Raidium Publishes Promptable CT Foundation Model for Tumor Segmentation
Raidium, a Paris-based medical AI company founded in 2022 by radiologist Dr. Paul Hérent and data scientist Pierre Manceron, focuses on developing multimodal foundation models to support radiology and oncology workflows. The company is training its models on a proprietary dataset of 1 billion real-world medical images, aiming to assist radiologists with scalable precision imaging tools.
Raidium’s oncology-focused model, ONCOPILOT, has now been published in npj Precision Oncology. ONCOPILOT is an interactive CT-based foundation model built for 3D segmentation of solid tumors. It allows radiologists to semi-automatically segment lesions using visual prompts (bounding boxes, point-click inputs) and refine outputs interactively, helping integrate AI into existing clinical workflows.
ONCOPILOT Foundation Model Training and Evaluation.
The model was trained on over 8,000 CT scans and evaluated against state-of-the-art baselines. ONCOPILOT’s ability to support volumetric analysis and radiomics extraction is presented as a step forward from conventional RECIST evaluations based on linear measurements.
This milestone builds on Raidium’s momentum from late 2024, when it secured €16 million in seed funding from Newfund, Kurma Partners, and the European Innovation Council. The company’s broader goal is to create AI "co-pilots" for radiology, not to automate diagnosis, but to enhance efficiency and precision amid growing imaging demands. Initial partnerships include academic collaborations with AP-HP and clinical deployments in precision oncology initiatives.
With the ONCOPILOT model now validated in early studies, Raidium plans to expand its foundation model applications across oncology research, clinical trial support, and biomarker discovery, while continuing preparations for regulatory certifications in Europe and the U.S.
Topics: AI & Digital
