EQUR-XSM
Principal Investigator (PI)Evidential quantification of uncertainty for reliability in XAI saliency maps. The project develops methods to evaluate and improve uncertainty estimates and saliency explanations in medical imaging.
Research
Research & projects
My research examines uncertainty, explanation reliability, and data-efficient learning in medical and industrial computer vision.
Active funded projects
SpinePredict
Scientific leadMultimodal AI for decision support in spine surgery, combining consultation notes, preoperative images and dynamic posture measures to study curvature correction.
SAFE-Med
Project leadSensitivity Aware Fragility and Explainability for Medical imaging AI: a collaboration on trustworthy and explainable medical imaging methods.
Research themes
Uncertainty & explanations
Evaluate whether explanations reflect the features a model uses, and how confidence estimates behave when predictions fail.
Medical image understanding
Study segmentation, pathology localisation, and the relationship between medical images and generated radiology reports.
Learning with limited labels
Develop annotation-efficient and adaptive models for medical segmentation and heterogeneous industrial quality control.
Research collaborations
PHAIMS
PHysics-aware AI for iMaging Science: from space to atoms
PHAIMS combines artificial intelligence with physics-based models to address imaging problems where measurements have complex relationships with physical quantities and reference data are scarce. It studies multispectral and hyperspectral imaging, thermal imaging and tomography, with applications in remote sensing, biomedical imaging and materials science. The project aims to develop interpretable AI methods that use computational resources efficiently.
Adaptive industrial quality control
Computer vision for subtle defects, scarce annotations and hardware constraints. Doctoral work by Rogerio Bombardelli, with GIPSA-Lab and Schneider Electric.
DeepKnee
Deep learning applied to biomechanical data for knee injury prevention, diagnosis and monitoring, in collaboration with the MOVE team. Postdoctoral researcher: Oussama Messai.
QUAMS & CORAL
QUAMS: automatic quantification of skeletal muscle.
CORAL: angiography-based microvascular perfusion assessment.
Master’s research in collaboration with CHU Grenoble Alpes.
Earlier work
Explainable chest radiology
Doctoral research by Sayeh Gholipour-Picha on generated report evaluation, semantic similarity and visual grounding in chest X-rays. Thesis defended in 2025.
MILCOM: medical segmentation with fewer annotations
At LS2N / École Centrale de Nantes, work on interactive few-shot muscle segmentation, annotation efficiency and unsupervised domain adaptation.
Macro- and micro-expression analysis
Doctoral research on facial expression representations, domain adaptation, zero-shot learning and anomaly-based micro-expression spotting.