REAL-Colon dataset
收藏资源简介:
REAL-Colon数据集是一个公开的医学影像数据集,由Cosmo Intelligent Medical Devices和Milan Center for Neuroscience共同构建。该数据集包含了60段完整的结肠镜检查视频,共计270万个帧,跨越六个中心的病人群体。数据集经过专业标注,标注内容包括三个结肠镜阶段(外部、插入、撤退)和七个不同的结肠段(盲肠、回肠、升结肠、横结肠、降结肠、乙状结肠和直肠)。这些标注使得计算结肠镜检查的时间、评估结肠段的识别算法以及计算质量度量成为可能,如盲肠和回肠的插管率和BBPS评分。该数据集的建立为结肠镜视频时间分割任务提供了一个重要的基准,推动了相关研究的进展。
The REAL-Colon dataset is a publicly available medical imaging dataset jointly constructed by Cosmo Intelligent Medical Devices and the Milan Center for Neuroscience. This dataset contains 60 complete colonoscopy video clips, totaling 2.7 million frames, spanning patient cohorts from six medical centers. The dataset has been professionally annotated, covering three colonoscopy phases (external, insertion, withdrawal) and seven distinct colonic segments: cecum, ileum, ascending colon, transverse colon, descending colon, sigmoid colon, and rectum. These annotations enable the calculation of colonoscopy procedure durations, the evaluation of colonic segment recognition algorithms, and the computation of quality metrics such as cecal and ileal intubation rates and the BBPS score. The establishment of this dataset provides a critical benchmark for colonoscopy video temporal segmentation tasks, advancing the progress of related research.

- 1A Temporal Convolutional Network-Based Approach and a Benchmark Dataset for Colonoscopy Video Temporal SegmentationCosmo Intelligent Medical Devices, Dublin, Ireland;Milan Center for Neuroscience, University of Milano–Bicocca, Milano, Italy · 2025年



