CAG-VLM
收藏资源简介:
CAG-VLM数据集是由东京大学和东京大学医院的研究团队创建的,用于冠状动脉造影图像识别的双语(日语/英语)图像-报告数据集。数据集包含539次检查的14,686帧图像,其中1,114帧图像被标记为关键帧,并与其术前报告和专家验证的诊断和治疗摘要配对。数据集的创建过程包括两个阶段:首先使用卷积神经网络进行关键帧检测和左右侧别标注,然后使用三种开源视觉语言模型进行微调,以生成临床报告和治疗建议。该数据集旨在解决冠状动脉造影图像解释和治疗计划制定中依赖专家的问题,通过AI技术提供辅助决策支持。
The CAG-VLM dataset was developed by a research team from The University of Tokyo and The University of Tokyo Hospital, serving as a bilingual (Japanese/English) image-report dataset for coronary angiography image recognition. It contains 14,686 image frames from 539 examinations, among which 1,114 frames are labeled as key frames and paired with their preoperative reports and expert-validated diagnostic and treatment summaries. The dataset construction process includes two stages: first, Convolutional Neural Networks (CNNs) are employed to conduct key frame detection and left-right laterality annotation; second, three open-source Vision-Language Models (VLMs) are fine-tuned to generate clinical reports and treatment recommendations. This dataset aims to address the expert-dependent challenge in coronary angiography image interpretation and treatment planning, providing auxiliary decision support via artificial intelligence (AI) technologies.




