Internship Proposal (M2 / Engineering Final Year) — PhD track
Description
TheraPanacea is a medical technology company using AI to improve cancer care, with a focus on radiotherapy. We build solutions that work alongside existing systems to help clinicians plan and deliver treatment more efficiently and consistently. Our goal is simple: support care teams by reducing routine workload and helping them focus on what matters most - patients.
Internship Proposal (M2 / Engineering Final Year) — PhD track
Latent Diffusion Models for CT–Histology Registration and Immune Microenvironment Mapping
TheraPanacea / Institut Gustave Roussy / Centre Leon Berard
Duration & Start
5–6 months · Start: Mar - May 2027 · Publication target: Oct 2027
Context
The spatial distribution of tumor-infiltrating lymphocyte (TIL) subtypes within the tumor microenvironment (TME) is a key determinant of treatment response. Histological whole-slide imaging (WSI) characterizes the TME at cellular resolution, but CT remains the clinical reference modality. Mapping immune microenvironment information from histology onto CT — across a resolution gap of ~100×, with tissue deformation and no shared intensity statistics — is an open and high-impact problem.
This project is grounded in a unique paired CT–histology cohort assembled at IGR/CLB/HCL, annotated with whole-tumor contours and per-lymphocyte-subtype sub-regional zones — a dataset rare in its multimodal depth and biological richness.
Objectives
- CT–Histology Registration in Latent Space — Design a registration pipeline based on a modality-agnostic latent representation learned by a jointly trained LDM encoder, enabling deformable alignment without pixel-level intensity matching across domains.
- Cross-modal Tumor Delineation— Transfer histology-space TME annotations (tumor contours, lymphocyte sub-zones) into CT volume coordinates via the learned deformation field.
- Conditional LDM for TME Prediction from CT — Train a conditional latent diffusion model to synthesize immune sub-regional biological pathways directly from CT, using transferred contours as supervision.
- Publication — Target one clinical paper (Radiotherapy and Oncology) and one methodological paper (Medical Image Analysis).
Requirements
- M.Sc. in Machine Learning, Computer Vision, or Medical Image Analysis (ongoing or completed)
- Strong hands-on experience with deep generative models (diffusion models, VAEs) and PyTorch
- Solid background in image registration and/or segmentation (classical and learning-based)
- Familiarity with 3D medical imaging (SimpleITK, MONAI) and WSI processing (OpenSlide) desirable
PhD Opportunity
Exceptional candidates might be be offered a CIFRE PhD continuation at TheraPanacea, extending this work towards immune biomarker discovery and integration into holistic treatment optimization oncology workflows.
Apply: CV + transcripts + cover letter → LinkedIN
Skills
Sign in to read and join the discussion.