Bio. With a background in computer vision and ML/DL and a post-doctoral experience in IA for ophthalmology, I am willing to support healthcare specialists in their practices to have a significant impact using data analysis and algorithms. Indeed, through my experiences as PhD student in EnCoV (Endoscopy and Computer Vision) research group and postdoctoral researcher in the AIMI (Artificial Intelligence in Medical Imaging) research group, I benefited from experts in their fields to gather various knowledge and skills to contribute on theoretical problems of computer vision and support clinicians on their practices with learning-based algorithms.

Current situation. Since April 2023, I am working as a Data Scientist in the Data Science And Innovation Team in the Hôpital Fondation Adolphe de Rothschild in Paris, France. One of my main projects is the EviRed project, which aims to replace the current classification of diabetic retinopathy using IA.

See my CV for more details.

Publications: Peer-Reviewed International Journals

AI-Driven OCT Biomarkers for Choroidal Neovascularization Assessment in Punctate Inner Choroidopathy and Multifocal Choroiditis
To assess potential optical coherence tomography (OCT) biomarkers associated with the presence and future development of secondary choroidal neovascularization (CNV) in inflammatory lesions using artificial intelligence (AI)-based segmentation in patients with punctate inner choroidopathy (PIC) and multifocal choroiditis (MFC).
Lorenzo Ferro Desideri, Mathias Gallardo, Muriel Ott, Ariel Schlaen, Debra A Goldstein, H Nida H Sen, Maurizio Battaglia Parodi, Vita LS Dingerkus, Yael Sharon, Michal Kramer, Siqing Yu, Lisa Borner, Adrian Kaufman, Joseph Blair, Sandro de Zanet, Meng Tian, Mario Peronetti, Elisabetta Miserocchi, Maria Vittoria Cicinelli, Martin Zinkernagel, Marion R Munk
American Journal of Ophthalmology, published October 2025
Deep learning for retinal non-perfusion and foveal avascular zone analysis in wide-field OCTA in diabetic retinopathy
We developed and tested two models working on widefield OCTA images: Q-NET, for low-quality areas segmentation; and NPA-NET, for retinal non-perfusion and FAZ segmentation. Combined, these models provide the first published solution for the automatic extraction of two metrics, retinal NPI and FAZ area, which are relevant for DR monitoring. This deep learning-based approach is time-saving, reliable, and could be integrated into clinical practice in order to improve medical decisions.
Hugo Le Boité, Sophie Bonnin, Mathias Gallardo, Matthieu Lamard, Aude Couturier, Gwenolé Quellec
Scientific Reports, published August 2025
Applications of artificial intelligence to inherited retinal diseases: A systematic review
Artificial intelligence(AI)-based methods have been extensively used for the detection and management of various common retinal conditions, but their targeted development for inherited retinal diseases (IRD) is still nascent. In the context of limited availability of retinal subspecialists, genetic testing and genetic counseling, there is a high need for accurate and accessible diagnostic methods. The currently available AI studies, aiming for detection, classification, and prediction of IRD, remain mainly retrospective and include relatively limited numbers of patients due to their scarcity. We summarize the latest findings and clinical implications of machine-learning algorithms in IRD, highlighting the achievements and challenges of AI to assist ophthalmologists in their clinical practice.
Mohamad Issa, Georges Sukkarieh, Mathias Gallardo, Ilias Sarbout, Sophie Bonnin, Ramin Tadayoni, Dan Milea
Survey of Ophthalmology, published November 2024
Predicting OCT biological marker localization from weak annotations
In this paper, we propose a method that automatically locates biological markers to the ETDRS rings, only requiring B-scan-level presence SRF and IRF annotations. The neural network outputs were mapped into the corresponding ETDRS rings. We incorporated the class annotations and domain knowledge into a loss function to constrain the output with biologically plausible solutions. The method outperforms previous baselines even in the most challenging scenarios and also shows consistent en-face segmentation despite not incorporating volume information in the training process.
Javier Gamazo Tejero, Pablo Marquez-Neila, Thomas Kurmann, Mathias Gallardo, Martin Sebastian Zinkernagel, Sebastian Wolf, Raphael Sznitman
Scientific Reports, published November 2023
Learning how to robustly estimate camera pose in endoscopic videos
Surgical scene understanding plays a critical role in the technology stack of tomorrow’s intervention-assisting systems in endoscopic surgeries. For this, tracking the endoscope pose is a key component, but remains challenging due to illumination conditions, deforming tissues and the breathing motion of organs. We propose a solution for stereo endoscopes that estimates depth and optical flow to minimize two geometric losses for camera pose estimation. Most importantly, we introduce two learned adaptive per-pixel weight mappings that balance contributions according to the input image content. To do so, we train a Deep Declarative Network to take advantage of the expressiveness of deep learning and the robustness of a novel geometric-based optimization approach.
Michel Hayoz, Christopher Hahne, Mathias Gallardo, Daniel Candinas, Thomas Kurmann, Maximilian Allan, Raphael Sznitman
International Journal of Computer Assisted Radiology and Surgery, published May 2023
Evaluation of an Artificial Intelligence-based Detector of Sub- and Intra-Retinal Fluid on a large set of OCT volumes in AMD and DME
In this retrospective cohort study, we wanted to evaluate the performance of an artificial intelligence (AI) algorithm in detecting retinal fluid in spectral-domain OCT volume scans from a large cohort of patients with neovascular age-related macular degeneration (AMD) and diabetic macular edema (DME).
Oussama Habra, Mathias Gallardo, Till Meyer zu Westram, Sandro De Zanet, Damian Jaggi, Martin S. Zinkernagel, Sebastian Wolf, Raphael Sznitman
Ophthalmologica, published October 2022
Machine learning can predict anti-VEGF treatment demand in a Treat-and-Extend regimen for patients with nAMD, DME and RVO associated ME
In this work, we aimed to observe the practical feasibility of machine learning-based prediction technique for routinely collected retrospective clinical cohorts considering three different pathologies (nAMD, DME and RVO related ME). Precisely, we wish to observe the feasibility of predicting the long-term demand of anti-VEGF medication at the early stage of a one-year TER regimen in a routine clinical setting. For each pathology group, we trained two Random Forest classifiers for identifying low and high demanders and analyzed in detail their performance and the consistency of the most important features used with those leveraged by clinicians.
Mathias Gallardo, Marion R. Munk, Thomas Kurmann, Sandro De Zanet, Agata Mosinska, Isıl Kutlutürk Karagoz, Martin S. Zinkernagel, Sebastian Wolf, Raphael Sznitman
Ophthalmology Retina, published July 2021
Shape-from-Template with Curves
We propose a considerable extension of the work on “Shape-from-Template in Flatland” by [Gallardo et al., 2015] in four ways. The first way is to extend the solutions and the theoretical analysis to all sub-cases of Curve SfT; in [Gallardo et al., 2015] only the case of 2D curve reconstruction was studied. The second way is our discrete graphical method that can generate all candidate solutions; in [Gallardo et al., 2015], only a single solution could be generated. The third way is an improved method to detect critical points which has better stability than the method in [Gallardo et al., 2015]. The fourth way is a larger quantitative evaluation on real and simulated datasets.
Mathias Gallardo, Daniel Pizarro, Toby Collins, Adrien Bartoli
International Journal Computer Vision, accepted August 2019

Publications: Peer-Reviewed International Conferences

CataNet: Predicting remaining cataract surgery duration
Cataract surgery is a sight saving surgery that is performed over 10 million times each year around the world. With such a large demand, the ability to organize surgical wards and operating rooms efficiently is critical to delivery this therapy in routine clinical care. In this context, estimating the remaining surgical duration (RSD) during procedures is one way to help streamline patient throughput and workflows. To this end, we propose CataNet, a method for cataract surgeries that predicts in real time the RSD jointly with two influential elements: the surgeon’s experience, and the current phase of the surgery.
Andrés Marafioti, Michel Hayoz, Mathias Gallardo, Pablo Márquez Neila, Sebastian Wolf, Martin Zinkernagel, Raphael Sznitman
MICCAI - International Conference on Medical Image Computing and Computer Assisted Intervention, 2021
Dense Non-Rigid Structure-from-Motion and Shading with Unknown Albedos
Non-Rigid Structure-from-Motion (NRSfM) is a 3D reconstruction method which only uses a sequence of images where a surface deforms and recovers the 3D shape of the surface visible in each image. Current limitation is that NRSfM methods do not handle poorly-textured surfaces that deform non-smoothly. We show that combining NRSfM with shading constraint allows to estimate dense surfaces under non-smooth deformations.
Mathias Gallardo, Toby Collins, Adrien Bartoli
ICCV - IEEE International Conference on Computer Vision, 2017
Poster - Acceptance rate 28.9%
Using Shading and a 3D Template to Reconstruct Complex Surface Deformations
We propose to push one of the limitations of current Shape-from-Template (SfT) methods: reconstructing complex deformations on poorly-textured surfaces. For this, we combine SfT with shading constraints in an integrated optimization framework. We also propose a cascaded initialization: it uses a batch of images to estimate the 3D surface shapes (visible in each image), the illumination, the camera responses and the surface albedos which all are required to use shading.
Mathias Gallardo, Toby Collins, Adrien Bartoli
BMVC - British Machine Vision Conference, 2016
Poster - Acceptance rate 39%
Can we Jointly Register and Reconstruct Creased Surfaces by Shape-from-Template Accurately?
We investigate how to solve the Shape-from-Template problem for creased surfaces. We propose two new components and add them in a non-convex refinement: a crease-preserving smoothing term based on M-estimator and a robust boundary term to improve the surface registration.
Mathias Gallardo, Toby Collins, Adrien Bartoli
ECCV - European Conference on Computer Vision, 2016
Poster - Acceptance rate 26.6%
Shape-from-Template in Flatland
Shape-from-Template (SfT) is the problem of inferring the shape of a deformable object as observed in an image using a shape template. We address the special case of SfT, called 1DSfT, where the template is a 1D line and the input image is 2D. 1DSfT appears to be not so easy compared to the usual SfT since multiple solutions exist. We propose here a theoretical study and two computational solutions with simulated and real datasets.
Mathias Gallardo, Daniel Pizarro, Toby Collins, Adrien Bartoli
CVPR - IEEE International Conference on Computer Vision and Pattern Recognition, 2015
Poster - Acceptance rate 28.4%

Abstracts

Modèle de Deep Learning pour la détection et l’orientation clinique de la rétinopathie de la prématurité à l’aide de l’imagerie du pôle postérieur et du fond d’œil périphérique : résultats préliminaires
Évaluer la faisabilité et la précision d’un modèle de Deep learning visant à la détection de la rétinopathie de la prématurité sur les images de la rétine périphérique et des pôles postérieurs obtenues à l’aide d’un appareil d’imagerie largement utilisé en pratique clinique.
Fiammetta Catania, Mathias Gallardo, Lucile Sénicourt, Giuseppe Albanese, Amandine Barjol, Florence Metge, Thibaut Chapron, Georges Caputo
131e Congrès de Société Française d'Ophtalmologie, Mai 2025
Abstract
A customized algorithm for image quality improvement in retinopathy of prematurity increases agreement in plus disease assessment
The presence of plus disease is a key factor in determining the need for treatment in retinopathy of prematurity (ROP). However, there is poor agreement in the assessment of plus disease, even among expert graders, which remains a significant challenge in ROP management. This study aims to evaluate the impact of a customized image processing algorithm on the consistency of plus disease assessment in ROP images.
Fiammetta Catania, Thibaut Chapron, Amandine Barjol, Florence Metge, Mathias Gallardo, Lucile Sénicourt, Georges Caputo
Investigative Ophthalmology & Visual Science, June 2025
Abstract
Deep learning-based classification of acute anterior optic neuropathies in the Emergency Room, on images acquired with a portable nonmydriatic camera: a prospective study
Neuro-ophthalmological emergencies presenting to the Emergency Room (ER) require rapid and accurate diagnosis. Our aim was to develop and prospectively test a deep learning system (DLS) able to accurately identify vision-and life-threatening anterior optic neuropathies in the ER, based on fundus images acquired with a portable nonmydriatic camera (PNC).
Samy Zaher, Ayse Gungor, Ilias Sarbout, Stefana Croitoru, Diana Raicu, Bouchra Touzani, Lucile Sénicourt, Mathias Gallardo, Cédric Lamirel, Daniel Racoceanu, François Audren, Dan Milea
Investigative Ophthalmology & Visual Science, June 2025
Abstract
Device-Agnostic Deep Learning Framework for Diabetic Retinopathy Grading Using Ultra-Widefield Fundus Photography
Ultra-widefield retinography (UWF-R) provides a broad view of the retina, enhancing the detection of diabetic retinopathy (DR). However, differences in acquisition technologies between UWF-R devices such as CLARUS 500 (Carl Zeiss Meditec Inc., Dublin, CA, USA) and 200Dx (OPTOS, Dunfermline, UK) for instance pose challenges for deep learning (DL) model generalization. This study investigates the benefit of training a single DL model on a combined dataset of CLARUS and OPTOS images to enhance DR grading accuracy.
Mostafa EL HABIB DAHO, Alireza Rezaei, Rachid Zeghlache, Laurent Borderie, Pierre Deman, Mathias Gallardo, Deborah Cosette, Stephanie Magazzeni, Abir Zureik, Capucine Lepicard, Alexandre Le Guilcher, François Potevin, Sophie Bonnin, Aude Couturier, Béatrice Cochener, Pierre-Henri Conze, Mathieu Lamard, Gwenolé Quellec
Investigative Ophthalmology & Visual Science, June 2025
Abstract
Cross-Device AI Fusion: Enhancing Diabetic Retinopathy Diagnosis with Combined Clarus and Optos Images
Ultra-widefield Retinography (UWF-R) is increasingly pivotal in ophthalmic diagnostics, distinguished by its broad field of vision and advanced imaging methodologies. This research acknowledges the distinct acquisition technologies employed by the Clarus 500 (Carl Zeiss Meditec Inc., Dublin, CA, USA) and the 200Dx (Optos, Dunfermline, UK) UWF-R systems, suggesting the possibility of complementary benefits. The study investigates whether a deep learning (DL) model trained on the fusion of UWF-R images from these technologically divergent systems can enhance automatic Diabetic Retinopathy (DR) severity assessment compared to using images from a single system individually.
Mostafa El Habib Daho, Yihao Li, Rachid Zeghlache, Alireza Rezaei, Hugo Le Boité, Aude Couturier, Sophie Bonnin, Stéphanie Magazzeni, Alexandre Le Guilcher, François Potevin, Mathias Gallardo, Ramin Tadayoni, Béatrice Cochener, Pierre-Henri Conze, Mathieu Lamard
Investigative Ophthalmology & Visual Science, June 2024
Abstract
Biomarker assessment for CNV development prediction in multifocal choroiditis (MFC) and punctate inner choroidopathy (PIC): A large, longitudinal, multicenter study on patients with MFC and PIC using an artificial intelligence-based OCT fluid and biomarker detector
Secondary choroidal neovascularization (CNV) represents the major cause of vision loss in idiopathic MFC and PIC. This study assessed potential biomarkers on OCT to predict the development of CNV using an artificial intelligence (AI)- based software.
Lorenzo Ferro Desideri, Mathias Gallardo, Muriel Ott, Ariel Schlaen, Debra Goldstein, H Nida Sen, Maurizio Battaglia Parodi, Vita S Dingerkus, Yael Sharon, Michal Kramer, Siqing Yu, Sandro De Zanet, Marion Ronit Munk
Investigative Ophthalmology & Visual Science, June 2023
Abstract
Ensemble and Majority-Vote Strategies for Deep-Learning Based Detection of Atrophy-Related Biomarkers in OCT Volumes
To build a detection model for atrophy-related biomarkers in OCT Bscans of AMD patients and explore multiple training strategies using multi-grader annotations.
Davide Scandella, Mathias Gallardo, Raphael Sznitman, Martin Sebastian Zinkernagel, Sebastian Wolf
Investigative Ophthalmology & Visual Science, June 2023
Abstract
Deep-learning model to localize biological markers on OCT volumes from weak annotations
Recent developments in deep learning have shown success in accurately predicting the location of biological markers in OCT volumes of patients with Age-Related Macular Degeneration (AMD) and Diabetic Retinopathy (DR). This approach has the potential to improve clinical practices and advance medical research. However, producing fine annotations to train these algorithms is burdensome for experts. We propose a method that automatically identifies and assigns biological markers to the ETDRS rings, only requiring B-scan-level presence annotations.
Javier Gamazo Tejero, Pablo Marquez-Neila, Thomas Kurmann, Mathias Gallardo, Martin Sebastian Zinkernagel, Sebastian Wolf, Raphael Sznitman
Investigative Ophthalmology & Visual Science, June 2023
Abstract
Evaluation of an Artificial Intelligence-based Detector of Sub- and Intra-Retinal Fluid on a large set of OCT volumes in AMD and DME Patients
To evaluate the performance of an artificial intelligence (AI) algorithm in detecting retinal fluid in spectral-domain OCT volume scans from a large cohort of patients with neovascular age-related macular degeneration (AMD) and diabetic macular edema (DME).
Oussama Habra, Mathias Gallardo, Till Meyer zu Westram, Damian Jaggi, Sandro De Zanet, Martin S. Zinkernagel, Sebastian Wolf, Raphael Sznitman
EURETINA, August 2022
Abstract
Evaluating an OCT-based Algorithm of Central Subfield Thickness Estimation on AMD and DME patients
To evaluate the accuracy of an algorithm to estimate Central Subfield Thickness from OCT volumes for patients with AMD or DME.
Mathias Gallardo, Oussama Habra, Till Meyer zu Westram, Sandro De Zanet, Sebastian Wolf, Raphael Sznitman, Martin S. Zinkernagel
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2993
Abstract
Automated fovea and optic disc detection in the presence of occlusions in Fundus SLO Images
To develop and validate a machine learning algorithm for accurate estimation of the optic disc and fovea center position in infra-red SLO fundus images including cases outside of the field of view or apparent occlusions of the landmarks.
Marc Stadelmann, Agata Mosinska, Mathias Gallardo, Raphael Sznitman, Marion Munk, Stefanos Apostolopoulos, Carlos Ciller, Sandro De Zanet
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 109
Abstract
Machine learning to predict anti-VEGF treatment response in a Treat-and-Extend regimen (TRE)
This abstract encompasses our findings on the capabilities of a machine learning approach to predict treatment response of patients with wAMD, DME and RVO treated according to a TER.
Mathias Gallardo, Marion Munk, Thomas Kurmann, Sandro De Zanet, Agata Mosinska, Mark van Grinsven, Clara I. Sanchez, Martin S. Zinkernagel, Sebastian Wolf, Raphael Sznitman
Investigative Ophthalmology & Visual Science June 2020, Vol.61, 1629
Abstract

Challenges

A Deep-Learning Based Cataract Workflow Analysis
In this work, we propose a deep-learning approach for surgical steps prediction in the cataract surgeries. Based on the previous work of (Y. Jin et al., 2019), we train an ensemble model to solve simultaneously the multi-label tool classification and the multi-task surgical step classification problems and use as regularizer a correlation loss to take advantage of the relationships between steps and tools. Each submodel combines a CNN for visual feature extraction and an RNN operating on the sequence of extracted features to perform the surgical step classification.
Michel Hayoz, Mathias Gallardo, Pablo Márquez Neila, Martin S. Zinkernagel, Raphael Sznitman
CATARACTS Workflow, part of the EndoVis Challenge, MICCAI 2020
Code & Report - Ranked 1st over 5 international participants

Book Chapter

Non-Rigid Structure-from-Motion and Shading
We show how photometric and motion-based approaches can be combined to reconstruct the 3D shape of deformable objects from monocular images. We start by motivating the problem using real-world applications. We give a comprehensive overview of the state-of-the-art approaches and discuss their limitations for practical use in these applications. We then introduce the problem of Non-Rigid Structure-from-Motion and Shading (NRSfMS), where photometric and geometric information are used for reconstruction, without prior knowledge about the shape of the deformable object. We present in detail the first technical solution to NRSfMS and close the chapter with the main remaining open problems.
Mathias Gallardo, Toby Collins, Adrien Bartoli
Advances in Photometric 3D-Reconstruction, J.-D. Durou, M. Falcone, Y. Quéau and S. Tozza (Eds.), Springer, 2020

National Journals and Conference Proceedings

Utilisation de la photométrie et d'un patron pour la reconstruction de surfaces pliées et la calibration photométrique
Mathias Gallardo, Toby Collins, Adrien Bartoli
Traitement du signal, GRETSI-CNRS, special issue: selected papers from RFIA 2016, accepted May 2017
Recalage et Reconstruction 3D de Surfaces Pliées par Shape-from-Template
Mathias Gallardo, Toby Collins, Adrien Bartoli
RFIA - Congrès Francophone de Reconnaissance des Formes et Intelligence Artificielle, Clermont-Ferrand, 2016 (Oral)
Shape-from-Template dans Flatland
Mathias Gallardo, Daniel Pizarro, Adrien Bartoli, Toby Collins
ORASIS - Congrès Francophone des Jeunes Chercheurs en Vision par Ordinateur, Amiens, 2015 (Oral)

PhD Dissertation

Contributions to
Monocular Deformable 3D Reconstruction:
Curvilinear Objects and Multiple Visual Cues
Mathias Gallardo
Université Clermont Auvergne, September 2018
Supervisors: Toby Collins and Adrien Bartoli

Work Experiences

Image restoration for biometrics - M2 internship
In 2013, I could complete my 6 months internship in Morpho (Osny, France), now Idemia, world leader in digital security and identification technologies. My project was to propose a method of estimation and correction of the inherent blur (Point Spread Function) for fingerprint modules.
Defects Detection on Silicon Wafers - M1 internship
In 2012, I could work during 3 months on an industrial project at Soitec (Bernin, France), world leader in semiconductor materials. The objective was to propose and implement an algorithm of automatic detection of manufacturing defects on silicon wafers using IR images.
© Mathias Gallardo, 2025