Colonoscopy Polyp Detection and Classification Dataset

本数据集名为“结肠镜息肉检测与分类数据集”，由堪萨斯大学医学中心等机构创建，包含37,899张图像，来源于不同医疗机构的结肠镜视频。数据集经过经验丰富的胃肠病学家标注，提供息肉的位置和分类信息，主要用于训练和评估深度学习模型，以提高结肠镜检查中息肉的检测和分类准确性。数据集涵盖多种息肉类型，包括增生性和腺瘤性息肉，旨在通过标准化训练提高医疗专业人员在胃肠内镜检查中的诊断能力，并辅助早期发现结直肠癌，降低其死亡率。

This dataset is named "Colonoscopic Polyp Detection and Classification Dataset", created by institutions such as the University of Kansas Medical Center. It contains 37,899 images sourced from colonoscopy videos collected from various medical institutions. The dataset has been annotated by experienced gastroenterologists, providing the location and classification information of polyps. It is primarily used for training and evaluating deep learning models to improve the accuracy of polyp detection and classification during colonoscopy. The dataset covers a variety of polyp types, including hyperplastic and adenomatous polyps. It aims to enhance the diagnostic capabilities of medical professionals in gastrointestinal endoscopy via standardized training, assist in the early detection of colorectal cancer, and reduce its mortality rate.

搜集汇总

数据集介绍

构建方式

The Colonoscopy Polyp Detection and Classification Dataset was meticulously constructed by aggregating endoscopic video sequences from diverse sources and meticulously annotating the ground truth of polyp locations and classifications with the aid of experienced gastroenterologists. This dataset was designed to serve as a benchmark platform for training and evaluating machine learning models, specifically tailored for the classification of polyps. The integration of data from various sources ensures a comprehensive representation of different polyp types and imaging conditions, thereby enhancing the dataset's utility for robust model training.

特点

The dataset is characterized by its comprehensive coverage of various polyp types, including hyperplastic and adenomatous polyps, which are critical for accurate colorectal cancer screening. The inclusion of multiple viewpoints and imaging conditions within the dataset ensures that models trained on it can generalize well to real-world scenarios. Additionally, the dataset's standardized annotation format, following the PASCAL VOC object detection task, facilitates easy integration and comparison with other datasets and models in the field.

使用方法

The Colonoscopy Polyp Detection and Classification Dataset can be utilized for training and evaluating deep learning models aimed at detecting and classifying polyps in colonoscopy images. Researchers can employ this dataset to fine-tune existing object detection models or develop new architectures specifically designed for polyp detection. The dataset's structured format and detailed annotations make it suitable for various tasks, including frame-based detection, sequence-based classification, and real-time polyp detection during colonoscopy procedures. Access to the dataset is provided via a public repository, ensuring transparency and reproducibility of research outcomes.

背景与挑战

背景概述

结肠镜检查息肉检测与分类数据集（Colonoscopy Polyp Detection and Classification Dataset）是由Kaidong Li等研究人员于2021年创建的，旨在解决结直肠癌（CRC）筛查中的关键问题。该数据集的构建得到了堪萨斯大学医学中心、华盛顿大学医学院和瑞尔森大学等多个机构的支持。结直肠癌是全球最常见的癌症之一，早期检测对于降低死亡率和治疗成本至关重要。因此，开发一个可靠的计算机辅助息肉检测与分类系统对于提高结肠镜检查的有效性具有重要意义。该数据集通过收集来自不同来源的内窥镜视频，并由经验丰富的胃肠病学家进行标注，为训练和评估机器学习模型提供了一个基准平台。

当前挑战

该数据集面临的挑战主要包括两个方面：一是解决领域问题，即在结肠镜检查中准确检测和分类息肉。由于息肉的大小、形状和颜色各异，且在视频中可能出现部分遮挡或远距离观察的情况，这增加了检测和分类的难度。二是数据集构建过程中的挑战，包括数据来源的多样性和图像质量的差异。不同设备采集的图像在分辨率和色彩温度上存在显著差异，这可能导致模型在实际应用中的泛化能力下降。此外，数据集中存在大量冗余帧，需要通过适当的采样策略来减少冗余，提高数据集的代表性。

常用场景

经典使用场景

Colonoscopy Polyp Detection and Classification Dataset 在结直肠癌筛查中扮演着关键角色。该数据集主要用于训练和评估深度学习模型，以实现结肠镜检查中息肉的自动检测和分类。通过使用该数据集，研究人员可以开发出能够在实时结肠镜检查中辅助医生定位和分类息肉的计算机辅助系统。这种系统的应用不仅能够提高筛查的效率，还能减少由于医生疲劳和注意力分散导致的漏诊率。

实际应用

在实际应用中，Colonoscopy Polyp Detection and Classification Dataset 可以用于开发和验证计算机辅助诊断系统，这些系统能够在结肠镜检查中实时检测和分类息肉。这种系统可以显著提高筛查的效率和准确性，减少漏诊率，从而降低结直肠癌的死亡率。此外，该数据集还可以用于培训医疗专业人员，提高他们在结肠镜检查中的操作技能和诊断能力。

衍生相关工作

基于 Colonoscopy Polyp Detection and Classification Dataset，研究人员已经开发出多种深度学习模型，用于息肉的检测和分类。例如，Faster RCNN、YOLOv3、SSD 和 RetinaNet 等模型在该数据集上进行了评估和比较，展示了深度学习在结直肠癌筛查中的潜力。此外，该数据集还促进了相关领域的研究，如图像分割、特征提取和多尺度学习等，进一步推动了计算机视觉技术在医学影像分析中的应用。