Pixel-wise Annotation for Clear and Contaminated Regions Segmentation in Wireless Capsule Endoscopy Images: A Multicentre Database

The first publicly available clear and contaminated regions segmentation mask multicentre dataset created by precisely annotating 17593 copyright-free CC BY 4.0 licensed small bowel capsule endoscopy images collected from Kvasir capsule endoscopy dataset (1), SEE-AI project database (2), and CECleanliness dataset (3) . The provided dataset consists of 4 main folders available in the data repository, namely "Kvasir capsule endoscopy dataset", "SEE-AI project database", "CECleanliness dataset", and "Combination". In the "Combination" folder, there are four subfolders namely "Images", "Gastroenterologist 1"," Gastroenterologist 2", and " Gastroenterologist 3". The "Images" subfolder includes the randomly selected 153 images from the pool of raw Kvasir, SEE-AI, and CECleanliness datasets. The three remaining subfolders contain the corresponding created masks of the randomly chosen images by the three gastroenterologists. Each one of the "Kvasir capsule endoscopy dataset", "SEE-AI project database", and "CECleanliness dataset" folders contain "Images", "Binary GT", "Tri-colour GT", and " Score" subfolders. In each subfolder, the "Images" contains raw frames from the related capsule endoscopy dataset. The "Binary GT" folder contains a binary ground truth segmentation mask for each individual image of the original images folder. In a black-and –white segmentation mask image, white pixels represent clear regions while contaminated regions have been indexed by black pixels. Considering the physiological meaning of bubbles and turbid fluids, the "Tri-colour GT" folder contain three-color manually annotated ground truth masks in which the bubble boundaries, turbid fluids, and clear tissue have been labeled by the blue, red, and white colors. Ground truth images in the binary masks, and tri-color masks folders share the same names as the raw images in the original images folder. The "Score" subfolder in each folder includes an Excel file in which the amount of clear area in each image and its cleanliness level has been reported. 1. Smedsrud PH, Thambawita V, Hicks SA, Gjestang H, Nedrejord OO, Næss E, et al. Kvasir-Capsule, a video capsule endoscopy dataset. Sci Data. 2021;8(1):1–10. 2. Yokote A, Umeno J, Kawasaki K, Fujioka S, Fuyuno Y, Matsuno Y, et al. Small bowel capsule endoscopy examination and open access database with artificial intelligence: The SEE‐artificial intelligence project. DEN Open. 2024;4(1):1–10. 3. Noorda R, Nevárez A, Colomer A, Pons Beltrán V, Naranjo V. Automatic evaluation of degree of cleanliness in capsule endoscopy based on a novel CNN architecture. Sci Rep [Internet]. 2020;10(1):1–13. Available from: https://doi.org/10.1038/s41598-020-74668-8本数据集是首个公开可用的清晰与污染区域分割掩码多中心数据集，通过精准标注17593张无版权、遵循知识共享署名4.0 (CC BY 4.0)协议授权的小肠胶囊内镜图像构建而成，这些图像采集自Kvasir胶囊内镜数据集（1）、SEE-AI项目数据库（2）及CECleanliness数据集（3）。 本数据集的数据仓库中共包含4个主文件夹，分别为「Kvasir胶囊内镜数据集」「SEE-AI项目数据库」「CECleanliness数据集」与「合并数据集」。 在「合并数据集」文件夹下，设有4个子文件夹，分别为「原始图像」「胃肠科医师1」「胃肠科医师2」与「胃肠科医师3」。其中「原始图像」子文件夹包含从原始Kvasir、SEE-AI及CECleanliness数据集池中随机抽取的153张图像；其余三个子文件夹则分别存储了三位胃肠科医师对上述随机抽取图像生成的对应分割掩码。 「Kvasir胶囊内镜数据集」「SEE-AI项目数据库」及「CECleanliness数据集」三个主文件夹，均下设「原始图像」「二值真值掩码 (Binary GT)」「三色真值掩码 (Tri-colour GT)」与「评分」四个子文件夹。其中： 「原始图像」子文件夹存储对应胶囊内镜数据集的原始帧图像； 「二值真值掩码 (Binary GT)」子文件夹存储与原始图像文件夹中每张图像对应的二值分割真值掩码：在黑白分割掩码图像中，白色像素代表清晰区域，黑色像素则对应污染区域； 考虑到气泡与浑浊液体的生理学特性，「三色真值掩码 (Tri-colour GT)」子文件夹存储三色手动标注真值掩码，其中气泡边界、浑浊液体与正常组织分别以蓝色、红色与白色进行标注； 二值掩码与三色掩码文件夹中的真值掩码图像，与原始图像文件夹中的原始图像命名完全一致。 各主文件夹下的「评分」子文件夹包含一个Excel文件，用于记录每张图像的清晰区域占比及其清洁度等级。 1. Smedsrud PH, Thambawita V, Hicks SA, Gjestang H, Nedrejord OO, Næss E, 等. Kvasir-Capsule：一款视频胶囊内镜数据集. Sci Data. 2021;8(1):1–10. 2. Yokote A, Umeno J, Kawasaki K, Fujioka S, Fuyuno Y, Matsuno Y, 等. 小肠胶囊内镜检查与人工智能开放获取数据库：SEE-人工智能项目. DEN Open. 2024;4(1):1–10. 3. Noorda R, Nevárez A, Colomer A, Pons Beltrán V, Naranjo V. 基于新型卷积神经网络 (Convolutional Neural Network, CNN) 架构的胶囊内镜清洁度自动评估. Sci Rep [Internet]. 2020;10(1):1–13. 可获取自：https://doi.org/10.1038/s41598-020-74668-8