TN3K: Thyroid Nodule Dataset for Segmentation and Classification

Overview

TN3K is a comprehensive open-access thyroid nodule dataset containing 3,493 thyroid ultrasound images with high-quality annotations for both segmentation and classification tasks. The dataset addresses the critical need for diverse, multi-center thyroid imaging data collected from various ultrasound devices and scanning views, reflecting real-world clinical scenarios.

Dataset Characteristics

Key Features

• Size: 3,493 thyroid nodule ultrasound images
• Multi-center: Images collected from multiple medical centers
• Multi-device: Data acquired from various ultrasound devices
• Multi-view: Different scanning views and perspectives
• High-quality annotations: Expert-labeled segmentation masks and classification labels

Annotation Details

Segmentation

• Precise pixel-level thyroid nodule masks
• Thyroid gland region annotations
• Expert-validated boundaries

Classification

• Binary labels:
  • 0: Benign nodules
  • 1: Malignant nodules
• Labels provided by experienced radiologists

Dataset Applications

1. Thyroid Nodule Segmentation

• Automatic nodule boundary detection
• Thyroid gland region identification
• Multi-task learning with region priors
• Domain adaptation across different devices

2. Thyroid Nodule Classification

• Benign vs. malignant nodule diagnosis
• Noisy label learning
• Curriculum learning strategies
• Multi-center generalization studies

Download

Option 1: Baidu Drive

Segmentation Dataset (TN3K)

• Link: https://pan.baidu.com/s/1byqO5sBlt6OQdOxC4-SYng
• Extract code: trfe

Classification Dataset (TNCD)

• Link: https://pan.baidu.com/s/1_pcVYndjTcBaPmI3nb6ObQ
• Extract code: eehb

Pretrained Models

• Link: https://pan.baidu.com/s/19AZx2gUvsOvyJeDvOW2PTg
• Extract code: dp1w

Option 2: Google Drive

Complete Dataset

• Link: https://drive.google.com/drive/folders/1YtZGTb-AOmVk_uRZGWsT-pjK8hOAUix6?usp=sharing

Segmentation Dataset (Alternative)

• Link: https://drive.google.com/file/d/1reHyY5eTZ5uePXMVMzFOq5j3eFOSp50F/view?usp=sharing

If you use this dataset, please cite the following related publications:

@article{gong2022thyroid,
  title={Thyroid Region Prior Guided Attention for Ultrasound Segmentation of Thyroid Nodules},
  author={Gong, Haifan and Chen, Jiaxin and Chen, Guanqi and Li, Haofeng and Chen, Fei and Li, Guanbin},
  journal={Computers in Biology and Medicine},
  pages={106389},
  year={2022},
  publisher={Elsevier}
}

@inproceedings{gong2021multi-task,  
  author={Gong, Haifan and Chen, Guanqi and Wang, Ranran and Xie, Xiang and Mao, Mingzhi and Yu, Yizhou and Chen, Fei and Li, Guanbin},  
  booktitle={2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)},   
  title={Multi-Task Learning For Thyroid Nodule Segmentation With Thyroid Region Prior},   
  year={2021}, 
  pages={257-261},  
  doi={10.1109/ISBI48211.2021.9434087}
}

@inproceedings{gong2022less,
  title={Less is More: Adaptive Curriculum Learning for Thyroid Nodule Diagnosis},
  author={Gong, Haifan and Cheng, Hui and Xie, Yifan and Tan, Shuangyi and Chen, Guanqi and Chen, Fei and Li, Guanbin},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
  pages={248--257},
  year={2022},
  organization={Springer}
}