3D-RAD
收藏资源简介:
3D-RAD是一个大规模数据集,旨在推动3D医学视觉问答(Med-VQA)领域的发展,使用了放射学CT扫描。该数据集包含六种多样化的VQA任务,支持开放和封闭式问题,并引入复杂的推理挑战,包括计算任务和多阶段时间分析,以实现全面基准测试。数据集遵循数据使用规定和伦理考虑,利用强大的大型语言模型进行标注,并通过严格的一致性检查确保高质量标注。此外,我们还提供了一个训练子集,以展示模型在3D诊断任务上的性能提升。
3D-RAD is a large-scale dataset intended to advance the field of 3D medical visual question answering (Med-VQA), which utilizes radiological CT scans. The dataset includes six diverse VQA tasks, supporting both open-ended and closed-ended questions, and incorporates complex reasoning challenges including computational tasks and multi-stage temporal analysis for comprehensive benchmarking. It adheres to data usage regulations and ethical considerations, employs powerful large language models for annotation, and ensures high-quality annotations through strict consistency checks. Additionally, a training subset is provided to showcase the performance improvements of models on 3D diagnostic tasks.




