🦴 Fractured, Or Not-Fractured — That Is The Question

An Image Processing and Computer Vision Project | Ironhack Data Science Bootcamp

📌 Introduction

Computer vision is a powerful application of deep learning that enables machines to interpret and understand visual data. In this project, I applied a computer vision approach to classify bone X-ray images into two categories: fractured and non-fractured. These X-rays come from various body parts, offering a diverse dataset for training and evaluation. The primary objective was to utilize a modern deep learning model to automate fracture detection, a task with significant potential in medical diagnostics.

🎯 Project Goals

• ✅ Preprocess the dataset by splitting it into training, validation, and test sets.
• ✅ Apply data augmentation techniques (e.g., image flipping and rotation) to increase model generalizability.
• ✅ Use a pre-trained state-of-the-art model (Transfer Learning) for efficient and accurate classification.
• ✅ Evaluate model performance using accuracy metrics and confusion matrix.
• ✅ Prepare a clear and concise presentation for stakeholders.

📂 Data Source

• Bone Fracture X-ray Dataset
  📎 Download here
  This dataset consists of X-ray images from various regions of the body, labeled as fractured or non-fractured.

⚙️ Methodology

1. Data Cleaning:

   • Removed corrupted or unreadable images using TensorFlow utilities.

2. Model Architecture:

   • Used MobileNetV2 (a lightweight, efficient CNN model) as the backbone for classification.
   • Applied transfer learning by fine-tuning only the top layers to adapt to our specific classification task.

3. Training Setup:

   • Trained for 30 epochs with early stopping based on validation loss.
   • Saved the best performing model as best_model.keras.

4. Evaluation:

   • Assessed model performance using accuracy, loss curves, and a confusion matrix.

💡 Key Insights

• MobileNetV2 achieved a high classification accuracy of 96%.
• The model showed minimal Type I and Type II errors, making it highly reliable.
• Transfer learning significantly improved efficiency and performance, even with a moderately sized dataset.

🧾 Project Structure

fracture-classifier/
│
├── data/                 # Download and keep your data here
├── code/                 # Jupyter Notebooks for EDA, training and evaluation, best model (.keras) is here
├── presentation/         # Presentation to the stakeholder 
├── README.md             # Project documentation (this file)
├── task.md               # Task given in the project 
└── requirements.txt      # Python dependencies

🚀 Getting Started

1. Clone the repository

   git clone https://github.com/abhishek-chattopadhyay/project-4-cv.git
   cd project-4-cv

2. Create a virtual environment and install dependencies:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   pip install -r requirements.txt

3. Download the data from the link and put it into the data folder.
4. Run the main.ipynb

📝 Extra Notes

• This project was developed as part of the Ironhack Data Science Bootcamp.
• Future enhancements could include:
  • Trying other models like EfficientNet, ResNet, or ensemble techniques.
  • Expanding the dataset to include more fracture types.
  • Deploying the model via a simple web app for demo purposes.