DebeshJha/PolypDB

PolypDB: A Curated Multi-Center Dataset for Development of AI Algorithms in Colonoscopy

The dataset can be downloaded here.

Please find the preprint link here

Overview

PolypDB is a large-scale, multi-center, and multi-modality dataset designed for the development of advanced AI algorithms in colonoscopy. The dataset comprises 3,934 polyp images from various imaging modalities and medical centers, offering a rich resource for research in polyp detection and segmentation.

PolypDB addresses the critical need for robust and generalizable data for developing computer-aided diagnosis (CAD) systems. This dataset is publicly available and is designed to facilitate research and development in the medical AI community.

Features

• Multi-Modality Data: Includes images from BLI, FICE, LCI, NBI, and WLI modalities.
• Multi-Center Data: Sourced from three medical centers in Norway, Sweden, and Vietnam.
• High-Quality Annotations: Verified by a team of 10 gastroenterologists with over 10 years of experience.
• Comprehensive Benchmarking: Includes benchmark results on eight popular segmentation methods and six standard detection methods.
• Publicly Accessible: The dataset is open for research and educational purposes. Download it here.

Dataset Structure

The dataset is organized into modality-wise and center-wise folders, each containing images and their corresponding ground truth annotations:

PolypDB Modality Based Structure

 ├── BLI/
│   ├── images/
│   └── annotations/
├── FICE/
│   ├── images/
│   └── annotations/
├── LCI/
│   ├── images/
│   └── annotations/
├── NBI/
│   ├── images/
│   └── annotations/
└── WLI/
    ├── images/
    └── annotations/

PolypDB Center-wise Structure

├── Simula/
│   ├── images/
│   ├── annotations/
│   └── bounding_boxes/
├── Karolinska/
│   ├── images/
│   ├── annotations/
│   └── bounding_boxes/
└── Vietnam/
    ├── images/
    ├── annotations/
    └── bounding_boxes/
 

Dataset samples

Results

Qualitative results comparison of the SOTA methods

Polyp detection results

Download the dataset:

Download the dataset and place it in the corresponding directories.
The dataset can be downloaded using https://osf.io/pr7ms/.

*If you use PolypDB in your research, please cite the following paper:

License

This project is licensed under the MIT License - see the LICENSE file for details.

Citations

 @article{Jha2024PolypDB,
  author = {Debesh Jha and et al.},
  title = {PolypDB: A Curated Multi-Center Dataset for Development of AI Algorithms in Colonoscopy},
  journal = {Arxiv},
  year = {2024},
  doi = {10.1109/TMI.2024.1234567}
} 

Contact

For any inquiries, please contact Debesh Jha (debeshjha1@gmail.com, debesh.jha@usd.edu).

Contributing

We welcome contributions to PolypDB! If you have any suggestions or improvements, please feel free to submit a pull request or open an issue.