Aymen Sadraoui, PhD

👨‍🔬 About Me

I am a Data Scientist at DEEMEA, where I contribute to the development of AI-driven solutions for the structuring and analysis of medical data. My work focuses on transforming complex, multi-modal healthcare data—particularly imaging data—into reliable, actionable insights that support biopharmaceutical and medical device clinical studies. My missions are to design and validate robust machine-learning models and data pipelines that enable collaborative, multi-centric medical data analysis while respecting the highest standards of data security, privacy, and sovereignty. I am particularly committed to ensuring that analytical models are trustworthy, well-calibrated, and interpretable in clinically relevant contexts.

👨‍🔬 PhD Thesis

I completed my PhD entitled AI-Driven Methods in Histopathology for the Diagnosis and Prognosis of Hepatocellular Carcinoma at CentraleSupélec, Université Paris-Saclay, where I conducted my research between the Center for Visual Computing Laboratory and Kremlin-Bicêtre Hospital. My thesis was co-directed by Prof. Jean-Christophe Pesquet (CVN, CentraleSupélec, Univ. Paris-Saclay, France) and Prof. Catherine Guettier (Department of Pathology, Kremlin-Bicêtre Hospital, France), and co-supervised by Prof. Mounir Kaaniche (Université Sorbonne Paris Nord, France) and Prof. Amel Benazza-Benyahia (SUP'COM, University of Carthage, Tunisia).

My doctoral research focused on developing AI-based diagnostic and prognostic models for Hepatocellular Carcinoma (HCC) using histopathological whole slide images (WSIs).
On the diagnostic side, I designed deep learning frameworks for the automated detection and classification of tumor architectures. This included fully supervised multi-scale models, few-shot learning approaches, and transductive learning strategies to address limited labeled data and improve spatial coherence.
For the prognostic component, I developed models to predict recurrence risk by integrating AI-derived features extracted from WSIs with clinical and macroscopic data. This work emphasized interpretability, linking visual histopathological phenotypes to patient outcomes.
In parallel, I investigated convex optimization and algorithm unrolling techniques for stain separation and normalization in histopathology. By unrolling proximal optimization algorithms into neural network architectures, I achieved more robust stain separation and improved stain consistency across histopathological images.

📚 Publications

✅ Sadraoui, A., Laurent-Bellue, A., et al. (2024). A Proximal Approach for_Stain Separation and Normalization of Whole Slide Histopathological Images. IEEE ISBI 2026

✅ Zhou, L., Sadraoui, A., et al. (2025). UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning. CVPR 2025

✅ Laurent-Bellue, A., Sadraoui, A., et al. (2024). Deep Learning Classification and Quantification of Pejorative and Nonpejorative Architectures in Resected Hepatocellular Carcinoma. The American Journal of Pathology

✅ Sadraoui, A., Martin, S., et al. (2024). A transductive few-shot learning approach for classification of digital histopathological slides. IEEE ISBI 2024

✅ Sadraoui, A., Laurent-Bellue, A., et al. (2024). Unrolled projected gradient algorithm for stain separation in digital histopathological images. IEEE ICIP 2024

🏆 Achievements

🥇 1st Place (out of 256 teams) — DigiLut Data Challenge for Lung Transplant Rejection organized by Hôpital Foch and Bpifrance. Represented CVN, CentraleSupélec, Université Paris-Saclay, Inria, and OPIS. DigiLut Data Challenge

✉️ Contact

📧 aymen.sadraoui@universite-paris-saclay.fr
📧 aymen.sadraoui@centralesupelec.fr