Kvasir (The Kvasir Dataset)

Kvasir 数据集由医生（经验丰富的内镜医师）注释和验证的图像组成，包括显示胃肠道解剖标志、病理学发现或内窥镜程序的几个类别，即每个类别有数百张图像。图像数量足以用于不同的任务，例如图像检索、机器学习、深度学习和迁移学习等。解剖标志包括 Z 线、幽门、盲肠等，而病理发现包括食管炎，息肉，溃疡性结肠炎等。此外，我们提供了几组与去除病变相关的图像，例如，“染色和解除息肉”，“染色切除边缘”等。数据集由不同分辨率的图像组成从 720x576 到 1920x1072 像素，并以根据内容命名的单独文件夹中的排序方式进行组织。通过使用可以支持图像解释的电磁成像系统（ScopeGuide，Olympus Europe），一些包含的图像类别有一个绿色的画中画，说明了肠内内窥镜的位置和配置。这种类型的信息对于以后的调查（因此包括在内）可能很重要，但必须小心处理以检测内窥镜检查结果。

The Kvasir dataset consists of images annotated and validated by senior endoscopists, covering several categories depicting gastrointestinal anatomical landmarks, pathological findings, or endoscopic procedures, with hundreds of images per category. The sufficient volume of images supports diverse tasks including image retrieval, machine learning, deep learning, transfer learning, and more. Anatomical landmarks include the Z-line, pylorus, cecum, among others, while pathological findings encompass esophagitis, polyps, ulcerative colitis, and other related conditions. Additionally, several sets of images related to lesion removal are provided, such as "stained and polypectomized", "stained resection margins", and other corresponding categories. The dataset comprises images with resolutions ranging from 720×576 to 1920×1072 pixels, and is organized into separate folders named based on their content. Using an electromagnetic imaging system (ScopeGuide, Olympus Europe) that supports image interpretation, some included image categories feature a green picture-in-picture overlay illustrating the position and configuration of the endoscope within the gastrointestinal tract. Such information may be valuable for future investigations (and thus is included in the dataset), but must be handled with caution when interpreting endoscopic findings.

搜集汇总

背景与挑战

背景概述

Kvasir数据集是一个用于计算机辅助胃肠道疾病检测的医学图像数据集，包含医生注释的内窥镜图像，涵盖解剖标志、病理发现和内窥镜程序等多个类别，图像分辨率多样且部分带有内窥镜位置信息。该数据集适用于图像检索、机器学习和深度学习等任务，由多个研究机构于2017年发布。