • Neurometers
  • Qairnel
  • inHEART
• Inria​ contact:
  stephane.cotin@inria.fr
• Summary:

  MIMESIS​‌ is involved in a​​ bilateral collaboration with Dassault​​​‌ Systèmes within the framework​ of the MediTwin project,​‌ funded by the BPI.​​ MediTwin aims at supporting​​​‌ the digital transformation of​ medical practices through a​‌ European and sovereign innovation​​ platform for medical decision​​​‌ support, based on patient-specific​ digital twins.

  Within this​‌ project, MIMESIS primarily contributes​​ to the modeling and​​​‌ simulation of microwave tumor​ ablation procedures. Our work​‌ focuses on biomechanical and​​ thermo-physical modeling, patient-specific digital​​​‌ twins, and the optimization​ of treatment planning, with​‌ the objective of improving​​ both pre-operative planning and​​​‌ intra-operative guidance. A particular​ emphasis is placed on​‌ the integration of simulation​​ results into the clinical​​​‌ workflow through augmented reality​ technologies.

  The collaboration currently​‌ supports two PhD theses​​ that started in October​​​‌ 2025. The PhD of​ Francesco Dettori focuses on​‌ modeling and simulation aspects​​ of microwave tumor ablation,​​​‌ while the PhD of​ Ilias Nahmed addresses treatment​‌ planning optimization and augmented​​ reality-based intra-operative guidance. This​​​‌ collaboration provides a direct​ pathway for transferring advanced​‌ research results from MIMESIS​​ towards an industrial-grade digital​​​‌ twin platform with strong​ clinical impact.

BPI –​‌ PREMYOM

Participants: Stéphane Cotin​​, Michel Duprez,​​​‌ Pablo Alvarez.

• Title:​
  Prise en charge et​‌ Ralentissement de l'Epidémie de​​ MYopie par l'Optique Médicale​​​‌
• Duration:
  2023 – 2028​
• Coordinator:
  EssilorLuxottica
• Partners:
  • Inria​‌
  • EssilorLuxottica
  • InSimo
  • Hôpital Fondation​​ Adolphe de Rothschild (HFAR)​​​‌
  • Fondation Voir et Entendre​ - Institut de la​‌ Vision
  • Institut Mines Telecom​​ (IMT)
• Inria contact:
  stephane.cotin@inria.fr​​​‌
• Summary:

  MIMESIS participates in​ the PREMYOM project, coordinated​‌ by EssilorLuxottica (2023 –​​ 2028) and co-financed by​​​‌ Bpifrance under the France​ 2030 plan. The project​‌ aims to slow the​​ global epidemic of myopia​​​‌ by establishing a therapeutic​ benchmark for personalized treatment,​‌ combining medical optics, clinical​​ expertise, and digital innovation.​​​‌

  Within PREMYOM, MIMESIS contributes​ its expertise in computational​‌ modeling and simulation to​​ support patient-specific optimization of​​​‌ myopia management strategies. The​ team focuses on designing​‌ and evaluating individualized treatment​​ plans based on biomechanical​​​‌ and optical models, integrating​ clinical and imaging data​‌ to improve the effectiveness​​ and safety of interventions.​​​‌

  The collaboration currently supports​ one PhD and one​‌ postdoctoral researcher. The PhD​​ of Even Harsigny,​​ started in October 2025,​​​‌ focuses on modeling and‌ optimization of myopia progression‌​‌ in pediatric patients. The​​ post-doc of Thomas Saigre​​​‌, started in November‌ 2025, works on implementing‌​‌ computational tools for personalized​​ treatment planning and evaluating​​​‌ the impact of optical‌ interventions. This partnership allows‌​‌ the transfer of advanced​​ simulation and modeling techniques​​​‌ from MIMESIS to industrial‌ and clinical applications in‌​‌ the field of vision​​ care.

BPI – POCUSI​​​‌

Participants: Stéphane Cotin.‌

• Title:
  Point of Care‌​‌ UltraSound for Screening and​​ Intervention
• Duration:
  2025 –​​​‌ 2030
• Coordinator:
  EssilorLuxottica
• Partners:‌
  • Inria
  • E-Scopics S.A.S
  • VERMON‌​‌
  • IHU Strasbourg
• Inria contact:​​
  stephane.cotin@inria.fr
• Summary:

  MIMESIS participates​​​‌ in the POCUSI project‌ (Point of Care UltraSound‌​‌ for Screening and Intervention),​​ a national consortium led​​​‌ by E-Scopics and co-funded‌ by Bpifrance under the‌​‌ France 2030 plan. POCUSI​​ aims to develop a​​​‌ new generation of software-based,‌ intelligent, and frugal ultrasound‌​‌ systems to support screening,​​ monitoring, and image-guided interventions,​​​‌ with a focus on‌ chronic liver diseases and‌​‌ associated metabolic and cardiovascular​​ risk factors.

  Within this​​​‌ project, MIMESIS contributes its‌ expertise in computational modeling,‌​‌ simulation, and digital twin​​ technologies to support real-time,​​​‌ quantitative assessment of liver‌ pathology. Our work focuses‌​‌ on integrating simulation-driven guidance​​ and decision support into​​​‌ the POCUSI platform, enabling‌ non-expert operators to accurately‌​‌ evaluate fibrosis, steatosis, and​​ other key disease markers,​​​‌ and to assist therapeutic‌ interventions.

  This collaboration provides‌​‌ a pathway for transferring​​ advanced research methods from​​​‌ MIMESIS to an industrial-grade,‌ clinically relevant, and accessible‌​‌ ultrasound solution, contributing to​​ the emergence of a​​​‌ French sovereign digital ultrasound‌ technology.

9.1.1​​ Scientific events: organisation

9.1.2​​ Journal

9.1.3 Invited talks​​​‌

• Michel Duprez gave a​ talk in the conference​‌ "PICOF" (October 2025, Hammamet,​​ Tunisie)
• Michel Duprez gave​​​‌ a talk in the​ conference "Enumath" (Sept. 2025,​‌ Heidelberg, Germany)
• Michel Duprez​​ gave a talk in​​​‌ the conference "The 30th​ Biennial Numerical Analysis Conference"​‌ (June 2025, Glasgow, UK)​​
• Camille Gontier 's coauthors​​​‌ presented a poster at​ the 11th International BCI​‌ Meeting (June 2025, Banff,​​ Canada)
• Camille Gontier gave​​​‌ an invited talk at​ the "CORTICO scientific days"​‌ (May 2025, Lyon)
• Michel​​ Duprez gave a talk​​​‌ in the conference "Coupled​ Problem" (May 2025 ,​‌ Villasimius, Sardinia)
• Michel Duprez​​ gave a talk in​​​‌ the conference "DTE AICOMAS​ 2025" (Feb. 2025 ,​‌ Paris)
• Stéphane Cotin gave​​ a talk at the​​​‌ "IHPBA Innovation Webinar". (Sept.​ 2025, online)
• Stéphane Cotin​‌ gave a keynote address​​ at the "12th Seoul​​​‌ Asan Hospital Medical Research​ Institute Symposium". (Nov. 2025,​‌ Seoul, South Korea)

9.1.4​​ Leadership within the scientific​​​‌ community

• Stéphane Cotin is​ scientific manager for the​‌ MediTwin project (2024-2029) involving​​ 9 Inria teams
• Stéphane​​​‌ Cotin is the main​ scientific lead for the​‌ PREMYOM project (2024-2029) involving​​ 3 Inria teams

9.1.5​​​‌ Scientific expertise

• Axel Hutt​ was reviewer for the​‌ PIQ (Programme Inria Quadrant)​​ and "NWO" (Dutch Research​​​‌ Council).

9.1.6 Research administration​

• Hadrien Courtecuisse is member​‌ of the ICube laboratory​​ council (Strasbourg)
• Hadrien Courtecuisse​​​‌ involved in the doctoral​ selection process of the​‌ MSII doctoral school
• Hadrien​​ Courtecuisse represents the MLMS​​​‌ team within the ICube​ Gaia computing platform
• Michel​‌ Duprez was elected co-head​​ of the MLMS ICube​​​‌ team with Hyewon Seo​ since the 22th November​‌ 2025

9.2.1 Teaching

Master:

• Michel​‌ Duprez , Optimal control,​​ 28h, M2 CSMI, University​​​‌ of Strasbourg.
• Michel Duprez​ , Optimisation, 28h, M1​‌ CSMI, University of Strasbourg​​
• Axel Hutt , Evoked​​​‌ Potentials, 2h, M2 Neuroscience,​ University of Strasbourg
• Hadrien​‌ Courtecuisse , 28h, CI​​ Real-time simulation, M2 HealthTech,​​​‌ University of Strasbourg
• Hadrien​ Courtecuisse , 24h, 3D​‌ modelisation and simulation, M2​​ IRMC, University of Strasbourg​​​‌

9.2.2 Supervision

Phd in​ progress

• Axel Hutt is​‌ supervising the PhD of​​ Negin Majzoubi together with​​​‌ Anne Bonnefond (INSERM 1329)​ (2024-2027). Title: "Effets neurophysiologiques​‌ et comportementaux du feedback​​ de performance: un nouvel​​​‌ outil numérique pour améliorer​ le traitement des symptômes​‌ attentionnels dans le TDAH"​​
• Axel Hutt is supervising​​​‌ the PhD of Telma​ Nette together with Anne​‌ Giersch (INSERM 1329) (2025-2028).​​ Title: "Modulation of temporal​​​‌ prediction by transcranial magnetic​ stimulation"
• Michel Duprez supervises​‌ the PhD of Frédérique​​ Lecourtier (2023-2026) together with​​​‌ Emmanuel Franck (MACARON, Inria)​ and Vanessa Lleras (IMAG,​‌ Montpellier). Title: "Finite element​​ methods and neural networks​​ for augmented surgery"
• Stéphane​​​‌ Cotin , Pablo Alvarez‌ and Michel Duprez co-supervise‌​‌ the PhD of Even​​ Harsigny (2025-2028). Title: "Customized​​​‌ modeling and control of‌ the eye-head-neck system in‌​‌ children with myopia"
• Stéphane​​ Cotin , Pablo Alvarez​​​‌ and Michel Duprez co-supervise‌ the PhD of Ilias‌​‌ Nahmed (2025-2028). Title: "Predicting​​ the outcomes of liver​​​‌ tumor ablation therapy using‌ an AI-driven digital twin"‌​‌
• Stéphane Cotin , Pablo​​ Alvarez and Michel Duprez​​​‌ co-supervise the PhD of‌ Francesco Dettorri (2025-2028). Title‌​‌ : "Personalized planning of​​ liver tumor ablation using​​​‌ digital twins"
• Stéphane Cotin‌ and Michel Duprez co-supervise‌​‌ the PhD of Louis​​ Ducongé (2025-2028) together with​​​‌ Yannick Privat (IECL, Nancy)‌ . Title : "Shape‌​‌ optimization for angioplasty balloon​​ and stent design"
• Stéphane​​​‌ Cotin co-supervises Leo Widmer‌ (2025-2029) from the Departement‌​‌ Biomedical Engineering, Basel University​​
• Hadrien Courtecuisse supervises the​​​‌ PhD of Noah Bertholon‌ (2025-2028). Title: "IDEAL –‌​‌ Haptic simulation of needle​​ insertion".

Phd defended

• Michel​​​‌ Duprez supervises the PhD‌ of Killian Vuillemot (2022-2025)‌​‌ together with Vanessa Lleras​​ and Bijan Mohammadi (IMAG,​​​‌ Montpellier) . Title: "Unfitted‌ finite element method for‌​‌ the development of organ​​ digital twins". PhD defensed​​​‌ the 18th Decembre 2025.‌
• Hadrien Courtecuisse has supervised‌​‌ the PhD of Claire-Martin​​ (2022-2025). Title: "IDEAL –​​​‌ Haptic simulation of needle‌ insertion". PhD defensed the‌​‌ 4th Decembre 2025.
• Hadrien​​ Courtecuisse has supervised the​​​‌ PhD of Thuc Long‌ Ha (2021-2025). Title: "AI-enhanced‌​‌ robotic needle steering". PhD​​ defensed the 9th Decembre​​​‌ 2025.

Phd not defended‌

• Stéphane Cotin supervised the‌​‌ PhD of Nicola Zotto​​ (2021-2025) Title: "Combining AI​​​‌ and biomechanics for computer-assisted‌ interventions"

9.2.3 Juries

• Hadrien‌​‌ Courtecuisse has been examinator​​ of the Kenza Oussalah​​​‌ PhD (Insa Lyon December‌ 2025)
• Stéphane Cotin was‌​‌ member of the Habilitation​​ thesis committee of Michel​​​‌ Duprez , Strasbourg University,‌ November 2025
• Stéphane Cotin‌​‌ was member (and reviewer)​​ of the PhD committee​​​‌ of Martha Duraes, Université‌ of Montpellier, December 2025.‌​‌
• Axel Hutt is member​​ of the Comité de​​​‌ Suivi Individuel de thèse‌ (CSI) of Chaoyang PENG‌​‌ (INSERM 1329).
• Axel Hutt​​ is member of the​​​‌ CSI of Rajat Chandra‌ MISHRA (INCI, Strasbourg).

9.2.4‌​‌ Educational and pedagogical outreach​​

Axel Hutt gave a​​​‌ 1-day workshop on Spectral‌ analysis of neural observations‌​‌ at the OpenLab Heidelberg​​ (May, Heidelberg, Germany)

9.3.1 Productions (articles,‌ videos, podcasts, serious games,‌​‌ ...)

• Camille Gontier :​​ Invited talk during the​​​‌ “Neurocareers: Doing The Impossible!”‌ podcast

9.3.2 Participation in‌​‌ Live events

• Axel Hutt​​ chaired a panel discussion​​​‌ on Dangers of AI‌ at the Franco-German dialogue‌​‌ on AI (March 27-28)​​ as part of the​​​‌ Trifels Spring School 2025‌ in Annweiler, Germany.
• Axel‌​‌ Hutt was participant of​​ the discussion panel It's​​​‌ a trap: E-Commerce &‌ Why We Still Consume‌​‌ (May 27) at the​​ public conference Re:publica in​​​‌ Berlin, Germany.

9.3.3 Others‌ science outreach relevant activities‌​‌

• Camille Gontier : Talk​​ “Sciences en boucle fermée​​​‌ : utiliser l'IA pour‌ optimiser les expériences et‌​‌ les soins médicaux” during​​ the Inria welcome day​​​‌ (June 2025, Strasbourg)

International journals​​

• 20 articleG.Gabriel​​​‌ Alves Castro, A.‌Anne Bonnefond, B.-T.‌​‌Bich-Thuy Pham and A.​​Axel Hutt. Sustained​​​‌ attention in attention-deficit subjects‌ and the impact of‌​‌ binaural beat stimulation evaluated​​ by behavior and EEG​​​‌.Experimental Brain Research‌September 2025, 211‌​‌HALDOIback to​​ text
• 21 articleR.​​​‌Roland Becker, F.‌Franz Chouly, M.‌​‌Michel Duprez, T.​​Thomas Richter, C.-Y.​​​‌ P.Christian Pierre-Yves Rohan‌ and T.Thomas Wick‌​‌. Dual Weighted Residual-driven​​ adaptive mesh refinement to​​​‌ enhance biomechanical simulations.‌Advances in Applied Mechanics‌​‌61April 2025,​​ 75-127HALDOIback​​​‌ to text
• 22 article‌H. P.Huu Phuoc‌​‌ Bui, M.Michel​​ Duprez, P.Pierre‐yves​​​‌ Rohan, A.Arnaud‌ Lejeune, S. P.‌​‌Stéphane Pierre Alain Bordas​​​‌, M.Marek Bucki​ and F.Franz Chouly​‌. Enhancing biomechanical simulations​​ based on a posteriori​​​‌ error estimates: the potential​ of Dual Weighted Residual-driven​‌ adaptive mesh refinement.​​International Journal for Numerical​​​‌ Methods in Biomedical Engineering​411January 2025​‌HALDOIback to​​ text
• 23 articleY.​​​‌Yves Dumont, M.​Michel Duprez and Y.​‌Yannick Privat. Sterile​​ insect technique in a​​​‌ patch system: influence of​ migration rates on optimal​‌ single-patch releases strategies.​​Journal of Mathematical Biology​​​‌916October 2025​, 71HALDOI​‌back to text
• 24​​ articleM.Michel Duprez​​​‌, V.Vanessa Lleras​, A.Alexei Lozinski​‌, V.Vincent Vigon​​ and K.Killian Vuillemot​​​‌. Phi-FD : A​ well-conditioned finite difference method​‌ inspired by phi-FEM for​​ general geometries on elliptic​​​‌ PDEs.Journal of​ Scientific Computing10423​‌May 2025HALDOI​​back to text
• 25​​​‌ articleM.Michel Duprez​, V.Vanessa Lleras​‌, A.Alexei Lozinski​​, V.Vincent Vigon​​​‌ and K.Killian Vuillemot​. Phi-FEM-FNO: a new​‌ approach to train a​​ Neural Operator as a​​​‌ fast PDE solver for​ variable geometries.Communications​‌ in Nonlinear Science and​​ Numerical Simulation152January​​​‌ 2026, 109131HAL​DOIback to text​‌
• 26 articleT.Thuc‐long​​ Ha, J.Juan​​​‌ Verde, J.Julien​ Bert and H.Hadrien​‌ Courtecuisse. Toward fluoroscopy​​ guided robotic needle insertion​​​‌ for radio frequency ablation​.Computer Animation and​‌ Virtual Worlds363​​2025, e70025HAL​​​‌DOIback to text​
• 27 articleA.Axel​‌ Hutt. Additive noise​​ shapes the state and​​​‌ stability of random networks​.The European Physical​‌ Journal. Special TopicsJuly​​ 2025HALDOIback​​​‌ to text
• 28 article​C.Chloé Isambert,​‌ J.Juan Verde,​​ B.Benoit Wach and​​​‌ L.Lennart Rubbert.​ Single-insertion deployment of multimodal​‌ fingerprinted fiducial markers for​​ enhanced SBRT guidance.​​​‌Journal of Medical Devices​November 2025, 1-11​‌HALDOIback to​​ text
• 29 articleF.​​​‌François Lecomte-Denis, J.​Juan Verde, J.-L.​‌Jean-Louis Dillenseger and S.​​Stéphane Cotin. Domain​​​‌ agnostic 2D-3D deformable registration​ Application to fluoroscopic guidance​‌ without contrast agent.​​Medical Image Analysis105​​​‌June 2025, 103688​HALDOIback to​‌ text
• 30 articleJ.​​Jérémie Lefebvre, A.​​​‌Andrew Clappison, A.​André Longtin and A.​‌Axel Hutt. Myelin-induced​​ gain control in nonlinear​​​‌ neural networks.Communications​ Physics81April​‌ 2025, 145HAL​​DOIback to text​​​‌
• 31 articleJ.Jeremie​ Lefebvre, A.Aref​‌ Pariz, J.-P.Jean-Philippe​​ Thivierge and A.Axel​​​‌ Hutt. Surfing brain​ waves without sinking.​‌Physics of Life Reviews​​55December 2025,​​​‌ 55-57HALDOIback​ to text
• 32 article​‌J.Jonas Mehtali,​​ J.Juan Verde and​​​‌ C.Caroline Essert.​ HEAT : high-efficiency simulation​‌ for thermal ablation therapy​​.International Journal of​​​‌ Computer Assisted Radiology and​ Surgery206April​‌ 2025, 1135-1143HAL​​DOIback to text​​
• 33 articleD.Daniel​​​‌ Trotter, A.Aref‌ Pariz, A.Axel‌​‌ Hutt and J.Jérémie​​ Lefebvre. Morphological variability​​​‌ may limit single-cell specificity‌ to electric field stimulation‌​‌.Frontiers in Synaptic​​ Neuroscience17August 2025​​​‌HALDOIback to‌ text
• 34 articleZ.‌​‌Z. Zeng and H.​​H. Courtecuisse. Dynamic​​​‌ cutting simulation using elastic‌ snapping for mesh quality‌​‌ optimization.Computer Graphics​​ ForumMarch 2025HAL​​​‌DOIback to text‌

International peer-reviewed conferences

• 35‌​‌ inproceedingsS.Sidaty EL​​ HADRAMY, N.Nicolas​​​‌ Padoy and S.Stéphane‌ Cotin. Deform any‌​‌ liver in real-time.​​IEEE International Symposium on​​​‌ Biomedical Imaging (ISBI 2025)‌Houston, United StatesApril‌​‌ 2025HALDOIback​​ to text
• 36 inproceedings​​​‌T.Thuc‐long Ha,‌ J.Julien Bert and‌​‌ H.Hadrien Courtecuisse.​​ Real-time Robotic Needle Insertion​​​‌ In Deformable and Moving‌ Structure Using Learning-by-Example Method‌​‌.2026 IEEE International​​ Conference on Robotics and​​​‌ Automation (ICRA)Vienne, Austria‌June 2026HALback‌​‌ to text

Conferences without​​ proceedings

• 37 inproceedingsK.-P.​​​‌Karl-Philippe Beaudet, S.‌Sidaty El Hadramy,‌​‌ P. C.Philippe C​​ Cattin, J.Juan​​​‌ Verde and S.Stéphane‌ Cotin. Optimization-based calibration‌​‌ for intravascular ultrasound volume​​ reconstruction.ASMUS 2025:​​​‌ The 6th International Workshop‌ on Advances in Simplifying‌​‌ Medical UltraSoundDeajon, South​​ KoreaSeptember 2025HAL​​​‌DOIback to text‌
• 38 inproceedingsA.Antoine‌​‌ Morin, J.Juan​​ Verde, L.Lennart​​​‌ Rubbert and C.Caroline‌ Essert. Towards non-linear‌​‌ needle trajectories: performance analysis​​ of the ARC passive-steerable​​​‌ needle in a phan‌.CARS 2025 Computer‌​‌ Assisted Radiology and Surgery​​ 39th International Congress and​​​‌ ExhibitionBerlin, GermanyJune‌ 2025HALback to‌​‌ text

Doctoral dissertations and​​ habilitation theses

• 39 thesis​​​‌M.Michel Duprez.‌ A priori and a‌​‌ posteriori estimates of finite​​ element schemes and development​​​‌ of commands for some‌ dynamic phenomena.Université‌​‌ de StrasbourgNovember 2025​​HALback to text​​​‌
• 40 thesisT. L.‌Thuc Long Ha.‌​‌ Fluoroscopy guided robotic needle​​ insertion for radio frequency​​​‌ ablation.Strasbourg University‌December 2025HALback‌​‌ to text
• 41 thesis​​C.Claire Martin.​​​‌ Modeling and resolution of‌ needle-tissue interactions for a‌​‌ fast and stable haptic​​ rendering: Application to hepatic​​​‌ percutaneous procedures.Université‌ de Strasbourg (Unistra), FRA.‌​‌December 2025HALback​​ to text
• 42 thesis​​​‌K.Killian Vuillemot.‌ Nonconformal Finite Element methods‌​‌ applied to the real​​ time design of digital​​​‌ twins of organs.‌Université de montpellierDecember‌​‌ 2025HALback to​​ textback to text​​​‌

Reports & preprints

• 43‌ miscH.Hélène Barucq‌​‌, M.Michel Duprez​​, F.Florian Faucher​​​‌, E.Emmanuel Franck‌, F.Frédérique Lecourtier‌​‌, V.Vanessa Lleras​​, V.Victor Michel-Dansac​​​‌ and N.Nicolas Victorion‌. Enriching continuous Lagrange‌​‌ finite element approximation spaces​​ using neural networks.​​​‌February 2025HALback‌ to text
• 44 misc‌​‌R.Roland Becker,​​ R.Raphaël Bulle,​​​‌ M.Michel Duprez and‌ V.Vanessa Lleras.‌​‌ Residual-based a posteriori error​​​‌ estimates with boundary correction​ for Phi-FEM.February​‌ 2025HALback to​​ text
• 45 miscA.​​​‌Alex Bespalov and R.​Raphaël Bulle. An​‌ adaptive multimesh rational approximation​​ scheme for the spectral​​​‌ fractional Laplacian.2025​HALDOIback to​‌ text
• 46 miscS.​​Sidaty El Hadramy,​​​‌ B.Belkacem Acidi,​ N.Nicolas Padoy and​‌ S.Stéphane Cotin.​​ Optimization in latent space​​​‌ for real-time intraoperative characterization​ of digital twins.​‌September 2025HALback​​ to text
• 47 misc​​​‌C.Camille Gontier,​ L.Laura Kalinski,​‌ J.Johannes Striebel,​​ M.Maximilian Sturm,​​​‌ Z.Zoe Meerholz,​ S.Sarah Schunk,​‌ Y.Yannick Lichterfeld and​​ C.Christian Liemersdorf.​​​‌ A computational model of​ altered neuronal activity in​‌ altered gravity.July​​ 2025HALDOIback​​​‌ to text

Software

• 48​ softwareR.Raphael Bulle​‌, M.Michel Duprez​​ and K.Killian Vuillemot​​​‌. phiFEM: a convenience​ package for using phiFEM​‌ with FEniCSx.October​​ 2025Inria Nancy Grand-Est​​​‌ lic: Creative Commons Attribution​ 2.5 Generic; GNU Affero​‌ General Public License v3.0​​.HALDOISoftware​​​‌ HeritageVCSback to​ text