MultiBypass-4C-T40
收藏资源简介:
MultiBypass-4C-T40是一个专为Roux-en-Y胃旁路手术设计的多中心密集手术动作三元组识别数据集,由多个欧洲研究机构联合创建。该数据集包含来自四个地理分布中心的腹腔镜手术视频,并伴有阶段和步骤层级标注,但未公开具体视频数量或帧数。数据集的构建过程涉及从不同医院收集手术视频,并对每个帧进行精细的⟨器械, 动作, 目标⟩三元组标注,以捕捉器械-组织交互的细微差异。该数据集旨在评估手术动作三元组识别模型在跨中心场景下的泛化能力,解决现有基准无法系统衡量机构间变异影响的问题,为安全监控、不良事件识别和技能评估等下游应用提供可靠测试平台。
MultiBypass-4C-T40 is a multi-center dense surgical action triplet recognition dataset specifically designed for Roux-en-Y gastric bypass surgery, jointly developed by multiple European research institutions. This dataset contains laparoscopic surgical videos from four geographically distributed centers, accompanied by phase and step-level annotations, while the exact number of videos and frames is not publicly disclosed. The dataset construction process involves collecting surgical videos from various hospitals and performing fine-grained ⟨instrument, action, target⟩ triplet annotations on each frame, to capture subtle differences in instrument-tissue interactions. It aims to evaluate the generalization ability of surgical action triplet recognition models in cross-center scenarios, addressing the limitation that existing benchmarks cannot systematically measure the impact of inter-institutional variability, and providing a reliable test platform for downstream applications such as safety monitoring, adverse event recognition and skill assessment.




