This project implements a U-Net model for breast cancer image segmentation using the PyTorch deep learning framework. The U-Net architecture is a popular choice for image segmentation tasks, showing success in medical image analysis, including breast cancer segmentation.
-
Clone the repository:
git clone https://github.com/parsakhavarinejad/Breast_cancer_image_segmentation.git cd Breast_cancer_image_segmentation -
Install the required dependencies:
pip install -r requirements.txt
-
To train the U-Net model and visualize images with masks, run:
python main.py
The dataset is available on Kaggle: https://www.kaggle.com/datasets/aryashah2k/breast-ultrasound-images-dataset/
The Breast Ultrasound Dataset, collected in 2018 from 600 female patients aged 25 to 75, consists of 780 ultrasound images in PNG format, each with an average size of 500x500 pixels. Enriched with ground truth images, it categorizes cases into normal, benign, and malignant, making it a valuable resource for machine learning models in breast cancer detection, classification, and segmentation. This well-annotated dataset contributes significantly to advancing research in medical imaging applications.
If you want to try the project, download the dataset and place it inside the data folder.
The U-Net model is well-suited for medical image segmentation. For more details on the U-Net architecture, refer to the image below:
After training, view some results, including images and loss plots:
- Example results:
- Loss plots:
Feel free to contribute to the project.