🩺 Breast Ultrasound Segmentation App

Live Demo (Gradio)

---

🚀 Project Overview

This project implements breast ultrasound image segmentation using a U-Net model with a ResNet50V2 backbone. The goal is to detect and segment lesions in ultrasound images with high accuracy.

Key features:

• U-Net architecture with ResNet50V2 encoder
• Dice coefficient and combined loss (BCE + Focal Tversky)
• Data augmentation for better generalization
• Interactive Gradio app for testing and visualization

---

📁 Folder Structure

BUSI_Segmentation/
│
├── README.md                          # This file
├── breast_segmentation_unet_resnet50v2.ipynb  # Training & evaluation notebook
├── Dataset_BUSI_with_GT/              # Ultrasound dataset (images + masks)
├── models/
│   └── unet_resnet50v2_final.keras    # Trained model
├── validation_images/                 # Preprocessed validation images
├── ultrasound_segmentation_app/       # Gradio app folder
│   ├── app.py
│   ├── utils.py
│   ├── custom_object.py
│   ├── unet_resnet50v2_final.keras
│   └── requirements.txt
└── venv/                               # Optional virtual environment (ignored in Git)

---

⚡ Features

• Train U-Net on BUSI dataset with augmented data
• Stratified train/validation split and oversampling to balance classes
• Evaluate model using Dice coefficient and visualize results
• Gradio app for easy image upload and segmentation visualization:
  • Upload an ultrasound image
  • Output segmentation mask
  • Overlay mask on original image

---

🛠 Installation

1. Clone the repository:

git clone https://github.com/danielakbank/BUSI_Segmentation.git
cd BUSI_Segmentation

2. Create a virtual environment:

python -m venv venv
source venv/bin/activate   # Linux/Mac
venv\Scripts\activate      # Windows

3. Install dependencies:

pip install -r ultrasound_segmentation_app/requirements.txt

---

▶️ Running the Gradio App

1. Navigate to the app folder:

cd ultrasound_segmentation_app

2. Launch the Gradio interface:

python app.py

3. Open the displayed URL in your browser to test segmentation.

---

📊 Training & Evaluation

• Notebook: breast_segmentation_unet_resnet50v2.ipynb

• Includes:

  • Data loading and preprocessing
  • U-Net model with ResNet50V2 backbone
  • Training (encoder freeze + fine-tuning)
  • Evaluation on validation images
  • Visualizations of predicted masks vs ground truth

• Model saved as: models/unet_resnet50v2_final.keras

---

📌 Notes

• The dataset folder Dataset_BUSI_with_GT/ should contain all BUSI images and masks.
• Validation images are stored in validation_images/ for quick testing.
• If the dataset or model is too large, consider hosting externally and linking in the README.

---

📜 License

This project is open-source under the MIT License. See LICENSE for details.

---

🙏 Acknowledgements

• BUSI Dataset
• TensorFlow and Keras libraries
• Gradio for interactive web interface
• Albumentations for data augmentation