Alexandre Stenger
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Alexandre Stenger has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
Benoît Naegel
1 shared publication- Benoît Naegel
ICube Laboratory, CNRS, University of Strasbourg, Strasbourg, France
1 shared publication Etienne Baudrier
1 shared publicationNicolas Passat
1 shared publication- Nicolas Passat
Université de Reims Champagne-Ardenne
1 shared publication - Étienne Baudrier
ICube - UMR7357, CNRS, University of Strasbourg, Strasbourg, France
1 shared publication
Latest Publications
- 9 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Deep learning methods and probabilistic perspectives in unsupervised domain adaptation for biomedical image segmentation
Méthodes d’apprentissage profond et perspectives probabilistes en adaptation de domaine non supervisée pour la segmentation d’images biomédicales: La segmentation sémantique est essentielle en imagerie biomédicale, mais la variabilité des acquisitions entraîne un fort décalage de domaine, dégradant les performances des modèles sur de nouvelles images. Les méthodes d’adaptation de domaine non supervisée (UDA) offrent une solution, mais les approches récentes, surtout conçues pour des données naturelles synthétique-vers-réel, s’adaptent mal aux images biomédicales. Nous proposons d’exploiter des modèles de diffusion pour tirer parti de l’hypothèse de décalage de covariables en générant des masques binaires conditionnés par l’image et intégrés dans un espace latent commun pour améliorer l’adaptativité. Un second axe s’appuie sur les modèles de fondation, notamment SAM, dont la segmentation zero-shot est utilisée dans RESAMPL-UDA comme pseudo-étiquettes raffinées. Les limites de SAM en non supervisé motivent PRISM-UDA, une approche hybride conciliant SAM et auto-apprentissage pour une adaptation plus fiable.
- 9 April 2026
- Published by Agence Bibliographique de l'Enseignement Supérieur
Deep learning methods and probabilistic perspectives in unsupervised domain adaptation for biomedical image segmentation
Méthodes d’apprentissage profond et perspectives probabilistes en adaptation de domaine non supervisée pour la segmentation d’images biomédicales: La segmentation sémantique est essentielle en imagerie biomédicale, mais la variabilité des acquisitions entraîne un fort décalage de domaine, dégradant les performances des modèles sur de nouvelles images. Les méthodes d’adaptation de domaine non supervisée (UDA) offrent une solution, mais les approches récentes, surtout conçues pour des données naturelles synthétique-vers-réel, s’adaptent mal aux images biomédicales. Nous proposons d’exploiter des modèles de diffusion pour tirer parti de l’hypothèse de décalage de covariables en générant des masques binaires conditionnés par l’image et intégrés dans un espace latent commun pour améliorer l’adaptativité. Un second axe s’appuie sur les modèles de fondation, notamment SAM, dont la segmentation zero-shot est utilisée dans RESAMPL-UDA comme pseudo-étiquettes raffinées. Les limites de SAM en non supervisé motivent PRISM-UDA, une approche hybride conciliant SAM et auto-apprentissage pour une adaptation plus fiable.
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 3690-3704
Unsupervised Domain Adaptation in Biomedical Images Segmentation With Guided Diffusion Generative Prior
Semantic segmentation has suffered for a while from a lack of datasets such as ImageNet for image classification. This issue was partially alleviated by the advent of the segment anything model (SAM), which provides a foundation model trained on the largest and most diverse segmentation dataset to date. However, the SAM often falls short in segmenting specific regions, mostly in regard to biomedical images; this is why unsupervised domain adaptation (UDA) remains the best option for addressing the challenge of generalization capabilities. Classical UDA methods might be ineffective in several biomedical segmentation cases because the gap between two datasets, named domain shift, is too high. To address this issue, we propose a strategy based on learning the source mask probability distribution with a segmentation diffusion model as a generative prior to propose accurate target segmentation at inference. This latter can be guided by supplementary inputs, which allows us to call for the rich information contained in SAM raw segmentation both to perform adaptation and to improve robustness. A study was conducted using a comprehensive collection of segmentation datasets: 3 domains for mitochondria, 2 for the endoplasmic reticulum, and 2 for brain tumors, allowing the creation of 10 adaptation scenarios and providing an extensive test basis. The results of the experiments reveal that our proposed method outperforms various state-of-the-art UDA methods. Furthermore, ablation studies highlight the significant role of each component of our presented strategy. The code will be made available.
- 3 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 3690-3704
Unsupervised Domain Adaptation in Biomedical Images Segmentation With Guided Diffusion Generative Prior
Semantic segmentation has suffered for a while from a lack of datasets such as ImageNet for image classification. This issue was partially alleviated by the advent of the segment anything model (SAM), which provides a foundation model trained on the largest and most diverse segmentation dataset to date. However, the SAM often falls short in segmenting specific regions, mostly in regard to biomedical images; this is why unsupervised domain adaptation (UDA) remains the best option for addressing the challenge of generalization capabilities. Classical UDA methods might be ineffective in several biomedical segmentation cases because the gap between two datasets, named domain shift, is too high. To address this issue, we propose a strategy based on learning the source mask probability distribution with a segmentation diffusion model as a generative prior to propose accurate target segmentation at inference. This latter can be guided by supplementary inputs, which allows us to call for the rich information contained in SAM raw segmentation both to perform adaptation and to improve robustness. A study was conducted using a comprehensive collection of segmentation datasets: 3 domains for mitochondria, 2 for the endoplasmic reticulum, and 2 for brain tumors, allowing the creation of 10 adaptation scenarios and providing an extensive test basis. The results of the experiments reveal that our proposed method outperforms various state-of-the-art UDA methods. Furthermore, ablation studies highlight the significant role of each component of our presented strategy. The code will be made available.
- Open Access
- 24 June 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 196, 221-227
RESAMPL-UDA: Leveraging foundation models for unsupervised domain adaptation in biomedical images
- 24 June 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 196, 221-227
Open AccessRESAMPL-UDA: Leveraging foundation models for unsupervised domain adaptation in biomedical images