BKAI IGH_NeoPolyp Small - sporedata/researchdesigneR GitHub Wiki
General description
The BKAI-IGH NeoPolyp-Small dataset is a colonoscopy polyp database developed collaboratively by BKAI, Hanoi University of Science and Technology, and the Institute of Gastroenterology and Hepatology (IGH) in Vietnam. It is part of a larger NeoPolyp dataset aimed at medical imaging research, specifically for polyp segmentation and detection during endoscopy.
The NeoPolyp-Small dataset contains 1,200 images: 1,000 White Light Imaging (WLI) images and 200 Flexible Imaging Color Enhancement (FICE) images. The dataset includes detailed segmentation annotations for these images, with polyps classified into two categories—neoplastic and non-neoplastic, marked by red and green, respectively. This makes it suitable for training and testing medical image segmentation models, often used for improving the accuracy and efficiency of polyp detection in colonoscopy procedures. The dataset has been used in various models, including Unet and Attention Unet, with different backbones, such as VGG-16, MobilenetV2, and EfficientNet-B0, for both training and segmentation purposes.
The dataset supports medical image segmentation research and is used to benchmark AI models like RaBiT for polyp segmentation and localization.
Related publications
Data access
For more information on the BKAI-IGH NeoPolyp-Small dataset, visit https://bkai.ai/research/bkai-igh-neopolyp-small-a-dataset-for-fine-grained-polyp-segmentation/?fbclid=IwAR0x-wOB-74KOPW9G2jNtSCcfM4ybiLNZZuPTbDCCxE4MZHRCXR7-GoAo94