EVENT DOSSIER
On September 7, 2026, arXiv released CT-ΔBench, a benchmark tool designed specifically to evaluate the ability of visual-language models in processing longitudinal three-dimensional medical image differences. This benchmark aims to address the challenges faced by existing models in analyzing comparative images from time series, and through the provision of standardized datasets and evaluation metrics, it promotes the application of related technologies in clinical diagnostic assistance and imaging follow-up monitoring.
医学影像领域引入 CT-$\Delta$Bench 基准，旨在评估基于视觉 - 语言模型进行纵向 3D 医学影像差异报告的能力。该基准采用患者级划分防止信息泄露，并开发了专门捕捉临床意义纵向变化的变化感知指标及独立医师验证流程。研究对比了直接成对 CT 推理与间接两阶段文本差异生成两种方法，提出了 DeltaMed 作为直接成对 CT 差异报告的基线模型并在基准训练集上进行训练。这些工作为构建能更好反映现实世界纵向临床推理的时间感知医学基础模型奠定基础。