partners :IBISC (univ Evry, université Paris-Saclay), centre hospitalier sud-francilien (CHSF)
Basic AI and Data Science : apprentissage statistique en grandes dimensions
Specialized ML and AI : signal, image, vision
Application domain : médecine de précision, imagerie par RM
Mots-clés deep learning, imagerie multi-modale, apprentissage faiblement supervisé
Key-wordsmachine learning, deep tech, neuroimaging, precision medicine, stroke
Laboratoires partenaires impliqués : IBISC (UEVE)
Total duration of internship: 6 months (graduate) from February 2026
Working period: durée totale du stage onwards (flexible depending on candidate profile)
Subject
Develop an uncertainty-informed multimodal fusion approach for thrombus and ischemic lesion segmentation on MRI in acute ischemic stroke. The project leverages SWAN/PHASE, DWI/ADC, and TOF-MRA to synthesize hypoperfusion-relevant information and improve clinical decision support [1–5].
Position of the problem
Stroke is the leading cause of acquired disability in adults and the second leading cause of death worldwide [6]. In ischemic stroke, a thrombus occludes a cerebral artery, depriving downstream tissue of oxygen. Treatment selection (thrombectomy or thrombolysis) requires accurate clot localization and estimation of the ischemic area from hyperacute multimodal MRI [1–5]. However, signal variability, partial redundancy across modalities, and regions with low contrast and high noise introduce segmentation uncertainty, which limits clinical accuracy [7–9]. Prior cross-modal attention work achieved reliable thrombus detection (rate ≈ 0.97) but only a Dice ≈ 0.65, underscoring the need to make uncertainty an explicit variable in fusion.
Internship objectives
A segmentation model based on cross-modal attention enabled reliable thrombus detection (rate ≈ 0.97) but achieved a Dice ≈ 0.65, highlighting the need to
address uncertain regions. The intern will assist in implementing an uncertainty-aware multimodal fusion system to enhance thrombus and ischemia segmentation.
— Uncertainty map generation. Compute voxel-wise entropy/Bayesian variance maps to identify high-uncertainty zones (e.g., thrombus borders) leveraging Bayesian deep learning tools for calibrated uncertainty [7–9].
— Uncertainty-guided attentive fusion. Inject uncertainty maps as attention masks and focus inter-modal fusion where uncertainty is high, so secondary modalities add complementary evidence (compatible with modern biomedical backb
ones such as U-Net) [10].
— Localized diffusion regularization (“focused blurring”). Apply a Gaussian blur weighted by local uncertainty and train a diffusion model on blurred images to reinforce contextual learning near the thrombus using recent denoising diffusion frameworks [11].
— Backbones & estimators. Use 3D U-Net / Cross-Modal Attention Network / Diffusion Model ; estimate uncertainty via MC-dropout, deep ensembles, and Bayesian learning [7–9] ; quantify cross-modal information with mutual information and partial information decomposition (PID) to dissect redundancy, complementarity, and synergy [12, 13].
— Experimental comparison. Compare focused blur vs. global uniform blur ; assess Dice, sensitivity, and clinical precision (with modality-specific cues from SWI/PHASE, DWI/ADC, and TOF-MRA) [1–5].
Application & Expected Impact
— Datasets & setup. Multimodal MRI from stroke patients (SWAN/PHASE, DWI/ADC, TOF-MRA) ; experiments on CHSF, MATAR, and ISLES2022. Environment : PyTorch/Python, RTX 3090, CHSF–IBISC collaboration.
— Clinical utility. Improved thrombus and ischemia segmentation enables more reliable estimation of penumbral “mismatch,” supporting reperfusion triage when onset is uncertain and enhancing treatment benefit prediction.
— Deliverables. (i) Prototype of an uncertainty-guided multimodal segmentation model ; (ii) quantitative analysis linking uncertainty to mutual information ; (iii) visual reports of clinically high-uncertainty regions ; (iv) manuscript targeting IEEE TMI or Frontiers in Neuroinformatics.
Expected results
— Prototype of an uncertainty-guided multimodal segmentation model ;
— Quantitative analysis of the link between uncertainty and mutual information ;
— Reporting and visualization of clinically high-uncertainty regions ;
— Preparation of a manuscript for IEEE Transactions on Medical Imaging or Frontiers in Neuroinformatics.
Candidate profile
We look for strongly motivated candidates (i) coming from a math, physics, computer science or engineering diploma (ii) having a strong mathematical background, notably in linear algebra, analysis, probability and statistics, in machine learning and deep learning (iii) having good programming skills on a scientific language, preferably Python.
Knowledge of medical imaging, particularly MRI, is not required, but is a strong plus. Knowledge of basic optimization theory is also appreciated.
Practical Information: The intern will be primarily hosted at the UFR science and technology (40 rue du Pelvoux), located close to the city center. However, he/she may also spend some periods at the Hospital of Corbeil.
The monthly internship gratification is about 670 euros.
Application procedure : send a motivation letter, a CV, and your University transcript (relevé de notes) since 1st year BSc to {vincent.vigneron,hichem.maaref}@univ-evry.fr.
What we offer
Hands-on experience with cutting-edge AI techniques for medical imaging.
Tackle real-world, high-impact healthcare problems using deep learning.
Close mentorship from experienced researchers at the IBISC laboratory.
Opportunities to co-author publications and present your work at conferences.
Continuation into PhD studies
Contact
{hichem.maaref, vincent.vigneron}@ibisc.univ-evry.fr, congej@yahoo.fr
Références
[1] E Mark Haacke, S Mittal, Z Wu, J Neelavalli, and Y-CN Cheng. Susceptibility weighted imaging (swi). Magnetic Resonance in Medicine, 52(3) :612–618, 2009.
[2] Àngel Rovira, Pilar Orellana, and José Alvarez-Sabín. The “susceptibility vessel sign” on t2*-weighted mri in acute ischemic stroke. Stroke, 40(2) :554–557, 2009.
[3] Steven Warach et al. Acute human stroke studied by diffusion-weighted mri. Annals of Neurology, 37(2) :231–241, 1995.
[4] DC Tong et al. Quantitative diffusion mri of acute ischemic stroke : Adc values predict tissue outcome. AJNR American Journal of Neuroradiology, 19(1) :104–110, 1998.
[5] Martin R Prince and Jeffrey Link. 3d contrast in time-of-flight mr angiography. Journal of Magnetic Resonance Imaging, 12(5) :776–783, 2000.
[6] Valery L Feigin et al. Global, regional, and national burden of stroke and its risk factors, 1990–2019. The Lancet Neurology, 20(10) :795–820, 2021.
[7] Alex Kendall and Yarin Gal. What uncertainties do we need in bayesian deep learning for computer vision ? In NeurIPS, 2017.
[8] Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation : Representing model uncertainty in deep learning. In ICML, pages 1050–1059, 2016.
[9] Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles. In NeurIPS, 2017.
[10] Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net : Convolutional networks for biomedical image segmentation. In MICCAI, pages 234–241, 2015.
[11] Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. In NeurIPS, 2020.
[12] Paul L Williams and Randall D Beer. Nonnegative decomposition of multivariate information. arXiv :1004.2515, 2010.
[13] Amer Makkeh, Dirk O Theis, and Raul Vicente. Broja-2pid : A robust estimator for bivariate partial information decomposition. Entropy, 23(10) :1274, 2021.
