tdsc-abus2023-pytorch 0.1.10

PyTorch dataset for TDSC ABUS 2023

TDSC-ABUS2023 PyTorch Dataset

A PyTorch-compatible dataset package containing volumetric data from the TDSC-ABUS2023 collection (Tumor Detection, Segmentation, and Classification Challenge on Automated 3D Breast Ultrasound).

---

📊 Dataset Description

The dataset consists of 200 3D ultrasound volumes collected using an Invenia ABUS (GE Healthcare) system at Harbin Medical University Cancer Hospital, China. All tumor annotations were created and verified by experienced radiologists.

Dataset Composition

Set	Cases	Malignant	Benign
Training	100	58	42
Validation	30	17	13
Test	70	40	30

Technical Specifications

• Image Dimensions: Vary between 843×546×270 and 865×682×354
• Pixel Spacing:
  • X-Y plane: 0.200 mm × 0.073 mm
  • Z-axis (between slices): ~0.475674 mm
• File Format: .nrrd
• Annotations: Voxel-level segmentation
  • 0: Background
  • 1: Tumor

---

📥 Installation

Install the package via pip:

pip install tdsc-abus2023-pytorch

Verify Installation

import tdsc_abus2023_pytorch
print("TDSC-ABUS2023 PyTorch Dataset is installed successfully!")

---

🚀 Usage

Loading the Original Dataset

from tdsc_abus2023_pytorch import TDSC, DataSplits

# Initialize dataset with automatic download
dataset = TDSC(
    path="./data",
    split=DataSplits.TRAIN,
    download=True
)

# Access a sample
volume, mask, label, bbx = dataset[0]

Using the Tumor-Only Dataset

This dataset contains only tumor data, suitable for classification and segmentation tasks.

from tdsc_abus2023_pytorch import TDSCTumors, DataSplits

# Initialize dataset with automatic download
dataset = TDSCTumors(
    path="./data",
    split=DataSplits.TRAIN,
    download=True
)

# Access a sample
volume, mask, label = dataset[0]

View Transformers for changing the view

from tdsc_abus2023_pytorch import TDSC, DataSplits, ViewTransformer, ViewTransposeConfig
from enum import Enum
import numpy as np

view_transformer = ViewTransformer(view=ViewTransposeConfig.CORONAL)
dataset = TDSC(path="./data", split=DataSplits.TRAIN, transforms=[view_transformer])

# Get transformed sample
vol, msk, label, bbx = dataset[0]

Custom Transformers

from tdsc_abus2023_pytorch import TDSC, DataSplits, ViewTransformer, ViewTransposeConfig
from enum import Enum
import numpy as np

class MyTransformer:
    def __init__(self, my_params):
        # your params work here...
        pass

    def transform(self, vol):
        # Your transform strategy here...
        pass

    def __call__(self, vol, mask):
        # you should return vol & mask after transform
        return transform(vol), transform(mask)

view_transformer = ViewTransformer(view=ViewTransposeConfig.CORONAL)
my_transformer = MyTransformer(*args)


dataset = TDSC(path="./data", split=DataSplits.TRAIN, transforms=[view_transformer, my_transformer])

# Get transformed sample
vol, msk, label, bbx = dataset[0]

---

📂 Data Structure

data/
  ├── Train/
  │   ├── DATA/
  │   └── MASK/
  ├── Validation/
  │   ├── DATA/
  │   └── MASK/
  └── Test/
      ├── DATA/
      └── MASK/

---

📖 Citation

If you use this dataset in your research, please cite:

@misc{luo2025tumordetectionsegmentationclassification,
    title={Tumor Detection, Segmentation and Classification Challenge on Automated 3D Breast Ultrasound: The TDSC-ABUS Challenge},
    author={Gongning Luo and others},
    year={2025},
    eprint={2501.15588},
    archivePrefix={arXiv},
    primaryClass={eess.IV},
    url={https://arxiv.org/abs/2501.15588},
}

---

🤝 Contributing

We welcome contributions! To contribute, please fork the repository, make your changes, and submit a Pull Request.