Beyond Traditional Approaches: Multi-Task Network for Breast Ultrasound Diagnosis

Overview

Nothing.

Requirements

• Python >=3.8
• PyTorch >=1.9
• Torchvision > 0.16
• segmentation-models-pytorch

Getting Started

0. Download dataset from Merged mask BUSI dataset and affirm number of images in the dataset by:

python check_dataset.py

If there are any missing, you can download the original dataset from here then run the merge mask code to generate same our experimental dataset by:

python merge_mask_BUSI.py

1. Clone the repository

git clone https://github.com/datct00/breast_ultrasound.git
cd breast_us

2. Install the dependencies

pip install -r requirements.txt

3. Adjust ARCH, ENCODER_NAME and OUTPUT_DIR parameter in hyper.py file to indicate different architectures and backbones.

4. Train classification

cd single_model
python train_classification.py

5. Train segmentation

cd single_model
python train_segmentation.py

Acknowledgement

@misc{Iakubovskii:2019,
  Author = {Pavel Iakubovskii},
  Title = {Segmentation Models Pytorch},
  Year = {2019},
  Publisher = {GitHub},
  Journal = {GitHub repository},
  Howpublished = {\url{https://github.com/qubvel/segmentation_models.pytorch}}
}