veyselozdemir/nnUNet-Breast-Cancer-Ultrasound

🎗️ Breast Ultrasound Dataset for nnU-Net v2

This repository contains the formatted dataset for breast ultrasound segmentation, prepared for use with nnU-Net v2. The dataset has been preprocessed and structured according to the nnUNet v2 dataset format for easy training and inference.

📌 Dataset Overview

• Original Dataset: Breast Ultrasound Images Dataset
• Task: Breast Ultrasound Segmentation
• Input Data: 2D ultrasound images of the breast
• Target Labels: Segmentation masks for lesion regions
• Dataset ID for nnU-Net: 101
• Dataset Name for nnU-Net: Breast

---

📁 Folder Overview

Original Dataset (combined multi-masked images)

ℹ️ About

The dataset comprises breast ultrasound images obtained from women aged 25 to 75 years. The data collection took place in 2018 and includes records from 600 female patients. It contains a total of 780 images, with an average resolution of 500×500 pixels, all stored in PNG format. Each ultrasound image is accompanied by its corresponding ground truth segmentation. The images are classified into three categories: normal, benign, and malignant.

📊 Statistics

• 📸 Total Images: 780
• ✅ Normal: 133
• 🌿 Benign: 437
• ⚠️ Malignant: 210

Breast_Ultrasound_Images/
├── benign/
│   ├── breast_001_0000.png
│   ├── breast_002_0000.png
│   ├── ...
│   └── breast_437_0000.png
├── benign_masks/
│   ├── breast_001.png
│   ├── breast_002.png
│   ├── ...
│   └── breast_437.png
├── malignant/
│   ├── breast_438_0000.png
│   ├── breast_439_0000.png
│   ├── ...
│   └── breast_647_0000.png
├── malignant_masks/
│   ├── breast_438.png
│   ├── breast_439.png
│   ├── ...
│   └── breast_647.png
├── normal/
│   ├── breast_648_0000.png
│   ├── breast_649_0000.png
│   ├── ...
│   └── breast_780_0000.png
└── normal_masks/
    ├── breast_648.png
    ├── breast_649.png
    ├── ...
    └── breast_780.png

Formatted Raw Dataset for nnU-Net v2

ℹ️ About

This directory contains the original dataset modified according to the dataset format specified in nnU-Net's documentation. The nnUNet_raw folder can be used as-is for further nnU-Net preprocess inference, as it meets the desired folder structure and filename convention format.

The label images might appear completely black, but don’t worry—they’re not. Each pixel in the label images can have one of three values: 0, 1, or 2. Here, 0 represents the background, 1 denotes benign, and 2 indicates malignant. If you want to view the labels in a human-readable black-and-white format, please refer to the original dataset.

The name of the dataset is Breast, whereas the ID is 101. This information will be required for the nnUNet preprocess inference.

nnUNet_raw/
└── Dataset101_Breast/
    ├── imagesTr/  # Training images
    │   ├── breast_001_0000.png
    │   ├── breast_002_0000.png
    │   └── ...
    ├── labelsTr/  # Training segmentation masks
    │   ├── breast_001.png
    │   ├── breast_002.png
    │   └── ...
    └── dataset.json  # Dataset metadata

Preprocessed Dataset

ℹ️ About

This directory contains the preprocessed dataset, where the nnU-Net preprocessing inference is applied to the raw dataset (nnUNet_raw). The nnUNet_preprocessed folder can be used as-is for further nnU-Net training inference on the 2d configuration.

Note that the name of the dataset is Breast, whereas the ID is 101. This information will be required for the nnUNet train inference.

nnUNet_preprocessed/
└── Dataset101_Breast/
    ├── dataset.json  # Metadata file used for training
    ├── dataset_fingerprint.json  # Dataset fingerprint
    ├── nnUNetPlans.json  # Training plans
    ├── splits_final.json  # Train/val/test split info
    ├── gt_segmentations/  # Ground truth segmentations
    │   ├── breast_001.png
    │   ├── breast_002.png
    │   └── ...
    └── nnUNetPlans_2d/  # Preprocessed data for training
        ├── breast_001.b2nd
        ├── breast_001.pkl
        ├── breast_001_seg.b2nd
        ├── breast_002.b2nd
        ├── breast_002.pkl
        ├── breast_002_seg.b2nd
        └── ...

---

🤝 Citation

If you use this dataset, please cite the original dataset:

📄 Al-Dhabyani W, Gomaa M, Khaled H, Fahmy A. Dataset of breast ultrasound images. Data in Brief. 2020 Feb;28:104863. DOI: 10.1016/j.dib.2019.104863.

---

👥 Contact

For questions or collaboration, reach out at:

• Email: ozdemirsoftware.dev@gmail.com
• GitHub: veysel-ozdemir