A promptable CT foundation model for solid tumor evaluation

作者：Léo Machado, Léo Alberge, Hélène Philippe, Élodie Ferreres, Julien Khlaut, Julie Dupuis, Korentin Le Floch, Denis Habip Gatenyo, Pascal Roux, Jules Grégory, Maxime Ronot, Corentin Dancette, Tom Boeken, Daniel Tordjman, Pierre Manceron, Paul Hérent · 发表于：npj Precision Oncology · 年份：2025 · DOI：10.1038/s41698-025-00903-y · 被引用次数：5 · 研究领域：Radiomics and Machine Learning in Medical Imaging、Medical Imaging Techniques and Applications、Lung Cancer Diagnosis and Treatment

Carcinogenesis is inherently complex, resulting in heterogeneous tumors with variable outcomes and frequent metastatic potential. Conventional longitudinal evaluation methods like RECIST 1.1 remain labor-intensive and prone to measurement errors, while existing AI solutions face critical limitations due to tumor heterogeneity, insufficient annotations, and lack of user interaction. We developed ONCOPILOT, an interactive CT-based foundation model dedicated to 3D tumor segmentation, significantly refining RECIST 1.1 evaluations with active radiologist engagement. Trained on more than 8000 CT scans, ONCOPILOT employs intuitive visual prompts, including point-click, bounding boxes, and edit-points. It attains segmentation accuracy that matches or exceeds state-of-the-art methods, provides radiologist-level precision for RECIST 1.1 measurements, reduces inter-observer variability, and enhances workflow efficiency. Integrating clinical expertise with interactive AI capabilities, ONCOPILOT facilitates widespread access to advanced biomarkers, notably volumetric tumor analyses, thereby supporting improved clinical decision-making, patient stratification, and accelerating advancements in oncology research.