🩺 BUSI Breast Ultrasound Image Segmentation

🧠 Overview

This project focuses on medical image segmentation for breast ultrasound scans using deep learning.

The goal is to automatically detect and segment lesion regions (benign, malignant, normal) from ultrasound images using a convolutional neural network.

---

🌐 Live Demo

🚀 Try the deployed app here:
https://huggingface.co/spaces/daniel-akbank/ultrasound-segmentation

---

📁 Repository

GitHub Repo:
https://github.com/danielakbank/BUSI-Segmentation

---

📊 Dataset

BUSI (Breast Ultrasound Images Dataset)

• Benign cases
• Malignant cases
• Normal cases
• Each image has a corresponding segmentation mask

---

🧠 Model Architecture

The model uses a U-Net style architecture with a pretrained encoder:

• Encoder: :contentReference[oaicite:0]{index=0}
• Decoder: U-Net upsampling path with skip connections
• Input size: 256 × 256 × 3
• Output: binary segmentation mask (256 × 256 × 1)

---

⚙️ Training Pipeline

1. Data Preparation

• Image-mask pairing from BUSI dataset
• Resize to 256×256
• Normalization to [0,1]
• Binary mask conversion

2. Data Splitting

• Stratified train/validation split
• Class balancing via oversampling

3. Data Augmentation

• Horizontal & vertical flips
• Rotation
• Brightness/contrast adjustment
• Elastic deformation

4. Training Strategy

• Phase 1: Encoder frozen, decoder training
• Phase 2: Full model fine-tuning (low learning rate)

---

📈 Evaluation Metrics

• Dice Coefficient (primary metric)
• Focal Tversky Loss
• Binary Cross-Entropy
• SSIM (Structural Similarity Index)
• PSNR (Peak Signal-to-Noise Ratio)

---

📊 Results

• Final Validation Dice Score: ~0.71
• Stable convergence across training phases
• Strong segmentation performance on validation data

---

🧪 Notebook Contents

The Jupyter notebook includes:

• Dataset loading & preprocessing
• Exploratory data analysis
• Data augmentation pipeline
• U-Net model implementation
• Training (Phase 1 + Phase 2)
• Evaluation & visualization
• Prediction overlays

---

🛠 Tech Stack

• Python
• TensorFlow / Keras
• NumPy
• Matplotlib
• Albumentations
• Gradio
• Hugging Face Spaces

---

⚠️ Disclaimer

This project is intended for research and educational purposes only.

It is not a medical device and should not be used for clinical diagnosis or treatment decisions.

Always consult qualified medical professionals for healthcare decisions.

---

👨‍💻 Author

Developed as a deep learning project for medical image segmentation using convolutional neural networks.