HyperKvasir Images Dataset

The authors address the prominence of artificial intelligence in the medical field, specifically in the context of sparse medical data due to legal restrictions and the labor-intensive nature of labeling training data. They present the HyperKvasir dataset, which is currently the most extensive collection of images and videos related to the gastrointestinal tract. This dataset, comprising 110,079 images and 374 videos, was obtained from real gastro- and colonoscopy examinations at Bærum Hospital in Norway and partially labeled by experienced gastrointestinal endoscopists. The dataset encompasses a variety of anatomical landmarks, normal findings, and pathological conditions. The combined image and video frames amount to approximately 1 million.

作者探讨了人工智能在医疗领域的显著地位，尤其是在法律限制导致医学数据稀疏和标注训练数据劳动密集型性质的大背景下。他们介绍了HyperKvasir数据集，该数据集是目前关于胃肠道图像和视频的最为丰富的集合。该数据集包含110,079张图像和374个视频，这些数据均来自挪威Bærum医院的实际胃肠道和结肠镜检查，并由经验丰富的胃肠道内窥镜医师部分标注。该数据集涵盖了多种解剖标志、正常发现和病理状态。图像和视频帧的总数约为100万。