Breast Cancer Image Segmentation Using Attention UNet

This project implements a custom Attention U-Net architecture for the segmentation of breast cancer images using the Breast Ultrasound Images Dataset. The model focuses on accurately segmenting breast ultrasound images by combining the strengths of U-Net and attention mechanisms to enhance segmentation performance.

Project Features

• Model Architecture: The model is based on U-Net, enhanced with attention gates for better focusing on important regions in the images.
• Dataset: Utilizes the Breast Ultrasound Images Dataset, which includes 780 images categorized as normal, benign, and malignant.
• Training Strategy: The model is trained with binary cross-entropy loss and optimized using the Adam optimizer. Metrics such as accuracy and Intersection over Union (IoU) are used to evaluate model performance.
• Evaluation: Model performance is validated using unseen data to ensure generalization and segmentation quality.

Technologies Used

• TensorFlow & Keras: For model implementation.
• Attention Mechanism: To selectively focus on relevant parts of the input image during segmentation.
• Python Libraries: Includes NumPy, TensorFlow, Matplotlib, and more for data processing, model training, and visualization.

How to Run

1. Clone the repository:

   git clone https://github.com/yourusername/breast-cancer-segmentation.git

2. Install the required dependencies:

   pip install -r requirements.txt

3. Download the dataset from Kaggle and place it in the data/ folder.

4. Run the training script:

   python train_model.py

5. Visualize results using:

   python visualize_results.py

Conclusion

This project demonstrates the effectiveness of combining attention mechanisms with U-Net architecture for semantic segmentation, achieving high performance in breast cancer image segmentation tasks.