MORPHEME

MORPHEME - 2025

2025Activity‌ reportTeamMORPHEME

RNSR:‌ 201120999G

Research center Inria‌‌ Centre at Université Côte d'Azur
In partnership with:‌CNRS, Université Côte d'Azur‌
Team name: Morphologie et‌‌ Images
In collaboration with:Laboratoire informatique, signaux systèmes‌ de Sophia Antipolis (I3S),‌ Institut de Biologie de‌‌ Valrose

Creation of the Team: 2013 July 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

A3.4.‌ Machine learning and statistics
A5.3. Image processing and‌ analysis
A5.3.2. Sparse modeling and image representation
A5.3.3.‌ Pattern recognition
A5.3.4. Registration
A5.9. Signal processing
A5.9.3.‌ Reconstruction, enhancement
A5.9.5. Sparsity-aware processing
A5.9.6. Optimization tools‌
A6.1. Methods in mathematical modeling
A6.1.1. Continuous Modeling‌ (PDE, ODE)
A6.1.2. Stochastic Modeling
A6.3.1. Inverse problems‌
A9.2.1. Supervised learning
A9.2.2. Unsupervised learning
A9.2.4. Optimization‌ and learning
A9.2.6. Neural networks
A9.2.7. Kernel methods‌
A9.2.8. Deep learning
A9.12.3. Content retrieval
A9.12.4. 3D‌ and spatio-temporal reconstruction
A9.12.5. Object tracking and motion‌ analysis
A9.12.6. Object localization

1 Team members, visitors,‌ external collaborators

Research Scientists

Xavier Descombes [Team‌ leader, INRIA, Senior Researcher, HDR‌]
Florence Besse [CNRS, Senior Researcher‌, HDR]
Laure Blanc-Féraud [CNRS,‌ Senior Researcher, HDR]
Luca Calatroni [‌CNRS, Researcher, until Jan 2025]‌
Eric Debreuve [CNRS, Researcher, HDR‌]
Grégoire Malandain [INRIA, Senior Researcher‌, HDR]
Caroline Medioni [CNRS,‌ Researcher, HDR]
Ellen Van Obberghen [‌INSERM, Emeritus, HDR]

Faculty Members‌

Thomas Boudier [Ecole Centrale Méditerranée, Associate‌ Professor, until Jul 2025, HDR]‌
Imen Chtourou [UNIV COTE D'AZUR, from‌ Sep 2025]
Fabienne De Graeve [UNIV‌ COTE D'AZUR, HDR]
Salvish Goomanee [‌UNIV COTE D'AZUR, from Oct 2025]‌

Post-Doctoral Fellow

Emmanuel Bouilhol [UNIV COTE D'AZUR‌, Post-Doctoral Fellow]

PhD Students

Moncef Belaskri‌ [UNIV TLEMCEN, from Sep 2025]‌
Morgane Fierville [CNRS]
Haydar Jammoul [‌UNIV COTE D'AZUR]
Faisal Jayousi [CNRS‌, until Sep 2025]
Anna Kestel [‌INRIA]
Inès Landolsi [CNRS, from‌ Nov 2025]
Alexandre Martin [INRIA,‌ until Apr 2025]
Hamza Mentagui [CNRS‌]
Mohamad Mohamad [UNIV COTE D'AZUR]‌
Meryem Sikouky [UNIV COTE D'AZUR]
Aneva‌ Doliciane Tsafack [UNIV COTE D'AZUR]

Interns‌ and Apprentices

Ivan Magistro Contenta [INRIA,‌ Intern, until Apr 2025]
Raffaele Martone‌ [INRIA, Intern, from Sep 2025]
Sheyenne Nguyen [‌INRIA, Intern,‌ from Feb 2025 until‌‌ Jun 2025]
Cristiano Parenti [ Modena university‌, from Mar 2025‌ until May 2025]‌‌

Administrative Assistants

Marylène Fontana [INRIA, from‌ Sep 2025]
Belfegas‌ Nadia [CNRS]‌‌
Stéphanie Verdonck [INRIA, until Aug 2025‌]

External Collaborator

Francesco‌ Ponzio [Politecnico di‌‌ Torino, from May 2025]

2 Overall‌ objectives

Morpheme is a‌ joint project between INRIA,‌‌ CNRS and the University of Côte d'Azur (UniCA);‌ Signals and Systems Laboratory‌ (I3S) (UMR 6070); the‌‌ Institute for Biology of Valrose (iBV) (CNRS/INSERM).

The‌ scientific objectives of Morpheme‌ are to characterize and‌‌ model the development and the morphological properties of‌ biological structures from the‌ cell to the supra-cellular‌‌ scale. Being at the interface between computational science‌ and biology, we plan‌ to understand the morphological‌‌ changes that occur during development combining in vivo‌ imaging, image processing and‌ computational modeling.

The morphology‌‌ and topology of mesoscopic structures, indeed, do have‌ a key influence on‌ the functional behavior of‌‌ organs. Our goal is to characterize different populations‌ or development conditions based‌ on the shape of‌‌ cellular and supra-cellular structures, including micro-vascular networks and‌ dendrite/axon networks. Using microscopy‌ or tomography images, we‌‌ plan to extract quantitative parameters to characterize morphometry‌ over time and in‌ different samples. We will‌‌ then statistically analyze shapes and complex structures to‌ identify relevant markers and‌ define classification tools. Finally,‌‌ we will propose models explaining the temporal evolution‌ of the observed samples.‌ With this, we hope‌‌ to better understand the development of normal tissues,‌ but also characterize at‌ the supra-cellular level different‌‌ pathologies such as the Fragile X Syndrome, Alzheimer‌ or diabetes.

3 Research‌ program

3.1 Research program‌‌

The recent advent of an increasing number of‌ new microscopy techniques giving‌ access to high throughput‌‌ screenings and micro or nano-metric resolutions provides a‌ means for quantitative imaging‌ of biological structures and‌‌ phenomena. To conduct quantitative biological studies based on‌ these new data, it‌ is necessary to develop‌‌ non-standard specific tools. This requires using a multi-disciplinary‌ approach. We need biologists‌ to define experiment protocols‌‌ and interpret the results, but also physicists to‌ model the sensors, computer‌ scientists to develop algorithms‌‌ and mathematicians to model the resulting information. These‌ different expertises are combined‌ within the Morpheme team.‌‌ This generates a fecund frame for exchanging expertise,‌ knowledge, leading to an‌ optimal framework for the‌‌ different tasks (imaging, image analysis, classification, modeling). We‌ thus aim at providing‌ adapted and robust tools‌‌ required to describe, explain and model fundamental phenomena‌ underlying the morphogenesis of‌ cellular and supra-cellular biological‌‌ structures. Combining experimental manipulations, in vivo imaging, image‌ processing and computational modeling,‌ we plan to provide‌‌ methods for the quantitative analysis of the morphological‌ changes that occur during‌ development. This is of‌‌ key importance as the morphology and topology of‌ mesoscopic structures govern organ‌ and cell function. Alterations‌‌ in the genetic programs‌ underlying cellular morphogenesis have been linked to a‌ range of pathologies.

Biological questions we will focus‌ on include:

what are the parameters and the‌ factors controlling the establishment of ramified structures? (Are‌ they really organized to ensure maximal coverage? How‌ are genetic and physical constraints limiting their morphology?),‌
how are newly generated cells incorporated into reorganizing‌ tissues during development? (is the relative position of‌ cells governed by the lineage they belong to?)‌

Our goal is to characterize different populations or‌ development conditions based on the shape of cellular‌ and supra-cellular structures, e.g. micro-vascular networks, dendrite/axon networks,‌ tissues from 2D, 2D+t, 3D or 3D+t images‌ (obtained with confocal microscopy, video-microscopy, photon-microscopy or micro-tomography).‌ We plan to extract shapes or quantitative parameters‌ to characterize the morphometric properties of different samples.‌ On the one hand, we will propose numerical‌ and biological models explaining the temporal evolution of‌ the sample, and on the other hand, we‌ will statistically analyze shapes and complex structures to‌ identify relevant markers for classification purposes. This should‌ contribute to a better understanding of the development‌ of normal tissues but also to a characterization‌ at the supra-cellular scale of different pathologies such‌ as Alzheimer, cancer, diabetes, or the Fragile X‌ Syndrome. In this multidisciplinary context, several challenges have‌ to be faced. The expertise of biologists concerning‌ sample generation, as well as optimization of experimental‌ protocols and imaging conditions, is of course crucial.‌ However, the imaging protocols optimized for a qualitative‌ analysis may be sub-optimal for quantitative biology. Second,‌ sample imaging is only a first step, as‌ we need to extract quantitative information. Achieving quantitative‌ imaging remains an open issue in biology, and‌ requires close interactions between biologists, computer scientists and‌ applied mathematicians. On the one hand, experimental and‌ imaging protocols should integrate constraints from the downstream‌ computer-assisted analysis, yielding to a trade-off between qualitative‌ optimized and quantitative optimized protocols. On the other‌ hand, computer analysis should integrate constraints specific to‌ the biological problem, from acquisition to quantitative information‌ extraction. There is therefore a need of specificity‌ for embedding precise biological information for a given‌ task. Besides, a level of generality is also‌ desirable for addressing data from different teams acquired‌ with different protocols and/or sensors. The mathematical modeling‌ of the physics of the acquisition system will‌ yield higher performance reconstruction/restoration algorithms in terms of‌ accuracy. Therefore, physicists and computer scientists have to‌ work together. Quantitative information extraction also has to‌ deal with both the complexity of the structures‌ of interest (e.g., very dense network, small structure‌ detection in a volume, multiscale behavior, $...$ )‌ and the unavoidable defects of in vivo imaging‌ (artifacts, missing data, $...$ ). Incorporating biological expertise‌ in model-based segmentation methods provides the required specificity‌ while robustness gained from a methodological analysis increases‌ the generality. Finally, beyond image processing, we aim‌ at quantifying and then statistically analyzing shapes and‌ complex structures (e.g., neuronal or vascular networks), static or in evolution, taking‌ into account variability. In‌ this context, learning methods‌‌ will be developed for determining (dis)similarity measures between‌ two samples or for‌ determining directly a classification‌‌ rule using discriminative models, generative models, or hybrid‌ models. Besides, some metrics‌ for comparing, classifying and‌‌ characterizing objects under study are necessary. We will‌ construct such metrics for‌ biological structures such as‌‌ neuronal or vascular networks. Attention will be paid‌ to computational cost and‌ scalability of the developed‌‌ algorithms: biological experimentations generally yield huge data sets‌ resulting from high throughput‌ screenings. The research of‌‌ Morpheme will be developed along the following axes:‌

Imaging: this includes i)‌ definition of the studied‌‌ populations (experimental conditions) and preparation of samples, ii)‌ definition of relevant quantitative‌ characteristics and optimized acquisition‌‌ protocol (staining, imaging, $...$ ) for the specific‌ biological question, and iii)‌ reconstruction/restoration of native data‌‌ to improve the image readability and interpretation.
Feature‌ extraction: this consists in‌ detecting and delineating the‌‌ biological structures of interest from images. Embedding biological‌ properties in the algorithms‌ and models is a‌‌ key issue. Two main challenges are the variability,‌ both in shape and‌ scale, of biological structures‌‌ and the huge size of data sets. Following‌ features along time will‌ allow to address morphogenesis‌‌ and structure development.
Classification/Interpretation: considering a database of‌ images containing different populations,‌ we can infer the‌‌ parameters associated with a given model on each‌ dataset from which the‌ biological structure under study‌‌ has been extracted. We plan to define classification‌ schemes for characterizing the‌ different populations based either‌‌ on the model parameters, or on some specific‌ metric between the extracted‌ structures.
Modeling: two aspects‌‌ will be considered. This first one consists in‌ modeling biological phenomena such‌ as axon growing or‌‌ network topology in different contexts. One main advantage‌ of our team is‌ the possibility to use‌‌ the image information for calibrating and/or validating the‌ biological models. Calibration induces‌ parameter inference as a‌‌ main challenge. The second aspect consists in using‌ a prior based on‌ biological properties for extracting‌‌ relevant information from images. Here again, combining biology‌ and computer science expertise‌ is a key point.‌‌

4 Application domains

Among the applications addressed by‌ Morpheme team we can‌ cite:

Kidney cancer classification‌‌ from histological images
IMP-RNA (ribonucleicc acid) granules detection‌ and classification from confocal‌ image
Extra-cellular matrix detection‌‌ and characterization from confocal images
Axon growth modeling‌
Glial cell detection and‌ characterization for the study‌‌ of high-fat diets
Death and division time detection‌ and type classification of‌ cells in microscopy time-lapses‌‌
Plankton images analysis and classification
Morphogenesis and embryogenesis‌
Numerical super-resolution techniques
Convex‌ and non-convex sparse optimization‌‌ with applications to biomedical imaging
Statistical and learning-based‌ approaches for parameter selection‌ in imaging inverse problems‌‌
Physics-inspired machine learning for fluorescence microscopy

5 Highlights‌ of the year

5.1‌ Awards

Mohamad Mohamad obtained‌‌ the Best Student paper award in the 25th‌ Bioimaging Conference (Porto) for‌ his work "investigating Reinforcement‌‌ Learning for Histopathological Image‌ Analysis" (joint work with Xavier Descombes , Francesco‌ Ponzio ). He was also awarded the Prix‌ d’excellence of Université Côte d'Azur on December 11,‌ 2025 for this work.

5.2 New team

Morpheme‌ team will end during 2026. We have submitted‌ a new research proposal for the Morpheme team‌ renewal, updating our research axis.

6 Latest software‌ developments, platforms, open data

6.1 Latest software developments‌

6.1.1 Obj.MPP

Keywords:
Object detection, Marked Point Process,‌ Parametric model
Functional Description:
Obj.MPP implements the detection‌ of parametric objects using a Marked Point Process‌ (MPP). A parametric object is an n-dimensional piece‌ of signal defined by a finite set of‌ parameters. Detecting an object in a signal amounts‌ to finding a position at which the signal‌ can be described well enough by a specific‌ set of parameters (unknowns of the detection problem).‌ The detection task amounts to finding all such‌ objects. Typically, the signal is a 2-dimensional grayscale‌ image and the parametric objects are bright disks‌ on a dark background. In this case, each‌ object is defined by a single parameter: the‌ disk radius. Note however that the core function‌ of Obj.MPP is not tied to a particular‌ context (2-dimensional imaging is just an example).
News‌ of the Year:
The software was updated to‌ handle the 3 start methods of Python's multiprocessing‌ module in order to be able to use‌ parallel processing on all Python supported OS platform.‌
URL:
https://src.koda.cnrs.fr/eric.debreuve/obj.mpp/-/wikis/home
Publications:
hal-02193101, hal-01383165, hal-00801448‌, hal-00616371, hal-00793677
Contact:
Eric Debreuve

6.1.2‌ RCC-VascularMorphClassify

Name:
Renal Cell Carcinoma Classification from Vascular‌ Morphology
Keywords:
Machine learning, Cancer, Biomedical imaging
Functional‌ Description:
Our proposed two sets of hand-crafted features,‌ skeleton, and lattice features, which are extracted from‌ the vascular network segmentation images, can classify RCC‌ subtypes robustly.
URL:
https://github.com/medxiaorudan/RCC-VascularMorphClassify
Contact:
Xavier Descombes

6.1.3‌ Ascidian

Name:
Ascidian package
Keywords:
Embryogenesis, Morphogenesis
Scientific‌ Description:
This suite exploits the results issued from‌ the Astec suite (https://astec.gitlabpages.inria.fr/astec/) for image series of‌ developing ascidian embryos. It allows to name individual‌ cells (with the Conklin nomenclature), to assess the‌ segmentation issued from Astec, and to perform population‌ based studies.
Functional Description:
The Ascidian package aims‌ at exploiting results issued from the processing of‌ 3D+t sequences of developing embryos imaged by a‌ light-sheet microscope. The processing, done by the ASTEC‌ suite (https://astec.gitlabpages.inria.fr/astec/), results in a so-called property file,‌ which is the input of Ascidian package procedure.‌
URL:
https://astec.gitlabpages.inria.fr/ascidian/
Publications:
hal-04609482, hal-02903409
Contact:
Grégoire‌ Malandain
Participants:
Grégoire Malandain, Haydar Jammoul
Partner:
CRBM‌ - Centre de Recherche en Biologie cellulaire de‌ Montpellier

6.1.4 Astec

Name:
Adaptative Segmentation and Tracking‌ of Embryonic Cells
Keywords:
3D, 4D, Data fusion,‌ Image segmentation, Fluorescence microscopy, Morphogenesis, Embryogenesis
Scientific Description:‌
ASTEC stands for Adaptive Segmentation and Tracking of‌ Embryonic Cells, and was first developed during L.‌ Guignard PhD thesis, "Quantitative analysis of animal morphogenesis:‌ from high-throughput laser imaging to 4D virtual embryo‌ in ascidians, Léo Guignard, 2015". It was later published in "Contact area–dependent‌ cell communication and the‌ morphological invariance of ascidian‌‌ embryogenesis, Léo Guignard at al., Science 2020"
Functional‌ Description:
This software suite‌ aims at providing quantitative‌‌ analysis of multi-angle acquisitions of SPIM images, and‌ the segmentation of the‌ temporal series of 3D‌‌ images, together with quantitative informations.
URL:
https://astec.gitlabpages.inria.fr/astec/
Publications:‌
hal-02903409v1, hal-01938126v2,‌ tel-01278725v2, hal-00915000v2,‌‌ 2016AZUR4088
Contact:
Grégoire Malandain
Participants:
Patrick Lemaire, Leo‌ Guignard, Emmanuel Faure, Gaël‌ Michelin
Partners:
CRBM -‌‌ Centre de Recherche en Biologie cellulaire de Montpellier,‌ LIRMM

6.1.5 vt-python

Keywords:‌
Image analysis, Image filter,‌‌ Image registration, Registration of 2D and 3D multimodal‌ images, Image processing, Biomedical‌ imaging, Medical imaging
Scientific‌‌ Description:
Python interface for some functionalities of the‌ vt image processing library.‌
Functional Description:
Python interface‌‌ for some functionalities of the vt image processing‌ library.
Release Contributions:

**API,‌ memory, and core semantics**‌‌ Stabilized the vt–Python bridge by defining a clear‌ ownership model for NumPy/Image‌ conversions, separating view (borrow)‌‌ and move (transfer) semantics. Enabled robust zero-copy interoperability‌ based on vt >=‌ 1.7.x, while preserving an‌‌ explicit deep-copy option when required.

**Packaging, releases, and‌ quality** Structured releases (1.3.x–1.4.1),‌ migrated to a PEP‌‌ 517 / pyproject.toml build system, refined dependency constraints,‌ and hardened cross-platform conda‌ CI with expanded tests‌‌ focused on ownership semantics.
Contact:
Grégoire Malandain
Participants:‌
Manuel Petit, Jonathan Legrand‌
Partner:
Inria

6.1.6 vt‌‌

Keywords:
Image analysis, Image processing, Image registration, Registration‌ of 2D and 3D‌ multimodal images, Image filter,‌‌ Biomedical imaging, Medical imaging
Scientific Description:
2D and‌ 3D image processing library‌
Functional Description:
2D and‌‌ 3D image processing library
Release Contributions:

**API and‌ memory semantics** Stabilization of‌ the vtImageBridge through a‌‌ unified memory model: non-owning inputs (View) and owning‌ outputs (Move), ensuring safe‌ zero-copy interoperability and reliable‌‌ usage in downstream libraries (vt-python, timagetk)

**CI, tests,‌ and documentation** Improvements to‌ cross-platform CI, strengthened test‌‌ coverage, and clearer documentation, with better control over‌ dependencies and release processes.‌
Contact:
Grégoire Malandain
Participants:‌‌
Manuel Petit, Grégoire Malandain, Jonathan Legrand

6.1.7 FibreSight‌

Keywords:
Biomedical imaging, Biostatistics‌
Functional Description:
The Python‌‌ module aims to facilitate the study of fibrillar‌ proteins in the tumour‌ extracellular matrix using fluorescence‌‌ microscopy images. It includes a preprocessing step to‌ remove non-fibrillar aggregates, detects‌ and characterises fibres using‌‌ Gabor filters, and partitions the image into locally‌ homogeneous regions using graphs‌ and Voronoi diagrams. The‌‌ module computes statistics within these regions and proposes‌ an alignment index to‌ quantify fibre organisation.
Contact:‌‌
Faisal Jayousi

6.1.8 Mufasa

Name:
Fluorescence Fluctuations Simulation:‌ MUFASA Simulator
Keywords:
Blinking‌ simulation, SMLM, Super-resolution
Functional‌‌ Description:
The simulation software takes into account :‌ - different laser powers,‌ - different camera types,‌‌ blur and noise levels - different fluorophores, -‌ Multi-protocol support: Includes blinking‌ and fluorescence fluctuation protocols:‌‌ Fluorescence Fluctuations (FF), SMLM (STORM, PALM), Blinking. The‌ software models molecule transitions‌ using continuous-time Markov chains‌‌ (CTMC)
URL:
https://github.com/Joywessim/MUFASA_Fluorescence_Fluctuations_Simulation
Contact:
Wessim Omezzine

7 New‌ results

7.1 Detection, characterization,‌ and clustering of mouse‌‌ glial cells

Participants: Eric‌ Debreuve, Carole Rovère [IPMC, Sophia Antipolis],‌ Clara Sanchez [IPMC, Sophia Antipolis].

Overweight and‌ obesity are major public health issues affecting respectively‌ 39% and 13% of the world population (from‌ World Health Organization, 2016). They constitute prominent risks‌ for numerous chronic diseases, including diabetes, cardiovascular diseases,‌ and cancer. Studies in animal models and humans‌ reveal that excess fat diets promote both a‌ peripheral chronic inflammation and a hypothalamic neuroinflammation, which‌ possibly leads to feeding behavior deregulation. Ascertaining whether‌ the inhibition of early activation of two major‌ brain cells involved in feeding behavior (glial cells,‌ more specifically astrocytes and microglia) in the hypothalamus‌ could prevent obesity would offer new prospects for‌ therapeutic treatments. To understand the mechanisms pertaining to‌ obesity-related neuroinflammatory response, we developed a fully automated‌ algorithm, NutriMorph (see Figure 1). Although some‌ algorithms were developed in the past decade to‌ detect and segment neural cells, they are highly‌ specific, not fully automatic, and do not provide‌ the desired morphological analysis. Our algorithm cope with‌ these issues and performs the analysis of cells‌ images (here, microglia of the hypothalamic arcuate nucleus),‌ and the morphological clustering of these cells through‌ statistical analysis and machine learning. Using the k-Means‌ algorithm, it clusters the microglia of the control‌ condition (healthy mice) and the different states of‌ neuroinflammation induced by high-fat diets (obese mice) into‌ subpopulations. Here we show that early hypothalamic inflammation‌ could be already set on within a few‌ hours through modification of microglia subpopulation proportions, instead‌ of a couple of months as previously hypothesized‌ and that the activated microglia show specific morphological‌ characteristics. See 13.

Figure‌ 1:
Pipeline for microglia detection, analysis, and‌ clustering.

7.2 Identifying autofluorescence in biological samples using‌ hyperspectral imaging

Participants: Eric Debreuve, Sébastien Schaub‌ [IMEV, Villefranche-sur-mer], Jenifer Croce [IMEV, Villefranche-sur-mer].‌

Fluorescence imaging of marine samples (animals or plants)‌ remains a challenge due to the inevitable endogenous‌ fluorescence (or autofluorescence) common in these samples, for‌ example due to an animal ingesting algae exhibiting‌ endogenous fluorescence. The autofluorescence superimposes with the fluorescence‌ of the probes which are the target of‌ a specific study. The aim of this work‌ is to take advantage of recent improvements in‌ fluorescence imaging to identify and subtract sample autofluorescence‌ from probe fluorescence using hyperspectral imaging, i.e. using‌ so-called $Λ$ $λ$ acquisitions (confocal acquisitions varying both excitation and detection wavelengths).‌ This requires to identify‌ the various excitation and‌‌ emission spectra, and the corresponding concentration maps. A‌ first approach using blind‌ source separation by ICA‌‌ (Independent Component Analysis) has been developed. Initial results‌ are encouraging (see Figure‌ 2). The challenge‌‌ now is to improve the approach by imposing‌ constraints linked to the‌ physics of the problem,‌‌ notably the fact that the emission wavelength is‌ necessarily larger than the‌ excitation wavelength.

Figure 2: Separation of‌ two fluorophores (from left‌ to right and top‌‌ to bottom): Acquired concentration map, concentration maps identified‌ of each fluorophores, and‌ the corresponding excitation (solid‌‌ lines) and emission (dashed lines) spectra (blue for‌ one fluorophore, orange for‌ the other one).

7.3‌‌ Organoid phenotyping

This section is devoted to the‌ ANR project MORPHEUS dedicating‌ to the classification of‌‌ the different prostate organoids phenotypes. This project is‌ a collaboration with IPMC‌ (Stephan Clavel) and Metatox‌‌ (Xavier Coumoul). These two teams have provided the‌ data and their expertise‌ on organoids.

7.3.1 Knowledge‌‌ Distillation for Efficient 3D Segmentation on Fluorescence Images‌

Participants: Ivan Magistro Contenta‌, Xavier Descombes.‌‌

Cell segmentation consists of analyzing and identifying the‌ most relevant features of‌ biological image stacks. One‌‌ of the main state-of-the-art models for this task‌ is Cellpose, which produces‌ high-quality 3D segmentation masks.‌‌ However, it is less efficient on resource-constrained devices.‌ Furthermore, the scarcity of‌ labeled 3D image datasets‌‌ makes supervised training expensive and time-consuming preventing to‌ train a new lighter‌ network for a given‌‌ application. In this work, we present DistilledCellpose, a‌ lightweight version of Cellpose.‌ Our model was designed‌‌ to efficiently segment fluorescence images of organoids with‌ nuclear biomarker. We adopted‌ a lightweight model design‌‌ and knowledge distillation to reduce the model size‌ and the inference time,‌ while maintaining baseline performance.‌‌ As a result, DistilledCellpose is 56 × smaller‌ than Cellpose and even‌ lighter than FastCellpose, the‌‌ latest compressed variant. Our model matches baseline performance‌ on our dataset, consisting‌ of DAPI 3D images‌‌ of prostate organoids, and generalizes as well as‌ Cellpose on well-known benchmarks.‌ Our work focuses on‌‌ a set of high-quality confocal images of mice‌ prostate organoids. Each organoid‌ was imaged on the‌‌ 7th day. These samples were collected in two‌ laboratories (IPMC in Sophia‌ Antipolis and Metatox in‌‌ Paris. The image stacks were acquired using confocal‌ microscopy with two different‌ objectives: 20 times and‌‌ 40 times magnification. To‌ compare quantitively the models’ performances, we extracted some‌ 2D slices from selected datasets of both laboratories‌ and labeled nuclei using multi-point tool of Fiji.‌ The results are summarized in Tables 1 and‌ 2

Table 1: Mean values on 2D slices‌ with magnification 20× (17 slices)
Model	Precision (‌ $%$ )	Recall ( $%$ )	F1 score‌ ( $%$ )	Inference time (s)
Cellpose	89.02‌	85.32	86.11	13.54
Dist. FastCP	89.04	86.65	86.69‌	4.38
DistilledCellpose	90.25	85.00	85.87	3.52

Table 2:‌ Mean values on 2D slices with magnification 40×‌ (8 slices)
Model	Precision ( $%$ )	Recall‌ ( $%$ )	F1 score ( $%$ )‌	Inference time (s)
Cellpose	92.12	92.17	92.10	14.13‌
Dist. FastCP	93.16	92.95	93.01	4.47
DistilledCellpose	94.06‌	90.325	92.08	3.30

7.3.2 Organoid Image Classification Using‌ Deep Learning

Participants: Raffaele Martone, Xavier Descombes‌.

We have compared attention-based mechanisms with traditional‌ convolutional approaches for 3D organoid image classification. We‌ have systematically explored the effectiveness of transformer architectures‌ (SwinViT), hybrid models (SwinUnetr), and classical CNN architectures‌ (ResNet, DenseNet) across multiple experimental configurations, ultimately aiming‌ to determine optimal strategies for organoid classification tasks‌ by considering three classes (compact, cystic and cauliflower).‌ The compact samples have a spherical morphology with‌ two layers of cells whereas the cystic samples‌ have more layers and the cauliflowers exhibit a‌ shape with protrusions (see an example on Figure‌ 3). We have tested several input image‌ dimensions by applying subsampling, batch size, training modes‌ (from scratch or pre-trained), loss functions and classification‌ heads. Surprisingly, the best results, using a k-fold‌ cross-validation, were obtained with Resnet18 with an accuracy‌ on the test set of $94 %$ .‌ This result was obtained after image down sampling‌ and a cross-entropy loss function.

Figure 3‌: An example of a 3D oganoid dataset‌ (phenotype: cauliflower)

7.3.3 Graph Neural Network for organoid‌ classification

Participants: Alexandre Martin, Xavier Descombes.‌

Organoids—miniaturized, three-dimensional in vitro cultures that replicate the‌ complexity of human tissues—are revolutionizing biomedical research. Yet‌ their analysis remains heavily reliant on manual methods‌ that are time-consuming, low-throughput, and prone to interpretative‌ bias. These structures, composed of cells organized into‌ spatial and functional interaction networks, demand analytical tools‌ capable of capturing not only their morphology but also the cellular relationships‌ that govern their behavior.‌ In this context, Graph‌‌ Neural Networks (GNNs) emerge as a particularly well-suited‌ solution, enabling organoids to‌ be modeled not as‌‌ static images but as relational systems, where each‌ cell is a node‌ connected to its neighbors‌‌ via edges representing biological interactions. This work introduces‌ an innovative framework for‌ the automated modeling and‌‌ classification of organoids using cellular graphs, fully leveraging‌ the potential of GNNs.‌ Unlike conventional approaches—based on‌‌ manual descriptors or convolutional neural networks (CNNs), which‌ analyze images pixel-by-pixel—GNNs integrate‌ structural and contextual information‌‌ by representing each organoid as a network. In‌ this framework, nodes encode‌ cellular properties (e.g., size,‌‌ shape, marker expression) while edges capture spatial relationships.‌ This relational representation enables‌ finer and more interpretable‌‌ classification, capable of distinguishing subtle phenotypes—such as early‌ differentiation stages or pathological‌ alterations—that elude traditional methods.‌‌ To address challenges posed by limited annotated data‌ and the intrinsic variability‌ of organoids, this work‌‌ develops a comprehensive pipeline, from constructing cellular graphs‌ from microscopy images to‌ robust GNN training. Particular‌‌ emphasis is placed on synthetic data generation via‌ graph generative models to‌ augment training sets and‌‌ explore rare or extreme scenarios. The applications of‌ this approach are far-reaching:‌ high-throughput drug screening, early‌‌ disease diagnosis from patient-derived organoids, and optimization of‌ culture protocols to standardize‌ organoid production. In the‌‌ long term, this work lays the groundwork for‌ holistic multi-modal analysis—integrating imaging,‌ cellular graphs, and omics‌‌ data—to deepen our understanding of underlying biological mechanisms‌ and advance precision medicine.‌

A comparison of the‌‌ obtained results with different architectures is given in‌ Table 3. The‌ best results are obtained‌‌ with GAT, showing the contribution of the attention‌ mecanism. Final classification results‌ are summarized in Table‌‌ 4.

Table 3: Mean Square Error on‌ a test set contining‌ 15 000 organoïds
Modèle‌‌	MSE	Nbr of Params
GCN	0.198	250K
DeepSets‌	0.145	280K
EGNN	0.137‌	800K
GAT	0.118	320K‌‌

Table 4: Results with GAT pre-trained on synthetical‌ data and fine-tuned with‌ real data (overall accuracu‌‌ : 84 $%$ )
Phenotype	Precision	Recall	F1-score‌	Support
Cauliflower	0.93	0.74‌	0.82	38
Cystic	0.78‌‌	0.95	0.85	37
Total mean	0.86	0.84	0.84‌	75

7.4 Computational histopathology‌

This section describes the‌‌ work within a collaboration with Nice CHU concerning‌ kidney and Bichat Hospital‌ (Paris) concerning lung cancer.‌‌ The AI developements result from a collaboration with‌ Polito (Torino, Italy).

7.4.1‌ Toward a numerical BANFF‌‌ for renal Histology

Participants: Meryem Sikouky, Xavier‌ Descombes, Francesco Ponzio‌, Damien Ambrosetti [CHU,‌‌ Nice], Giorgio Toni [CHU, Nice], Paul‌ Hannetel [CHU, Nice].‌

In recent years, deep‌‌ learning has improved the instance segmentation of histology‌ images. However, existing instance‌ segmentation networks often struggle‌‌ to accurately capture the intricate geometry and topology‌ of tubular structures. Conventional‌ methods that rely solely‌‌ on semantic approaches, boundary or distance maps remain‌ susceptible to the merging‌ of adjacent instances or‌‌ breakings in connectivity. In‌ this work, we introduced a geometry-aware multi-task deep‌ network that concurrently predicts semantic probability and an‌ energy map that encodes both skeleton and boundary‌ information. The proposed multitask framework, as shown in‌ Figure 4, utilizes distance-based geometric supervision to‌ incorporate structural priors, thereby enhancing topological continuity while‌ maintaining object separation. During inference, instances are recovered‌ automatically via a module that we call cross-talk.‌ When tested on renal tubule histology data, the‌ proposed approach demonstrates superior performance compared to state-of-the-art‌ deep learning models based on panoptic-style and boundary-sensitive‌ metrics, with minimal architectural complexity. This will be‌ presented at ISBI 2026 under the title: Shape-Aware‌ Multi-task Instance Segmentation for tubules in renal Histology.‌

Figure 4.a — Figure 4: Overview of the proposed architecture.‌ The network jointly learns semantic and geometric representations,‌ which are fused to guide instance prediction. Each‌ decoder head is supervised with its dedicated loss,‌ while skeletal and boundary consistency terms reinforce alignment‌ between semantic and geometric cues. The resulting maps‌ are concatenated and passed through a lightweight cross-talk‌ neck that produces the final instance segmentation mask‌ (top) and visual examples of tubule instance segmentation.‌ Black circles indicate under-segmentation, white indicate over-segmentation, red‌ mark false negatives, and yellow highlight semantically incomplete‌ structures (bottom).

Figure 4.b — Figure 4: Overview of the proposed architecture.‌ The network jointly learns semantic and geometric representations,‌ which are fused to guide instance prediction. Each‌ decoder head is supervised with its dedicated loss,‌ while skeletal and boundary consistency terms reinforce alignment‌ between semantic and geometric cues. The resulting maps‌ are concatenated and passed through a lightweight cross-talk‌ neck that produces the final instance segmentation mask‌ (top) and visual examples of tubule instance segmentation.‌ Black circles indicate under-segmentation, white indicate over-segmentation, red‌ mark false negatives, and yellow highlight semantically incomplete‌ structures (bottom).

We are currently developing a two-stage‌ fusion network. In the first stage, four specialized‌ Attention-UNets are trained to detect glomeruli, tubules, vessels,‌ and peritubular capillaries (PTC). These models are trained‌ on patches sampled at appropriate magnifications for each‌ structure (5x for glomeruli and vessels, 10x for‌ tubules and 40x for PTC) and fine-tuned with‌ structure-specific hyperparameters.

Initially, we used simple union and‌ rule-based conflict resolution to merge predictions. However, this approach could not recover‌ missing detections or correct‌ structural overlaps. Thus, we‌‌ are transitioning to a more advanced fusion model‌ incorporating uncertainty maps. Each‌ expert now generates both‌‌ a segmentation map and an uncertainty score. These‌ are input to a‌ transformer-inspired module using cross-attention‌‌ between features and uncertainty weights. As a matter‌ of fact, in a‌ Vision Transformer, each layer‌‌ relies on scaled dot-product attention to fuse contextual‌ information. This mechanism assigns‌ larger weights to patches‌‌ whose features are most relevant to the current‌ query, capturing long-range dependencies‌ that are critical in‌‌ histological imagery.

7.4.2 Quantification of immunohistochimical slices

Participants:‌ Sheyenne Nguyen, Xavier‌ Descombes, Damien Ambrosetti‌‌ [CHU, Nice], Paul Hannetel [CHU, Nice].‌

We address the problem‌ of renal carcinoma classification,‌‌ a disease comprising several tumor subtypes that are‌ difficult to identify and‌ characterize through morphology alone.‌‌ To offset these limitations, pathologists typically use a‌ dual-stain strategy: a broad‌ structural overview with hematoxylin-eosin‌‌ (H&E) and a more specific, chemistry-based approach with‌ immunohistochemistry (IHC). Although visual‌ assessment of histopathology is‌‌ indispensable, it remains complex and subject to substantial‌ inter- and intra-observer variability.‌ Our goal is therefore‌‌ to automate IHC quantification, delivering time savings, greater‌ robustness, and enhanced diagnostic‌ reliability all of which‌‌ can improve patient care. First, we have tailored‌ the VGG16 model to‌ detect tumor regions in‌‌ IHC slides. We then have developed a pipeline‌ that isolates the chromogen‌ associated with the considered‌‌ IHC staining on non-tumor areas and classifies tumor‌ tissue areas according to‌ their staining response in‌‌ four classes (negative answer and three levels of‌ positivity). The resulting performance,‌ empirically evaluated by a‌‌ medical expert, is promising: it demonstrates feasibility for‌ cytoplasmic markers and sets‌ the stage for adaptation‌‌ to other IHC targets. Overall, the tools developed‌ here offer strong potential‌ for reproducible, quantitative analysis‌‌ of diverse immunohistochemical markers (see Figure 5).‌

Figure 5: Colorimetric analysis (a) Reference‌ color estimation on non-tumor‌ area (b) Quantification on‌‌ tumor areas.

7.4.3 Domain transfer in histopatholgy

Participants:‌ Imen Chtourou, Xavier‌ Descombes, Damien Ambrosetti‌‌ [CHU, Nice], Giorgio Toni [CHU, Nice].‌

Hematoxylin and eosin (H&E)‌ staining is the most‌‌ widely used technique in histopathology, as it provides‌ a comprehensive overview of‌ tissue morphology. However, special‌‌ stains such as Periodic acid–Schiff (PAS) play a‌ crucial role in clinical‌ diagnosis by highlighting specific‌‌ histological structures, including basement‌ membranes and glycogen, which are particularly relevant in‌ renal pathology. In routine practice, acquiring multiple stains‌ requires additional tissue sections and increases processing time‌ and costs. Recent advances in deep learning, particularly‌ in generative models such as generative adversarial networks‌ (GANs) and diffusion models, have enabled virtual stain-to-stain‌ translation, allowing PAS-like images to be synthesized directly‌ from H&E slides. These models learn complex non-linear‌ mappings between staining domains and have demonstrated promising‌ results in generating visually realistic and diagnostically relevant‌ PAS images. Despite this progress, H&E-to-PAS translation remains‌ a challenging task due to inter-dataset variability, differences‌ in tissue preparation protocols, and staining heterogeneity. We‌ have developed a novel approach, illustrated in Figure‌ 6, based on a dedicated preprocessing step‌ applied prior to the training of a diffusion‌ model. This preprocessing is designed to mitigate domain‌ shifts and facilitate the learning of robust and‌ transferable stain mappings. We have evaluated our approach‌ on two distinct datasets from CHU Nice :‌ (i) H&E whole-slide images (WSIs) from cancer tissue‌ samples, and (ii) H&E WSIs from renal biopsy‌ samples. In both cases, PAS patches extracted from‌ diabetic patients are used as the target domain.‌ Several preprocessing strategies to normalize the images' color‌ have been investigated. Figure 7 illustrates the preliminary‌ results demonstrating color transfer from H&E to PAS.‌

Figure 6:‌ General Workflow of the proposed approach.

Figure 7‌: Examples of H&E (top line) to PAS‌ (bottom line) Color Transfer.

7.4.4 Transformers for kidney‌ cancer subtype classification

Participants: Moncef Belaskry, Xavier‌ Descombes, Mohamed Lamine Benomar [Tlemcen University],‌ Damien Ambrosetti [CHU, Nice].

In this study,‌ we developed an renal cell carcinoma (RCC) histopathology‌ classification pipeline based on 224×224 RGB patches extracted‌ from whole-slide images, with strict patient-level partitioning to‌ prevent data leakage. To mitigate inter-site staining variability‌ caused by differences in scanners, protocols, and laboratory conditions, we applied a‌ stain normalization, which standardizes‌ color appearance by matching‌‌ each patch’s channel-wise mean and standard deviation to‌ those of a fixed‌ reference. Importantly, the same‌‌ reference statistics were reused consistently during both model‌ development and inference to‌ ensure a stable and‌‌ reproducible preprocessing workflow. The proposed framework employs a‌ dual-branch Vision Transformer (ViT)‌ architecture to capture complementary‌‌ multi-scale representations. The ViT-B/16 branch operates on 16×16‌ patch embeddings, producing a‌ denser token sequence that‌‌ increases sensitivity to fine-grained local morphological cues. In‌ contrast, the ViT-B/32 branch‌ processes 32×32 patch embeddings,‌‌ yielding a more compact tokenization that emphasizes global‌ tissue organization and architectural‌ context. To integrate these‌‌ heterogeneous feature streams, we introduce a bidirectional cross-attention‌ fusion module that enables‌ mutual information exchange, local‌‌ features are contextualized by global representations, while global‌ descriptors are refined using‌ discriminative local evidence. This‌‌ learned cross-scale fusion provides a principled alternative to‌ naive aggregation strategies. Model‌ optimization leveraged transfer learning‌‌ from ImageNet-pretrained ViTs, together with data augmentation to‌ improve robustness and class-weighted‌ training to address subtype‌‌ imbalance. Crucially, we evaluated the approach not only‌ on internal splits but‌ also on multiple independent‌‌ multi-center datasets comprising large numbers of images acquired‌ under heterogeneous conditions that‌ were never observed during‌‌ training, providing a stringent assessment of generalization under‌ domain shift. Performance was‌ quantified using accuracy, precision,‌‌ recall, and F1-score, and further analyzed via confusion‌ matrices to characterize subtype-specific‌ error patterns and inter-class‌‌ confusions. Figure 8 summarizes the model architecture, Grad-Rollout‌ visual explanations, and multi-center‌ confusion-matrix evaluation.

Figure 8‌‌: Overview of the proposed dual-branch ViT fusion‌ approach (top), confusion matrices‌ on RCC subtype classification‌‌ (bottom).

7.4.5 Reinforcement Learning in histopathology

Participants: Mohamad‌ Mohamad, Xavier Descombes‌, Francesco Ponzio,‌‌ Nicolas Pote [Hôpital Bichat, Paris], Maxime Gassier‌ [Hôpital Bichat, Paris].‌

We first focused on‌‌ transferring an agent previously trained for WSI (Whole‌ Slide Images)localization to a‌ real-world case study in‌‌ tumor analysis. The objective is to train an‌ agent capable of selecting‌ tumor regions patch by‌‌ patch, across multiple magnification levels, while minimizing the‌ time required for exploration.‌ This work involved defining‌‌ the problem formulation, specifying the model inputs, designing‌ the reward signal, and‌ formalizing the dynamics of‌‌ a tumor-segmentation environment. In‌ addition, we explored the associated challenges and collected‌ additional data to improve the generalization of the‌ agent’s learned behavior to unseen patients. Some results‌ are shown on Figure 9.

Figure 9: This figure illustrates the desired‌ behavior and the skills effectively learned by the‌ tumor segmentation agent. The agent starts from the‌ top-left image with no prior segmentation and progressively‌ segments the tumor region patch by patch. The‌ bottom-left image shows the ground truth segmentation.

We‌ then addressed a fundamental issue arising from the‌ use of Batch Normalization layers within reinforcement learning‌ agents. These architectures initially caused training instability. During‌ this year, we identified the underlying cause of‌ this behavior and demonstrated that the issue generalizes‌ across different applications, not limited to histopathology. We‌ also derived a practical solution that allows effective‌ use of such normalization layers without compromising training‌ stability. This solution is currently being benchmarked, and‌ a paper describing this work is in preparation.‌

7.5 Spatial transcriptomics

7.5.1 Estimation of the subcellular‌ distribution of RNA molecules at the population level‌ using optimal transport

Participants: Morgane Fierville, Xavier‌ Descombes, Pascal Barbry [IPMC, Sophia Antipolis],‌ Kevin Lebrigand [IPMC, Sophia Antipolis].

Spatial transcriptomics‌ enables the mapping of gene expression within tissues‌ at subcellular resolution. This recent technology provides direct‌ access to the detection of RNA molecules in‌ individual cells, allowing the investigation of localized expression‌ mechanisms. The subcellular localization of RNA molecules aims‌ to elucidate how they are distributed and expressed‌ in specific regions of the cell, as well‌ as the particular modifications that may be observed‌ depending on cell type and cellular state (healthy‌ or pathological). However, despite technological advances, the number‌ of RNA molecules detected per cell remains limited,‌ thereby constraining the accuracy of subcellular localization for‌ genes of interest. In this context, we propose‌ an innovative method to enhance the understanding of‌ specific localizations by aggregating information from multiple cells‌ belonging to the same cell type (see Figure‌ 10). Our approach is based on optimal‌ transport, and more specifically on the Fused Gromov–Wasserstein‌ (FGW) distance. This approach enables the representation of‌ cell geometry through their external shape defined by‌ the cell membrane, together with subcellular structures defined by the nuclear membrane.‌ This strategy is inspired‌ by the recent work‌‌ by Govek et al. 35, who leverage‌ Gromov–Wasserstein optimal transport to‌ align and classify neuronal‌‌ morphologies. We extend this paradigm to spatial transcriptomics‌ by adapting the algorithm‌ to account for biological‌‌ constraints specific to subcellular data. In our method,‌ each cell is represented‌ by a distance matrix‌‌ between descriptive points outlining the geometry of the‌ cell membrane and the‌ nuclear membrane. To preserve‌‌ biological consistency during alignment, a labeling scheme enforces‌ that points on the‌ cell membrane are transported‌‌ only to other cell membrane points, and likewise‌ for nuclear membrane points,‌ thereby ensuring strict adherence‌‌ to the cell–nucleus correspondence. The use of FGW‌ makes it possible to‌ simultaneously integrate these structural‌‌ constraints and local features by identifying an optimal‌ transport plan between a‌ source cell and a‌‌ target cell.

Figure 10: a)‌‌ Computation of optimal transport between cells sharing the‌ same cellular state using‌ the FGW distance. Visualization‌‌ of transcripts of gene 1 within the cells.‌ b) Extraction of the‌ medoid shape based on‌‌ the distance matrix. c) Aggregation of transcripts.

7.5.2‌ High-resolution 3D spatial transcriptomics‌ of Drosophila Kenyon cells‌‌

Participants: Fabienne De Graeve, Florence Besse.‌

This preliminary work has‌ motivated the RNALOC project,‌‌ which has been accepted for funding by ANR.‌ The project kick-off has‌ been scheduled in early‌‌ January 2026.

We are interested in the molecular‌ mechanisms underlying memory, using‌ gamma neurons in the‌‌ mushroom body of Drosophila as a model. The‌ results accumulated to date‌ have highlighted: i) the‌‌ existence of RNA compartmentalization along axons, ii) the‌ importance of neuronal activity‌ in this process, and‌‌ iii) the importance of the Imp protein in‌ the transport of a‌ subpopulation of mRNAs 36‌‌. The small number of mRNAs visualized to‌ date does not allow‌ us to obtain an‌‌ overall view of the molecular composition of the‌ different compartments along the‌ axons and their remodeling‌‌ in response to the activity of afferent neurons.‌ To answer these questions,‌ we are developing a‌‌ high-resolution 3D spatial transcriptomics protocol for stimulated or‌ inhibited gamma neurons. Two‌ spatial transcriptomics methods have‌‌ caught our attention. The MERFISH method (Vizgen, Boston,‌ USA) is the most‌ sensitive approach for cryosections‌‌ of fixed tissue (see‌ Figure 11). This preliminary experiment confirms that‌ the MERFISH approach could meet our expectations. The‌ seq-FISH method (EMBL, Heidelberg, Germany) would allow us‌ to avoid the steps of cryosectioning and 3D‌ reconstruction of the volume occupied by the axons‌ of mushroom neurons, as it would be performed‌ on whole brains.

Figure 11: Results of the test experiment‌ using the MERFISH approach on frozen sections of‌ adult Drosophila. A: Histological section passing through 15‌ Drosophila brains analyzed using the MERFISH method. Scale‌ bar 1 mm. B: Zoom on a section‌ passing through the median lobes (red and blue‌ outlines) formed by the axons of gamma neurons‌ (green). Each colored spot (white arrow) reveals the‌ presence of an mRNA. Scale bar 100 µm.‌

7.6 From Photon Emission to Super-Resolution: The MUFASA‌ Simulator

Participants: Wessim Omezzine, Laure Blanc-Féraud,‌ Luca Calatroni, Sébastien Schaub.

We present‌ MUFASA (Multi-Protocol Unified Fluorescence-based‌ Advanced Simulation Algorithm), a physically‌ grounded, continuous-time simulator for super-resolution fluorescence microscopy. By‌ modeling fluorophore dynamics using continuous-time Markov chains, MUFASA'‌ simulation features yield realistic photon emission behavior across‌ both Single Molecule Localization Microscopy (SMLM) and fluorescence‌ fluctuation-based (FF-SRM) protocols, independently of frame duration and‌ sampling. The framework supports both individual emitters and‌ structure-level simulations, incorporating photophysical transitions, photobleaching, and camera‌ properties.

To quantitatively validate simulations with real data,‌ we introduce a novel validation metric based on‌ the 1-Wasserstein distance between simulated and experimental photon-count‌ distributions. In addition to simulation, another functionality estimates‌ key photophysical parameters (e.g., molar extinction coefficient) and‌ to suggest optimal light-source power ranges from fluctuation‌ data. An intuitive Python-based graphical interface enables real-time‌ parameter tuning, visualization, and TIFF export. Designed for‌ biologists, physicists, microscopists, and numerical imaging engineers, MUFASA‌ offers a practical platform for microscopy experiment design,‌ hypothesis testing and the generation of realistic training‌ data for data-driven microscopy methods across modalities (see‌ Fig. 12).

Figure 12: Pipeline‌ of the MUFASA simulator.‌‌

7.7 Off-the-grid dynamic super-resolution for fluorescence microscopy

Participants:‌ Aneva Tsafack, Laure‌ Blanc-Féraud, Gilles Aubert‌‌.

We introduce a new off-the-grid variational framework‌ for reconstructing curves (one-dimensional‌ geometric objects, such as‌‌ filaments) from blurry and noisy images. Such objects‌ are naturally modeled by‌ Radon measures supported on‌‌ curves. A key theoretical contribution is a new‌ Smirnov-type decomposition theorem in‌ a space $S$ of‌‌ simple regular curves. It states that for every‌ $μ \in 𝒱 (‌ ℝ^{2})$ ,‌‌ the space of two-dimensional finite Radon measures with‌ finite divergence, there exists‌ a positive Radon measure‌‌ $σ \in ℳ^{+} (S)$ such‌ that

μ = {\int﻿﻿﻿‌}_{S} μ_{γ} d﻿‌​‌ σ (γ)﻿​​﻿, | μ |​​​‌ = \int_{S} |﻿﻿﻿‌ μ_{γ} | d﻿‌​‌ σ (γ)﻿​​﻿ .

This result allows‌ us to define a‌ physically consistent forward model‌‌ for blurred and noisy images: $O = D‌ (σ) +‌ η$ , where the‌‌ forward operator $D$ is defined by $D (‌ σ) : =‌ \int_{S} (|‌‌ μ_{γ} | * h) d σ‌ (γ),‌$ with $h$ a given‌‌ blur kernel. The associated inverse problem consists of‌ recovering the curves by‌ minimizing a new convex‌‌ functional, termed Curve LASSO (CLASSO):

ℰ (σ​​​‌) = \frac{1}{2﻿﻿﻿‌} {∥ O - D﻿‌​‌ (σ) ∥﻿​​﻿}_{2}^{2} + λ​​​‌ {∥ σ ∥}_{TV﻿﻿﻿‌} .

Minimizers of CLASSO‌‌ are shown to be finite combinations of Dirac‌ measures supported on curves‌ in $S$ , thereby‌‌ promoting sparse solutions. For numerical implementation, we use‌ a Sliding Frank-Wolfe algorithm,‌ which iteratively reconstructs the‌‌ solution curves. An illustrative example of the reconstructed‌ curves is shown in‌ Figure 13.

Figure 13.a — Figure‌ 13: Illustrative results of curve reconstruction using‌ the CLASSO framework. Left: input blurry and noisy‌ image; right: reconstruction using the Sliding Frank-Wolfe (SFW)‌ algorithm (red = reconstruction, blue = ground-truth).

Figure 13.b — Figure‌ 13: Illustrative results of curve reconstruction using‌ the CLASSO framework. Left: input blurry and noisy‌ image; right: reconstruction using the Sliding Frank-Wolfe (SFW)‌ algorithm (red = reconstruction, blue = ground-truth).

7.8‌ Embryogenesis

7.8.1 Motion Compensation in Multiview Light Sheet‌ Microscopy Temporal Series

Participants: Grégoire Malandain, Haydar‌ Jammoul, Kilian Biasuz [CRBM, Montpellier], Patrick‌ Lemaire [CRBM, Montpellier].

In developmental biology, the‌ study of growing organisms at cell level for‌ the understanding of morphogenesis is required to decipher‌ the underlying genetic mechanisms that govern development. Latest‌ microscopy techniques allow to acquire temporal series of‌ 3D images with both good spatial resolution and‌ imaging frequency, enabling to follow the development over‌ a long period of time. To that end,‌ it is crucial to have fast microscopy techniques,‌ to minimize both the sample deformation (due to‌ development) during the acquisition time and the phototoxicity‌ to ensure a normal development. Light sheet imaging‌ (or selective plane illumination microscopy (SPIM)) achieved both‌ goals. However, the image quality may be impaired‌ by poor sample transparency. Multiview lighsheet imaging (MuViSPIM)‌ addresses this drawback by providing images from 4‌ points of view. At each time point, a‌ 3D image is constructed from four acquisitions: the‌ first two are acquired simultaneously by two opposite‌ cameras, and the next two are acquired by‌ the same two cameras after a rotation of‌ 90 degrees of the stage (and thus of‌ the sample). Then, the stage rotates back to‌ its initial position for the next time point‌ acquisition.

Hence, an image is issued from the‌ fusion of the 4 acquisitions: it requires to‌ first co-register the 4 acquisitions (one of them‌ being considered as the reference), and then to‌ combine them, e.g. by a weighted linear combination,‌ the weights being calculated to emphasize acquired data‌ close to the cameras 5. Additionally, when‌ processing the temporal series, cell tracking (or lineage‌ calculation) is eased by co-registering couples of successive‌ images 5. These latter transformations can also‌ be combined to re-compute an artificially stabilized sample,‌ thus facilitating the visual inspection of the data.‌

In some series, we noticed that the imaged‌ embryo may undergo large rotation angles, either between‌ the two stage positions of a time point‌ or between two successive time points (when the‌ stage returns to its initial position), and this‌ may jeopardize either the fusion or the cell‌ tracking. Because of the efforts required to acquire‌ such data, it is crucial to find a‌ way to exploit them.

We proposed a strategy‌ to handle such large rotations. Instead of reconstructing‌ each time point independently from the 4 acquisitions,‌ stabilized temporal series for each stage position are‌ first reconstructed. To do so, the couples of mis-registered successive images are‌ identified thanks to a‌ dedicated merit function. A‌‌ dedicated registration strategy involving multiple initial positions is‌ designed to address these‌ mis-registrations. Last, it is‌‌ sufficient to co-register one couple of corresponding time‌ point fusions from the‌ two stabilized series to‌‌ get an estimation of the relative position of‌ the four acquisitions for‌ any time point, which,‌‌ in turn, allows to compute stabilized fused images‌ for the whole series.‌

This work has been‌‌ accpted for publication at ISBI 2026.

7.8.2‌ Surface preserving resampling of‌ labeled images

Participants: Grégoire‌‌ Malandain, Patrick Lemaire [CRBM, Montpellier].

Thanks‌ to the development of‌ microscopy techniques, the acquisition‌‌ of temporal series of 3D images is becoming‌ a standard for the‌ study of evolving phenomena‌‌ (e.g. developmental biology). In most cases, live samples‌ are moving/growing during long-term‌ imaging, therefore it is‌‌ desirable to compensate for this global 3D motion‌ for both a more‌ comfortable visualization and analysis.‌‌ It is implicitly assumed that the quantitative properties‌ of the resampled series‌ are similar (if not‌‌ equal) to those of the original one. Among‌ these properties, the surface‌ measurement is quite important‌‌ since it helps predicting cell behavior, fate, or‌ to compute symmetry axis.‌

We demonstrated that the‌‌ surface estimation of segmented objects may not be‌ preserved by the nearest‌ neighbor interpolation, the usual‌‌ technique used to resample labeled images, and demonstrated‌ that the gaussian based‌ interpolation, dedicated for multi-labeled‌‌ images, preserves the surface estimation, with relative errors‌ of the order of‌ 1 % for different‌‌ surface estimation methods 19.

7.8.3 Predicting cell‌ division orientation in ascidian‌ development

Participants: Haydar Jammoul‌‌, Grégoire Malandain, Kilian Biasuz [CRBM, Montpellier]‌, Benjamin Gallean [CRBM,‌ Montpellier], Patrick Lemaire‌‌ [CRBM, Montpellier].

The ascidian embryo exhibits a‌ highly reproducible pattern where‌ homologous cells can be‌‌ identified across different embryos, allowing consistent cell naming.‌ This developmental stereotypy of‌ the embryo depends strongly‌‌ on the orientation of cell divisions. To better‌ understand the embryonic organization,‌ identifying the cues that‌‌ control division orientation by attempting to predict it‌ is of great interest.‌ Before the 64-cell stage,‌‌ when the embryo still has a spherical shape,‌ the longest axis of‌ a cell's apical surface‌‌ during interphase (around 20 minutes before cell division)‌ predicts its division orientation‌ (Hertwig's rule). After the‌‌ 112-cell stage, testing this rule becomes more difficult‌ due to the local‌ tissue deformations occurring between‌‌ a cell's interphase and its division. We therefore‌ proposed a local registration‌ of the cell neighborhood‌‌ to compensate for these deformation. This allows to‌ compare the axis of‌ the apical surface at‌‌ interphase with the division direction. We therefore can‌ systematically test the Hertwig‌ rule in 3D+t ascidian‌‌ embryos between the 64- and 300-cell stages.

Results‌ are somewhat mixed:

for‌ 55 divisions (46%), the‌‌ division orientation could be predicted from the apical‌ surface longest axis (see‌ Figure 14 top),
for‌‌ 13 divisions (11%), the‌ division orientation was not aligned with the interphasic‌ apical surface longest axis (see Figure 14 bottom),‌
for the remaining 51 divisions (43%), either the‌ result were inconclusive (directions were neither aligned nor‌ orthogonal) or the apical surface was not well-defined.‌

Our results suggest that some neural plate cells‌ division orientation may follow non-geometric cues, and some‌ cell divisions exhibiting two distinct orientations may arise‌ from apical surface geometry. This quantitative analysis contributes‌ to our understanding of the factors controlling cell‌ division orientation.

Figure 14:
Top left: a7.16* cell in‌ Phmamm-1 during interphase, with the red line indicating‌ its longest apical surface axis and its neighboring‌ cells. Top right: the daughter cells of a7.16*‌ after division, with the red line showing the‌ division orientation, which follows the interphase longest apical‌ axis. Bottom left: b7.13* cell in Phmamm-1 during‌ interphase, with the red line indicating its longest‌ apical surface axis and its neighboring cells. Bottom‌ right: the daughter cells of b7.13* after division,‌ with the red line showing the division orientation,‌ which does not follow the interphase longest apical‌ axis.

7.8.4 Blastoderm apical cell shape in Drosophila‌ melanogaster embryo

Participants: Ines Landolsi, Grégoire Malandain‌, Barthélémy Delorme [IBV, Nice], Matteo Rauzi‌ [IBV, Nice].

The embryo of Drosophila melanogaster‌ starts its development with a syncytial blastoderm and‌ then undergoes a step of cellularization where all‌ the membranes of its blastoderm are created at‌ the same time around each nucleus. This development‌ suggest that the position of nuclei may highly‌ influence polygonal like apical cell shape of the‌ blastoderm.

This study first implies to be aware‌ of the apical cell shape of the different‌ tissues of the blastoderm, which is not described‌ in the literature. Our first results show that‌ around 50% of apical cell shape are hexagon-like,‌ around 20-25% pentagon and heptagon-like shape and lastly‌ around 5% are tetragon and octagon-like shape (Fig.‌ 15). These different shapes appear to be‌ evenly spread around the embryo. Next steps include‌ to test whether the cell membrane position solely‌ depends on the nuclei position.

Figure 15:
Left:‌ polygon-like apical cell shape‌‌ distribution in blastoderm; right: cell shape distribution of‌ blastoderm cells around Drosophila‌ embryo.

7.9 Quantifying ligno-cellulosic‌‌ enzymatic deconstruction

Participants: Anna Kestel, Grégoire Malandain‌, Yassin Refahi [INRAE,‌ Reims], Gabriel Paës‌‌ [INRAE, Reims].

In the framework of the‌ FillingGaps targeted project of‌ the PEPR B-Best, we‌‌ aim to quantify the enzymatic hydrolysis of maize‌ stems, a biomass with‌ heterogeneous tissues. For this,‌‌ temporal series of maize samples undergoing deconstruction are‌ imaged by confocal microscopy,‌ the cell wall autofluorescence‌‌ intensity providing a proxy of the cell wall‌ deconstruction.

The motion compensation‌ of the temporal followed‌‌ by the cell segmentation allowed us to quantify‌ the cell wall fluorescence,‌ thus the deconstruction. Figure‌‌ 16 presents the average autofluorescence intensity dynamics of‌ cell walls in processed‌ images for two regions‌‌ (the parenchyma in the pith (PM) and the‌ parenchyma below the rind‌ (PE)) for different experimental‌‌ conditions. As expected, the dynamics differ between the‌ two regions, with a‌ higher intensity reduction in‌‌ the PM region than in the PE region‌ for pretreated samples, while‌ the opposite trend is‌‌ observed for raw samples.

Figure‌‌ 16:
Maize cell walls autofluorescence average values,‌ for both the parenchyma‌ in the pith (PM)‌‌ and the parenchyma below the rind (PE), and‌ for different experimental conditions.‌

7.10 Deciphering the axonal‌‌ regeneration

Participants: Caroline Medioni, Grégoire Malandain.‌

We are interested in‌ deciphering the control of‌‌ axonal growth, particularly during developmental regeneration. More precisely,‌ we have investigated the‌ role of RNA transport‌‌ in axons during the stage of neuronal maturation.‌ The RNA-binding protein Imp‌ seems to be one‌‌ of the molecular players involved in this process.‌ We demonstrated that it‌ controlled axonal regeneration by‌‌ regulating the mRNA encoding profilin, a modulator of‌ actin cytoskeleton polymerization. These‌ discoveries represent an important‌‌ advance in understanding the mechanisms of neuronal regeneration‌ and in establishing diagnostics‌ and targeted treatment proposals‌‌ for patients with neurodegenerative diseases such as Alzheimer's‌ or Parkinson's disease. To‌ develop a more comprehensive‌‌ approach to elucidate the‌ mechanisms of axonal regeneration in greater detail, we‌ focused on a new, poorly characterized population of‌ neurons (the Bursicon neurons) that undergo a cycle‌ of degeneration/regeneration leading to a drastic remodeling of‌ axonal terminals. The role of the Imp RNA-binding‌ protein in the regrowth of these axons is‌ currently being characterized (see Fig. 17).

Figure 17:
Axonal‌ arborization of Bursicon neurons. Confocal image mosaic of‌ the entire thorax and abdomen of adult Drosophila‌. Bursicon neurons extend long and branched axons‌ that form a tree spreading in the fly‌ abdomen (left). Axonal arborization is strongly reduced (red‌ arrows, right) in imp mutant conditions.

7.11 Fibronectin‌ networks in the extra-cellular matrix

Participants: Faisal Jayousi‌, Emmanuel Bouilhol, Laure Blanc-Féraud, Ellen‌ Van Obberghen-Schilling, Xavier Descombes.

The extracellular‌ matrix (ECM) is a complex network of proteins‌ and carbohydrates, regulates key cellular and developmental processes.‌ While computational methods for characterizing collagen topology are‌ well-established, the organization of fibronectin (FN), another vital‌ ECM protein, remains comparatively underexplored. FN's more intricate‌ structure and thinner fibrillar arrays make existing collagen-based‌ methods less effective for its analysis. This work‌ aims to lay the groundwork for studying clinical‌ tumor images from head and neck cancer patients,‌ with the goal of integrating it into a‌ broader multimodal framework to predict resistance to immunotherapy.‌

In this work, we examine FN assembled by‌ normal fibroblasts cultured in either control (non-tumor) or‌ disease-mimicking (tumor-like) conditions to validate our method for‌ assessing fibre geometry. We first extract skeletons and‌ graph representations of the underlying fibres. We propose‌ discriminant geometric and topological features to characterise FN‌ configurations in both conditions. To validate the discriminative‌ power of these features, we compared our handcrafted‌ feature-based approach with a state-of-the-art (SOTA) classification methods.‌ While SOTA methods excel in many image classification‌ tasks, they underperformed in this specific context, likely‌ due to the unique structural complexity of FN‌ networks. In contrast, our approach demonstrated competitive classification‌ performance, achieving an F1-Score of 90%. Furthermore, a‌ significant advantage of our methodology lies in its‌ explainability. The features proposed are not only interpretable‌ but also provide meaningful insights into the underlying‌ structural characteristics of FN networks, thereby enhancing the‌ transparency of the classification process (see 18).‌

8 Partnerships and cooperations

8.1 National initiatives

8.1.1‌ ANR PRC MICROBLIND

Participants: Luca Calatroni, Laure‌ Blanc-Féraud.

This project is a collaborative project‌ led by Pierre Weiss (IMT, Toulouse)[PI].

Several recent‌ revolutions in imaging rely on numerical computations. One can think of single‌ molecule localization microscopy (Nobel‌ Prize 2014) or cryo-electron‌‌ microscopy (Nobel Prize 2017). What they have in‌ common is the need‌ to perform prior mathematical‌‌ modeling and calibration of the system. Although they‌ have made it possible‌ to observe phenomena that‌‌ were previously out of reach, their expansion is‌ currently limited by an‌ important problem: it is‌‌ difficult to precisely control the imaging conditions (e.g.‌ temperatures, wavelengths, refractive indices).‌ This results in modeling‌‌ errors that can have disastrous repercussions on the‌ quality of the images‌ produced. Thus, these technologies‌‌ are currently reserved for a handful of research‌ centers possessing state-of-the-art equipment‌ and considerable interdisciplinary experience.‌‌ The objective of this project is to bring‌ new theoretical and numerical‌ solutions to overcome these‌‌ difficulties, and then to apply them to different‌ optical microscopes. This should‌ allow to democratize their‌‌ use, to reduce their cost and the preparation‌ time of the experiments.‌

The central idea is‌‌ to characterize a measurement device, not by a‌ single operator (e.g. a‌ convolution), but by a‌‌ small dimensional family allowing to model all possible‌ states of the system.‌ To our knowledge, this‌‌ idea has been very little explored so far‌ and opens many difficult‌ questions: how to best‌‌ evaluate this family experimentally and numerically? How to‌ identify the state of‌ the system from indirect‌‌ noisy observations? How to exploit this information to‌ reconstruct images in short‌ computing times? We have‌‌ begun to explore these questions in recent works‌ and wish to continue‌ this effort using tools‌‌ from optimization, harmonic analysis, probability and statistics, algebraic‌ geometry, machine learning and‌ massively parallel computing. We‌‌ hope to make significant advances in the field‌ of blind inverse problems.‌ We will validate them‌‌ on photonic microscopy problems in collaboration with opticians,‌ responsible for two microscopy‌ platforms in Nice and‌‌ Toulouse. This allows us to obtain direct feedback‌ for real problems in‌ biology. We particularly study‌‌ the problems of super-resolution by single molecule, multi-focal‌ localization and blind structured‌ illumination. Moreover, several companies‌‌ in the Toulouse area (INNOPSYS, IMACTIV-3D, AGENIUM), provide‌ us with data from‌ their microscopes (line scanning‌‌ microscope, light sheet fluorescence microscope), which will ensure‌ direct transfers to industry.‌ A workshop has been‌‌ organized at CIRM from September 29 to October‌ 3, 2025 on (Blind)‌ inverse problems in imaging:‌‌ from foundations to applications, see Event CIRM.‌

8.1.2 ANR MORPHEUS

Participants:‌ Xavier Descombes [PI],‌‌ Grégoire Malandain, Alexandre Martin, Ivan Magistro‌ Cotenta, Raffaele Martone‌.

In this project,‌‌ we propose to use the cutting-edge organoid technology‌ to test the toxicity‌ of endocrine disruptors (EDCs)‌‌ on human organs. The aim is to develop‌ computational tools and models‌ to allow the use‌‌ of organoid technology for EDC toxicity testing. The‌ project is thus divided‌ in two main objectives:‌‌ to build up and analyze a phenotypic landscape‌ of EDC effect on‌ organoid and to develop‌‌ explicative or predictive models‌ for their growth. The first goal is to‌ define and construct a phenotypic map of organoids,‌ modeled as graphs (the nodes representing the cells‌ and edges adjacency between them) for classifying EDCs‌ families. The second is to classify organoid growth‌ trajectories on this map. We will consider two‌ organoid models, gastruloids and prostate organoids. To derive‌ the phenotypic map, we combine a graph representation‌ and a deep learning approach. The deep learning‌ approach is considered for its discriminating properties whereas‌ a correspondence between the bottleneck layer of the‌ chosen neural network and the stratified graph space‌ brings some explicability to the derived classification.

This‌ 4-years project started in november 2021 and is‌ leaded by X. Descombes. It involves 3 groups:‌ IPMC (S. Clavel, Nice), Metatox, Inserm (X. Coumoul,‌ Paris) and Morpheme.

8.1.3 Targeted Project Filling Gaps‌

Participants: Grégoire Malandain, Anna Kestel.

This‌ targeted project, "Filling the gaps between scales to‌ understand biomass properties", is issued from the PEPR‌ B-Best.

The architecture of biomass is highly‌ complex and can be defined as a continuum‌ of length-scales from molecules to particles, including polymers,‌ nano-structures, assemblies, cells, and/or tissues. These scales are‌ strongly interconnected and reflect not only chemical and‌ structural properties of biomass but most importantly their‌ reactivity to transformation processes such as chemical, physical,‌ mechanical or biological reactions.

The goal of this‌ project is to identify and quantify markers at‌ different scales in order to be able to‌ propose a generic model (at least for each‌ biomass type considered) that describes and predict their‌ properties and possibly their reactivity (at the chemical,‌ biological, physical levels), with a focus on lignocellulosic‌ and algal biomass. Morpheme team will address the‌ image analysis issues.

8.1.4 3IA Senior chair, "Imaging‌ for Biology"

Participants: Laure Blanc-Féraud.

Recent advances‌ in microscope technology provide outstanding images that allow‌ biologists to address fundamental questions. This project aims‌ at developing new AI methods and algorithms for‌ (i) novel acquisition setups for super-resolution imaging, and‌ (ii) extraction of valuable quantitative information from these‌ large heterogeneous datasets. More precisely we search for‌ biomarkers in multispectral fluorescence images of tumor tissues‌ to predict the response of immunotherapy in head‌ and neck cancers.

8.2 Regional initiatives

8.2.1 Dynabio‌

The Morpheme team belongs to the Dynamics of‌ Biomolecular Networks (DYNABIO) cluster of excellence at the‌ Université Côte d’Azur (Nice, France), which brings together‌ 85 research teams from six local biology institutes:‌ C3M (Centre Méditerranéen de Médecine Moléculaire); iBV (Institut‌ de Biologie Valrose); IPMC (Institut de Pharmacologie Moléculaire‌ et Cellulaire) ; IRCAN (Institute for Research on‌ Cancer and Aging, Nice); ISA (Institut Sophia Agrobiotech)‌ and LP2M (Laboratoire de PhysioMédecine Moléculaire) as well‌ as Inria.

9 Dissemination

9.1 Promoting scientific activities‌

9.1.1 Scientific events: organization

Member of the organizing‌ committees

Laure Blanc-Féraud has co-organized the workshop IABM‌ 2025 IABM 2025
Caroline Medioni has co-organized a‌ three-day international interdisciplinary workshop (ICON) on biophotonics, bringing together around‌ a hundred researcher specialists‌ in optics, microscopy and‌‌ optogenetics in particular.

9.1.2 Scientific events: selection

Eric‌ Debreuve served as a‌ reviewer for the conference‌‌ ICIP 2025.
Xavier Descombes served as a reviewer‌ for the conferences EMBC‌ 2025 and GRETSI 2025.‌‌
Laure Blanc-Féraud served as a reviewer for the‌ conferences IABM 2025, GRETSI‌ 2025.
Grégoire Malandain served‌‌ as a reviewer for the conferences ISBI 2025‌ and GRETSI 2025.

9.1.3‌ Journal

Member of the‌‌ editorial boards

Laure Blanc-Féraud is associate editor for‌ the encyclopedia SCIENCES edited‌ by ISTE-WILEY for the‌‌ image domain.

Reviewer - reviewing activities

Eric Debreuve‌ served as a reviewer‌ for International Journal of‌‌ Biomedical Imaging (Wiley) and Pattern Recognition (Elsevier).

9.1.4‌ Invited talks

Xavier Descombes‌ gave invited talks at‌‌ IABM on March 17th, at ReinCare on May‌ 14th and during the‌ I3S/Muenster unvisersity workshop on‌‌ March 28th.
Laure Blanc-Féraud gave invited talk at‌ "Blind Inverse Problem and‌ application" Workshop at CIRM,‌‌ Sept 29 - Oct 3.
Morgane Fierville gave‌ oral presentations at ISHG,‌ April 2025, France Génomique,‌‌ Paris and at Nice-Seq, May 2025, Nice. She‌ gave poster presentations at‌ JEDN, ED 85, May‌‌ 2025, Doctoral school in Nice and in Osaka‌ symposium between IPMC and‌ University of Osaka, June‌‌ 2025
Meryem Sikouky gave a presentation at the‌ French–Indian Campus: Colloquium, Delhi,‌ India, in November 2025.‌‌
Mohamad Mohamad presented a poster at the Sophia‌ Summit in November 2025‌ entitled Navigating WSI with‌‌ RL: Potentials and Challenges.

9.1.5 Leadership within the‌ scientific community

Xavier Descombes‌ is vice-president of the‌‌ ANR committee CE45.
Xavier Descombes is member of‌ the CPS of the‌ Structuring Idex programm Dynabio‌‌ from Université Côte d'Azur
Xavier Descombes is member‌ of the scientifc committee‌ of the TLE-SKIN.
Laure‌‌ Blanc-Féraud is Vice chair of Academy RISE of‌ Idex Université Côte d'Azur.‌
Laure Blanc-Féraud is member‌‌ of the steering committee of EUR DS4H of‌ Université Côte d'Azur
Grégoire‌ Malandain is a member‌‌ of the IEEE Biomedical Image and Image Processing‌ (BIIP) Technical Committee.

9.1.6‌ Scientific expertise

Eric Debreuve‌‌ served as a reviewer of 2 ANR projects‌ for the evaluation committee‌ “CE45 - Mathématiques et‌‌ sciences du numérique pour la biologie et la‌ santé”.
Xavier Descombes is‌ expert at MENSR for‌‌ CIR and JEI.
Laure Blanc-Féraud is reviewer for‌ the CEFIPRA (Indian) program‌ and AGAUR program (Spain).‌‌

9.1.7 Research administration

Eric Debreuve is a membre‌ of the “Comité des‌ postes, EUR DS4H, Université‌‌ Côte d'Azur”.
Xavier Descombes is member of the‌ bureau of the DS4H‌ PhD selection committee.
Laure‌‌ Blanc-Féraud is member of selection committee PRA (Programme‌ de Recherches Avancées) of‌ Université Côte d'Azur
Laure‌‌ Blanc-Féraud is member of CNRS AI Rising Talent‌ committee.
Laure Blanc-Féraud is‌ member of the CNRS‌‌ informatic sciences RIPEC 3 committee.
Laure Blanc-Féraud was‌ part of the visit‌ evaluation committee HCERES of‌‌ ENS Physics Lab in Lyon.
Laure Blanc-Féraud was‌ president of the MCU‌ selection committee of Toulouse‌‌ university.
Laure Blanc-Féraud was‌ one of the two experts for the evaluation‌ of the EPFL Imaging Center.
Haydar Jammoul contributed‌ to the organization of the INRIA PhD seminars‌ at INRIA Sophia Antipolis.

9.2 Teaching - Supervision‌ - Juries - Educational and pedagogical outreach

9.2.1‌ Teaching

Eric Debreuve taught Scientific image processing and‌ Machine learning, Master SVS, 15h EqTD, M1 +‌ M2, Université Côte d'Azur, France.
Xavier Descombes taught‌ Scientific image processing and Machine learning, Master SVS,‌ 15 EqTD, M1 + M2, Université Côte d'Azur‌
Xavier Descombes taught image processing, Master GBM, 9‌ EqTD, M2, Université Côte d'Azur
Xavier Descombes taught‌ Machine Learning for Image Analysis, last year engineer,‌ 9 EqTD, M2, Sophia Antipolis Polytech. He is‌ responsible of the MLIA modulus.
Xavier Descombes taught‌ Probabilistic Approaches in Image Processing, 9 EqTD, last‌ year engineer, ISAE Supaero
Xavier Descombes taught Bio-imagerie,‌ master IRIV, 6h EqTD, Niveau M2, Université de‌ Strasbourg
Xavier Descombes taught Artificial Intelligence for Histopathologist,‌ master oncology, 3h Eq. TD, Niveau M2, Université‌ Côte d'Azur.
Laure Blanc-Féraud taught Machine Learning for‌ Image Analysis, last year engineer, 12 EqTD, M2,‌ Sophia Antipolis Polytech
Laure Blanc-Féraud is responsible of‌ Modulus Inverse problems for image processing at Msc‌ Data Science Artificial Intelligence Master (M2) and taught‌ 12h EqTD.
Imen Chtourou taught 162 hours EqTD‌ at IUT. Her courses concern programming basis and‌ relational Databases: "Bases de la programmation", "Bases de‌ données relationnelles", "Bases de la programmation","SAE : Conception‌ et implémentation d’une base de données relationnelle, Qualité‌ de développement, Architecture logicielle, Programmation avancée, Automates et‌ Langages".
Fabienne De Graeve taught Formal Genetic, 10‌ EqTP, Life Sciences License (L1), Univ. Côte d'Azur.‌
Fabienne De Graeve taught Introduction to informatics (Python),‌ 20 EqTP, Life Sciences License (L2), Univ. Côte‌ d'Azur.
Fabienne De Graeve taught Molecular Actors, 15‌ EqTP, Life Sciences License (L3), Univ. Côte d'Azur.‌
Fabienne De Graeve taught Initiation to Biological Image‌ Processing, 12 EqTP, Life Sciences Master (M1 and‌ M2), Univ. Côte d'Azur.
Fabienne De Graeve taught‌ Cellular Signalisation, 7 EqTP, Life Sciences Master (M2),‌ Univ. Côte d'Azur.
Fabienne De Graeve taught Good‌ practice in programming, 15 EqTP, Polytech Engineer School,‌ Univ. Côte d'Azur.
Fabienne De Graeve taught Apoptosis‌ and Cancer, 6 EqTP, GBHQ Professional License, Univ.‌ Côte d'Azur.
Fabienne De Graeve taught Imagery and‌ Image Processing, 12 EqTP, GBHQ Professional License, Univ.‌ Côte d'Azur.
Aneva Tsafack taught at Polytech Nice‌ Sophia for MAM 4: "Stochastic Processes for Engineers"‌ 36 EqTD "Data Valorization" 20 EqTD.
Caroline Medioni‌ taught Tissue Imaging, (12h)
Caroline Medioni is head‌ of the ‘Life Imaging’ teaching unit in the‌ SVS Master's program (32h)
Morgane Fierville : Introduction‌ à l’informatique, Licence Sciences de la vie, 20h,‌ Niveau L2, Université Côte d’Azur.
Morgane Fierville :‌ Bio-informatique, Licence Sciences de la vie, 16h, Niveau‌ L3, Université Côte d’Azur.
Morgane Fierville Programmation Python‌ et environnement Linux, Licence Sciences du Vivant, 16h,‌ Niveau L3, Université Côte d’Azur.
Morgane Fierville Programmation‌ Python et environnement Linux, Master Bioinformatique et Biologie Computationnelle, EUR Life, 4h,‌ Niveau M1, Université Côte‌ d’Azur.
Morgane Fierville ECUE‌‌ Analyses bio-informatiques de séquences biologiques, Polytech GB3, 7h,‌ Université Côte d’Azur.
Meryem‌ Sikouky taught 63h EqTD,‌‌ Unix and Shell programmingin, L1, Université Côte d’Azur‌
Mohamad Mohamad delivered three‌ tutorial sessions (4.5 EqTD)‌‌ between September and October 2025 at Polytech Sophia‌ Antipolis, within the program‌ Formation Mathématiques Appliqu´ees –‌‌ M2

9.2.2 Supervision

Xavier Descombes was the PhD‌ supervisor of Alexandre Martin,‌ co-supervisor of Faisal Jayouisi.‌‌ He is currently the PhD supervisor of Morgane‌ Fierville, Meryem Sikouky and‌ Mohamad Mohamad. He was‌‌ the supervisor of the masters Sheyenne Nguyen, Yvan‌ Contenta Magistro and Raffaele‌ Martone.
Laure Blanc-Féraud is‌‌ the PhD supervisor of Aneva Doliciane Tsafack, and‌ was co-supervisor of Faisal‌ Jayousi. She was supervisor‌‌ of Cristiano Parenti master student of Modena university‌ in visit for 3‌ months.
Grégoire Malandain is‌‌ the PhD supervisor of Ines Landolsi, and the‌ PhD co-supervisor of Haydar‌ Jammoul and Anna Kestel.‌‌

9.2.3 Juries

Xavier Descombes was member of the‌ PhD committee of Faisal‌ Jayousi as co-supervisor, Alexandre‌‌ Martin as supervisor, Quentin Rapilly as reviewer, Christer‌ Lock as reviewer and‌ Fabrice Camilleri as president.‌‌ He was member of the medical thesis jury‌ of Paul Hannetel. He‌ is member of two‌‌ CSI PhD committees as expert (Younes Habbal and‌ Zhenyu Zhu) and two‌ others as student supervisor.‌‌
Laure Blanc-Féraud was member of the PhD committee‌ of Faisal Jayousi as‌ co-supervisor, was member of‌‌ the PhD committees of M. Mohammad (Aix-Marseille university)‌ and A. Jarret (EPFL)‌ as reviewer, B. Brument‌‌ (Toulouse university) as member, and was member of‌ the HDR committee of‌ J-B Courbot (Haute Alsace‌‌ university) as reviewer. She is member of CSI‌ PhD committee of Claire‌ Couvreur (I3S Lab)
Grégoire‌‌ Malandain was member of the HDR committee of‌ S. Prima (Rennes university).‌ He is member of‌‌ CSI PhD committee of Andrea Infanti (I3S Lab)‌

9.2.4 Specific official responsibilities‌ in science outreach structures‌‌

Laure Blanc-Féraud participated in a round table discussion‌ at the Saint Raphael‌ Technology Spring Event.
Caroline‌‌ Medioni : intervention in primary schools to conduct‌ experiments on pH and‌ learn how to measure‌‌ small volumes (from $μ L$ to $m$ ).‌
Caroline Medioni : Workshops‌ on how the brain‌‌ works with secondary school students as part of‌ the Cordées de la‌ Réussite program.
Caroline Medioni‌‌ : Participation in the regional Hackathon (a day‌ of scientific mediation on‌ how neurons work and‌‌ the mechanisms that enable them to regenerate): Winner‌ of the 1st prize‌ in the 2Keuros competition‌‌ for the creation of a model for the‌ 2025 Science Festival and‌ future school interventions.
Caroline‌‌ Medioni : Participation in International Day of Women‌ and Girls in Science,‌ in an online discussion‌‌ forum with the general public on the place‌ of women in science,‌ organized by Sciences Azur‌‌ at the University of Côte d'Azur.

10 Scientific‌ production

10.1 Major publications‌

1 articleM.Mayeul‌‌ Cachia, V.Vasiliki‌ Stergiopoulou, L.Luca Calatroni, S.Sébastien‌ Schaub and L.Laure Blanc-Féraud. Fluorescence image‌ deconvolution microscopy via generative adversarial learning (FluoGAN).‌Inverse Problems395April 2023, 054006‌HAL DOI
2 articleF.Fabienne De Graeve‌, E.Eric Debreuve, S.Somia Rahmoun‌, S.Szilvia Ecsedi, A.Alia Bahri‌, A.Arnaud Hubstenberger, X.Xavier Descombes‌ and F.Florence Besse. Detecting and quantifying‌ stress granules in tissues of multicellular organisms with‌ the Obj.MPP analysis tool.TrafficJuly 2019‌HAL DOI
3 articleX.Xavier Descombes.‌ Multiple objects detection in biological images using a‌ marked point process framework.Methods2016HAL‌DOI
4 articleG.Georgios Efthymiou, A.‌Agata Radwanska, A.-I.Anca-Ioana Grapa, S.‌Stéphanie Beghelli-de la Forest Divonne, D.Dominique‌ Grall, S.Sébastien Schaub, M.Maurice‌ Hattab, S.Sabrina Pisano, M.M.‌ Poët, D. F.Didier F Pisani,‌ L.Laurent COUNILLON, X.Xavier Descombes,‌ L.Laure Blanc-Féraud and E.Ellen Van Obberghen-Schilling‌. Fibronectin Extra Domains tune cellular responses and‌ confer topographically distinct features to fibril networks.‌Journal of Cell ScienceFebruary 2021HAL
5‌ articleL.Léo Guignard, U.-M.Ulla-Maj Fiuza‌, B.Bruno Leggio, J.Julien Laussu‌, E.Emmanuel Faure, G.Gaël Michelin‌, K.Kilian Biasuz, L.Lars Hufnagel‌, G.Grégoire Malandain, C.Christophe Godin‌ and P.Patrick Lemaire. Contact area-dependent cell‌ communication and the morphological invariance of ascidian embryogenesis‌.ScienceJuly 2020HAL DOI back to‌ text back to text
6 articleB.Bastien‌ Laville, L.Laure Blanc-Féraud and G.Gilles‌ Aubert. Off-the-grid curve reconstruction through divergence regularisation:‌ an extreme point result.SIAM Journal on‌ Imaging Sciences162June 2023, 867-885‌HAL DOI
7 articleC.Clara Sanchez,‌ M.Morgane Nadal, C.Céline Cansell,‌ S.Sarah Laroui, X.Xavier Descombes,‌ C.Carole Rovère and É.Éric Debreuve.‌ Computational detection, characterization, and clustering of microglial cells‌ in a mouse model of fat-induced postprandial hypothalamic‌ inflammation.Methods2362025, 28-38HAL‌DOI

10.2 Publications of the year

International journals‌

8 articleA.Alain Corinus, S.Sophie‌ Abélanet, J.Julia Dubreuil, Z.Zhenyu‌ Zhu, S.Sabrina Pisano, C.Christelle‌ Boscagli, A.-S.Anne-Sophie Gay, D.Delphine‌ Debayle, M.Marin Truchi, K.Kevin‌ Lebrigand, S.Sandra Lacas-Gervais, F.Frédéric‌ Brau, X.Xavier Descombes, P.Patricia‌ Rousselle, M.Michel Franco and F.Frédéric‌ Luton. Human mammary 3D spheroid models uncover‌ the role of filopodia in breaching the basement‌ membrane to facilitate invasion.Matrix Biology143‌December 2025, 36-47HAL DOI
9 article‌R.Roxane Elaldi, A.Aïda Meghraoui,‌ A.Axel Elaldi, L.Luciana Petti, A.Alizé Gouleau,‌ P.Patrice Hemon,‌ W.Wassila Khatir,‌‌ J.Jonas Meziane, X.Xavier Descombes,‌ H.Henri Montaudié,‌ J.Julien Boyer,‌‌ A.Alexandre Bozec, G.Gilles Poissonnet,‌ J.-O.Jacques-Olivier Pers,‌ A.Anne Sudaka,‌‌ V. M.Veronique M Braud and F.Fabienne‌ Anjuère. CD206+CD14− Skin-Resident‌ Macrophages and DC–T Cell‌‌ Clusters Are Spatial Features Characterizing Nonrelapsing Cutaneous Squamous‌ Cell Carcinoma.Cancer‌ Immunology ResearchNovember 2025‌‌, Online ahead of printHAL DOI
10‌ articleM.M'Hamed Essafri‌, L.Luca Calatroni‌‌ and E.Emmanuel Soubies. Exact continuous relaxations‌ of l0-regularized criteria with‌ non-quadratic data terms.‌‌Journal of Global Optimization2025HAL DOI
11‌ articleS. H.Solmaz‌ Hossein Khani, K.‌‌ O.Khadidja Ould Amer, N.Noah Remy‌, B.Berangère Lebas‌, A.Anouck Habrant‌‌, A.Ali Faraj, G.Grégoire Malandain‌, G.Gabriel Paës‌ and Y.Yassin Refahi‌‌. A distinct autofluorescence distribution pattern marks enzymatic‌ deconstruction of plant cell‌ wall.New Biotechnology‌‌88September 2025, 46-60HAL DOI
12‌ articleS. H.Solmaz‌ Hossein Khani, N.‌‌Noah Remy, K. O.Khadidja Ould Amer‌, B.Berangère Lebas‌, A.Anouck Habrant‌‌, G.Grégoire Malandain, G.Gabriel Paës‌ and Y.Yassin Refahi‌. Time-lapse 3D image‌‌ datasets of spruce tree wood enzymatic deconstruction.‌Data in Brief2025‌, 111618HAL DOI‌‌
13 articleC.Clara Sanchez, M.Morgane‌ Nadal, C.Céline‌ Cansell, S.Sarah‌‌ Laroui, X.Xavier Descombes, C.Carole‌ Rovère and É.Éric‌ Debreuve. Computational detection,‌‌ characterization, and clustering of microglial cells in a‌ mouse model of fat-induced‌ postprandial hypothalamic inflammation.‌‌Methods2362025, 28-38HAL DOI back‌ to text

International peer-reviewed‌ conferences

14 inproceedingsM.‌‌Moncef Belaskri, M. L.Mohammed Lamine Benomar‌, M.Mourtada Benazzouz‌, X.Xavier Descombes‌‌ and D.Damien Ambrosetti. Improving ViT Performance‌ for Colon Cancer Histopathological‌ Image Classification Using Transfer‌‌ Learning.IC2SDA 2025 - International Conference on‌ Intelligent Computer Systems, Data‌ Science and ApplicationsBlida,‌‌ AlgeriaIEEENovember 2025, 1-8HAL DOI‌
15 inproceedingsM.Morgane‌ Fierville, K.Kevin‌‌ Lebrigand, P.Pascal Barbry and X.Xavier‌ Descombes. Linear structure‌ unfolding : application to‌‌ mouse brain in spatial transcriptomics.2025 47th‌ Annual International Conference of‌ the IEEE Engineering in‌‌ Medicine and Biology Society (EMBC)EMBC 2025 -‌ 47th Annual International Conference‌ of the IEEE Engineering‌‌ in Medicine and Biology Society2025 47th Annual‌ International Conference of the‌ IEEE Engineering in Medicine‌‌ and Biology Society (EMBC)Copenaghen, DenmarkJuly 2025‌HAL DOI
16 inproceedings‌H.Haydar Jammoul,‌‌ K.Kilian Biasuz, B.Benjamin Gallean,‌ P.Patrick Lemaire and‌ G.Grégoire Malandain.‌‌ Developmental stereotypy assessment in ascidian embryos.ISBI‌ 2025 - IEEE International‌ Symposium on Biomedical Imaging‌‌Houston, TX, United States‌April 2025HAL
17 inproceedingsF.Faisal Jayousi‌, X.Xavier Descombes, E.Emmanuel Bouilhol‌, E.Ellen van Obberghen-Schilling and L.Laure‌ Blanc-Féraud. Caractérisation quantitative des réseaux de fibronectine‌ dans la matrice extracellulaire tumorale : vers des‌ biomarqueurs pronostiques et prédictifs de réponse à l'immunothérapie‌.GRETSI - XXXe Colloque Francophone de Traitement‌ du Signal et des ImagesStrasbourg, FranceAugust‌ 2025HAL
18 inproceedingsF.Faisal Jayousi,‌ X.Xavier Descombes, E.Emmanuel Bouilhol,‌ E.Ellen van Obberghen-Schilling and L.Laure Blanc-Féraud‌. Statistical Insights into Fibronectin Networks in the‌ Extracellular Matrix.EMBC 2025Copenhagen, DenmarkJuly‌ 2025HAL back to text
19 inproceedingsG.‌Grégoire Malandain and P.Patrick Lemaire. Surface‌ preserving resampling of labeled images.ISBI proceedings‌2025 IEEE 22nd International Symposium on Biomedical Imaging‌ (ISBI)2025 IEEE 22nd International Symposium on Biomedical‌ Imaging (ISBI)Houston, United StatesIEEEApril 2025‌, 1-4HAL DOIback to text
20‌ inproceedingsM.Mohamad Mohamad, F.Francesco Ponzio‌, M.Maxime Gassier, N.Nicolas Pote‌, D.Damien Ambrosetti and X.Xavier Descombes‌. Investigating Reinforcement Learning for Histopathological Image Analysis‌.SciTePress Digital LibraryBIOIMAGING 2025 - 12th‌ International Conference on BioimagingProceedings of the 18th‌ International Joint Conference on Biomedical Engineering Systems and‌ Technologies (BIOSTEC 2025) - Volume 1Porto, Portugal‌SCITEPRESS - Science and Technology PublicationsFebruary 2025‌, 369-375HAL DOI
21 inproceedingsA. D.‌Anéva Doliciane Tsafack, L.Laure Blanc-Féraud and‌ G.Gilles Aubert. Off-the-Grid Curve Reconstruction in‌ Inverse Problems.30° Colloque sur le traitement‌ du signal et des images (GRETSI 2025)Strasboug,‌ FranceAugust 2025HAL

Conferences without proceedings

22‌ inproceedingsA. D.Anéva Doliciane Tsafack, L.‌Laure Blanc-Féraud and G.Gilles Aubert. Numerical‌ Optimization for Off-the-Grid Curve Reconstruction in 2D Images‌.12e Biennale Française des Mathématiques Appliquées et‌ IndustriellesCarcans-Maubuisson, Gironde, FranceJune 2025HAL

Reports‌ & preprints

23 miscK.Kilian Biasuz,‌ H.Haydar Jammoul, B.Benjamin Gallean,‌ Y.Yanis Asloudj, J.Julien Laussu,‌ P.Patrick Lemaire and G.Grégoire Malandain.‌ Automated cell naming reveals reproducible and variable features‌ of ascidian embryogenesis.October 2025HAL DOI‌
24 miscH.Haydar Jammoul, K.Kilian‌ Biasuz, B.Benjamin Gallean, P.Patrick‌ Lemaire and G.Grégoire Malandain. Investigating temporal‌ stereotypy in ascidian embryos development.January 2026‌HAL
25 miscB.Bastien Laville and T.‌Théo Bertrand. Dynamic off-the-grid untangling of curves‌ with Reeds-Shepp metric..June 2025HAL
26‌ miscM.Marta Lazzaretti, C.Claudio Estatico‌, A.Alejandro Melero and L.Luca Calatroni‌. Off-the-grid regularisation for Poisson inverse problems.‌January 2025HAL
27 miscW.Wessim Omezzine‌, S.Sébastien Schaub, L.Laure Blanc-Féraud‌ and L.Luca Calatroni. MUFASA: A Continuous-Time‌ Stochastic Framework for Realistic Fluorescence Microscopy Simulation.June 2025HAL
28‌ miscA. D.Anéva‌ Doliciane Tsafack, L.‌‌Laure Blanc-Féraud and G.Gilles Aubert. A‌ Curve LASSO reconstruction in‌ 2D imaging inverse problems‌‌.January 2026HAL
29 miscA. D.‌Anéva Doliciane Tsafack,‌ L.Laure Blanc-Féraud and‌‌ G.Gilles Aubert. A SMIRNOV-TYPE DECOMPOSITION ON‌ OPEN SIMPLE REGULAR CURVES‌ AND PROPERTIES *.‌‌January 2026HAL
30 miscA. D.Anéva‌ Doliciane Tsafack, L.‌Laure Blanc-Féraud and G.‌‌Gilles Aubert. Curve Reconstruction in Inverse Problems:‌ From Divergence-Measure Vector Fields‌ to Density Lebesgue Measures‌‌.January 2025HAL

Other scientific publications

31‌ inproceedingsP.P. Assemat‌, C.Cecile Barron‌‌, M.Mirjam Czjzek, P.Paul Duru‌, F.Fabienne Guillon‌, S.Severine Humbert‌‌, F.Frédéric Jamme, D.David Legland‌, G.Grégoire Malandain‌, V.Valérie Méchin‌‌, M.Matthieu Reymond, R.Richard Sibout‌ and G.Gabriel Paës‌. Filling the gaps‌‌ between scales to understand biomass properties.International‌ Conference on Plant Cell‌ Wall BiologyVancouver, Canada‌‌July 2025HAL
32 inproceedingsS.Solmaz Hossein‌ Khani, K. O.‌Khadidja Ould Amer,‌‌ N.Noah Remy, B.Berangère Lebas,‌ A.Anouck Habrant,‌ A.Ali Faraj,‌‌ G.Grégoire Malandain, G.Gabriel Paës and‌ Y.Yassin Refahi.‌ A computational Framework Reveals‌‌ Autofluorescence Distribution Patterns Marking Enzymatic Deconstruction of Plant‌ Cell Wall.Plant‌ Cell Wall BiologyVancouver‌‌ (British Columbia), CanadaJuly 2025HAL
33 inproceedings‌H.Haydar Jammoul,‌ K.Kilian Biasuz,‌‌ B.Benjamin Gallean, P.Patrick Lemaire and‌ G.Grégoire Malandain.‌ Simulation of cell count‌‌ evolution in the ascidian embryo development.ICON‌ 2025 - Biophotonic Workshop‌Nice, FranceApril 2025‌‌HAL

Scientific popularization

34 inproceedingsA. D.Anéva‌ Doliciane Tsafack, L.‌Laure Blanc-Féraud and G.‌‌Gilles Aubert. Curve Reconstruction in Fluorescence Microscopy:‌ From Modeling to Numerical‌ Implementation.25e Forum‌‌ des jeunes mathématiciennes et mathématiciensTalence (Bordeaux), France‌November 2025HAL

10.3‌ Cited publications

35 article‌‌K. W.Kiya W. Govek, P.Patrick‌ Nicodemus, Y.Yuxuan‌ Lin, J.Jake‌‌ Crawford, A. B.Artur B. Saturnino,‌ H.Hannah Cui,‌ K. Z.Kristi Zoga‌‌ aand Michael P. Hart and P. G.Pablo‌ G. Camara. CAJAL‌ enables analysis and integration‌‌ of single-cell morphological data using metric geometry.‌Nature Communications142023‌, 3672DOI back‌‌ to text
36 articleB.B. de Queiroz‌, H.H. Laghrissi‌, S.S. Rajeev‌‌, F.F. Blot, F.F. De‌ Graeve, M.M.‌ Dehecq, M.M.‌‌ Hallegger, U.U. Dag, M.M.‌ Dunoyer de Segonzac,‌ M.M. Ramialison,‌‌ C.C. Cazevieille, K.K. Keleman,‌ J.J. Ule,‌ A.A. Hubstenberger and‌‌ F.F. Besse. Axonal RNA localization is‌ essential for long-term memory‌.Nature Commun.16‌‌2560DOI back to‌ text

MORPHEME - 2025

MORPHEME - 2025

2025Activity​​​‌ reportTeamMORPHEME

Keywords

Computer Science﻿​﻿﻿ and Digital Science

Other Research Topics and​​​‌ Application Domains

1 Team members, visitors,​​​‌ external collaborators

Research Scientists﻿​﻿﻿

Faculty Members​‌﻿﻿

Post-Doctoral Fellow

PhD Students

Interns​‌﻿﻿ and Apprentices

Administrative Assistants

External Collaborator

2 Overall​​​‌ objectives

3 Research﻿﻿﻿‌ program

3.1 Research program﻿‌​‌

4 Application domains

5 Highlights​​​‌ of the year

5.1﻿﻿﻿‌ Awards

5.2 New team

6 Latest software​‌﻿﻿ developments, platforms, open data​​﻿﻿

6.1 Latest software developments​​​‌

6.1.1 Obj.MPP

6.1.2​‌﻿﻿ RCC-VascularMorphClassify

6.1.3​‌﻿﻿ Ascidian

6.1.4 Astec

6.1.5 vt-python

6.1.6 vt﻿‌​‌

6.1.7 FibreSight​​​‌

6.1.8 Mufasa﻿​​﻿

7 New​​​‌ results

7.1 Detection, characterization,﻿﻿﻿‌ and clustering of mouse﻿‌​‌ glial cells

7.2 Identifying autofluorescence﻿​﻿﻿ in biological samples using​‌﻿﻿ hyperspectral imaging

7.3﻿‌​‌ Organoid phenotyping

7.3.1 Knowledge﻿‌​‌ Distillation for Efficient 3D﻿​​﻿ Segmentation on Fluorescence Images​​​‌

7.3.2​​﻿﻿ Organoid Image Classification Using​​​‌ Deep Learning

7.3.3 Graph​​﻿﻿ Neural Network for organoid​​​‌ classification

7.4 Computational histopathology﻿﻿﻿‌

7.4.1﻿﻿﻿‌ Toward a numerical BANFF﻿‌​‌ for renal Histology

7.4.2 Quantification﻿​​﻿ of immunohistochimical slices

7.4.3 Domain﻿​​﻿ transfer in histopatholgy

7.4.4 Transformers for kidney​​​‌ cancer subtype classification

7.4.5 Reinforcement Learning﻿​​﻿ in histopathology

7.5 Spatial transcriptomics

7.5.1﻿​﻿﻿ Estimation of the subcellular​‌﻿﻿ distribution of RNA molecules​​﻿﻿ at the population level​​​‌ using optimal transport

7.5.2​​​‌ High-resolution 3D spatial transcriptomics﻿﻿﻿‌ of Drosophila Kenyon cells﻿‌​‌

7.6 From Photon Emission​​﻿﻿ to Super-Resolution: The MUFASA​​​‌ Simulator

7.7 Off-the-grid dynamic super-resolution﻿​​﻿ for fluorescence microscopy

7.8​​​‌ Embryogenesis

7.8.1 Motion Compensation﻿​﻿﻿ in Multiview Light Sheet​‌﻿﻿ Microscopy Temporal Series

7.8.2​​​‌ Surface preserving resampling of﻿﻿﻿‌ labeled images

7.8.3 Predicting cell​​​‌ division orientation in ascidian﻿﻿﻿‌ development

7.8.4 Blastoderm apical﻿​﻿﻿ cell shape in Drosophila​‌﻿﻿ melanogaster embryo

7.9 Quantifying ligno-cellulosic﻿‌​‌ enzymatic deconstruction

7.10 Deciphering the axonal﻿‌​‌ regeneration

7.11 Fibronectin​​​‌ networks in the extra-cellular﻿​﻿﻿ matrix

8 Partnerships and cooperations​​﻿﻿

8.1 National initiatives

8.1.1​​​‌ ANR PRC MICROBLIND

8.1.2 ANR MORPHEUS

8.1.3﻿​﻿﻿ Targeted Project Filling Gaps​‌﻿﻿

8.1.4​​﻿﻿ 3IA Senior chair, "Imaging​​​‌ for Biology"

8.2​​﻿﻿ Regional initiatives

8.2.1 Dynabio​​​‌

9 Dissemination​​﻿﻿

9.1 Promoting scientific activities​​​‌

9.1.1 Scientific events: organization﻿​﻿﻿

Member of the organizing​‌﻿﻿ committees

9.1.2﻿​​﻿ Scientific events: selection

9.1.3﻿﻿﻿‌ Journal

Member of the﻿‌​‌ editorial boards

Reviewer -﻿​​﻿ reviewing activities

9.1.4​​​‌ Invited talks

9.1.5 Leadership within the​​​‌ scientific community

9.1.6﻿﻿﻿‌ Scientific expertise

9.1.7 Research administration

9.2 Teaching - Supervision​​​‌ - Juries - Educational﻿​﻿﻿ and pedagogical outreach

9.2.1​‌﻿﻿ Teaching

9.2.2 Supervision

2025Activity‌ reportTeamMORPHEME

Computer Science and Digital Science

Other Research Topics and‌ Application Domains

1 Team members, visitors,‌ external collaborators

Research Scientists

Faculty Members‌

Interns‌ and Apprentices

2 Overall‌ objectives

3 Research‌ program

3.1 Research program‌‌

5 Highlights‌ of the year

5.1‌ Awards

6 Latest software‌ developments, platforms, open data

6.1 Latest software developments‌

6.1.2‌ RCC-VascularMorphClassify

6.1.3‌ Ascidian

6.1.6 vt‌‌

6.1.7 FibreSight‌

6.1.8 Mufasa

7 New‌ results

7.1 Detection, characterization,‌ and clustering of mouse‌‌ glial cells

7.2 Identifying autofluorescence in biological samples using‌ hyperspectral imaging

7.3‌‌ Organoid phenotyping

7.3.1 Knowledge‌‌ Distillation for Efficient 3D Segmentation on Fluorescence Images‌

7.3.2 Organoid Image Classification Using‌ Deep Learning

7.3.3 Graph Neural Network for organoid‌ classification

7.4 Computational histopathology‌

7.4.1‌ Toward a numerical BANFF‌‌ for renal Histology

7.4.2 Quantification of immunohistochimical slices

7.4.3 Domain transfer in histopatholgy

7.4.4 Transformers for kidney‌ cancer subtype classification

7.4.5 Reinforcement Learning in histopathology

7.5.1 Estimation of the subcellular‌ distribution of RNA molecules at the population level‌ using optimal transport

7.5.2‌ High-resolution 3D spatial transcriptomics‌ of Drosophila Kenyon cells‌‌

7.6 From Photon Emission to Super-Resolution: The MUFASA‌ Simulator

7.7 Off-the-grid dynamic super-resolution for fluorescence microscopy

7.8‌ Embryogenesis

7.8.1 Motion Compensation in Multiview Light Sheet‌ Microscopy Temporal Series

7.8.2‌ Surface preserving resampling of‌ labeled images

7.8.3 Predicting cell‌ division orientation in ascidian‌ development

7.8.4 Blastoderm apical cell shape in Drosophila‌ melanogaster embryo

7.9 Quantifying ligno-cellulosic‌‌ enzymatic deconstruction

7.10 Deciphering the axonal‌‌ regeneration

7.11 Fibronectin‌ networks in the extra-cellular matrix

8 Partnerships and cooperations

8.1.1‌ ANR PRC MICROBLIND

8.1.3 Targeted Project Filling Gaps‌

8.1.4 3IA Senior chair, "Imaging‌ for Biology"

8.2 Regional initiatives

8.2.1 Dynabio‌

9 Dissemination

9.1 Promoting scientific activities‌

9.1.1 Scientific events: organization

Member of the organizing‌ committees

9.1.2 Scientific events: selection

9.1.3‌ Journal

Member of the‌‌ editorial boards

Reviewer - reviewing activities

9.1.4‌ Invited talks

9.1.5 Leadership within the‌ scientific community

9.1.6‌ Scientific expertise

9.2 Teaching - Supervision‌ - Juries - Educational and pedagogical outreach

9.2.1‌ Teaching

9.2.4 Specific official responsibilities‌ in science outreach structures‌‌

10 Scientific‌ production

10.1 Major publications‌

10.2 Publications of the year

International journals‌

International peer-reviewed‌ conferences

Reports‌ & preprints

10.3‌ Cited publications