Ziad Kheil

Ziad Kheil

Vulnerant omnes; interea discimus — building physics and biology into deep learning, one gradient step at a time.

Publications

Full list on Google Scholar.

Journal articles

Evaluation of Classical and Deep-Learning Deformable Registration for MR to 4D-CT Contour Mapping in Liver Stereotactic Body Radiotherapy
Z. Kheil, L. Risser, E. Cohen-Jonathan Moyal, S. Ken
Physics and Imaging in Radiation Oncology, 2026
A clinical benchmark of classical and deep learning methods for mapping contours from MRI onto each breathing phase of a 4D-CT, on 170 liver SBRT patients.
Paper

Incorporating Sliding and Local Rigidity Constraints in Neural-Based Image Registration
Z. Kheil, S. Ken, L. Risser
Scientific Reports, 2026
Builds physical priors into neural registration: organs that slide against each other and structures that should stay rigid, so predicted deformations stay physically plausible.
Preprint · Code · Playground

FGFR Inhibition as a New Therapeutic Strategy to Sensitize Glioblastoma Stem Cells to Tumor Treating Fields
P. Deshors, Z. Kheil, L. Ligat, V. Gouazé-Andersson, E. Cohen-Jonathan Moyal
Cell Death Discovery, 2025
Shows that FGFR inhibition with pemigatinib sensitizes glioblastoma stem cells to Tumor Treating Fields.
Paper

Conference papers

IMITATE: Image Registration with Context for Unknown Time Frame Recovery
Z. Kheil, L. Robinet, L. Risser, S. Ken
IEEE International Symposium on Biomedical Imaging (ISBI), 2025, oral presentation
Recovers missing or corrupted breathing phases in 4D-CT with context-aware registration.
Paper · Code

DRIM: Learning Disentangled Representations from Incomplete Multimodal Healthcare Data
L. Robinet, A. Berjaoui, Z. Kheil, E. Cohen-Jonathan Moyal
Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2024
Multimodal learning across MRI, histopathology and omics that separates shared and modality-specific information, and handles missing modalities.
Paper · Code

Code

  • biomechanical_DLIR: PyTorch framework for registration with physical constraints (rigid structures, sliding interfaces, learned tissue mechanics), with an interactive loss playground.
  • IMITATE: context-aware registration for recovering time frames in 4D-CT.
  • DRIM (co-author): disentangled multimodal learning across MRI, histopathology and omics, robust to missing modalities.
  • Elenchus: a multiplayer LLM workspace, where a team and its AI models share one thread.
  • Contributor to MONAI, the PyTorch framework for medical imaging.