🩻 Fracture Detection System using Dual CNN Models with PyQt5 GUI

This project is a deep learning-based desktop application for detecting bone fractures from X-ray images using two CNN models.
It provides a graphical interface built with PyQt5, enabling users to load images, perform automatic classification, and visualize confusion matrices dynamically.

---

🚀 Features

• 🧠 Two-Stage CNN Detection
  1. Model 1: Detects whether the input is a valid X-ray image.
  2. Model 2: Classifies the type of bone fracture (or detects if it’s normal).
• 🖼️ Interactive PyQt5 GUI for selecting, viewing, and classifying images.
• 📊 Dynamic metrics display (precision, recall, F1-score, etc.) per image.
• 🔥 Top-5 prediction results list for Model 2.
• 🎨 Automatic confusion matrix generation and visualization for each prediction.

---

🧰 Tech Stack

Component	Description
Language	Python 3.x
GUI Framework	PyQt5
Deep Learning	TensorFlow / Keras
Image Handling	Pillow (PIL)
Visualization	Matplotlib, Seaborn
Data Format	.keras, .json

---

📂 Project Structure

detection_fractureTkinter/
│
├── main_dual_cnn.py           # Main GUI file (PyQt5 interface)
├── model1_xray_filterCNN.keras # Model 1 - X-ray filter (binary classifier)
├── TfractureCNNRev_model.keras # Model 2 - Fracture type classifier
├── class_names_model1.json     # Class labels for Model 1
├── class_names_model2.json     # Class labels for Model 2
├── evaluation_metrics.json     # Model performance metrics
├── confusion_matrix_final.png  # Static confusion matrix example
├── Model1_CNNConv2D.ipynb      # Training notebook
├── requirements.txt            # Project dependencies
└── README.md                   # Documentation

---

🧠 How It Works

🩺 Model 1 — X-ray Filter (Binary CNN)

This model detects whether the uploaded image is a real X-ray or a non-X-ray image.
If the image is not an X-ray, the system will warn the user and stop further classification.

🦴 Model 2 — Fracture Classification (Multi-class CNN)

If the input is a valid X-ray, Model 2 classifies it into one of several fracture types, such as:

• Avulsion Fracture
• Comminuted Fracture
• Fracture Dislocation
• Greenstick Fracture
• Hairline Fracture
• Impacted Fracture
• Longitudinal Fracture
• Normal
• Oblique Fracture
• Pathological Fracture
• Spiral Fracture

It also displays top-5 probabilities and dynamic confusion matrix visualizations.

---

🧪 Model Training

Training was conducted in Model1_CNNConv2D.ipynb using Keras Sequential API.

🔹 General Workflow

1. Data Preparation

   • X-ray dataset split into train, validation, and test sets.
   • Images normalized and resized to (224, 224) for Model 1 and (256, 256) for Model 2.

2. Model Architecture

   model = Sequential([
       Conv2D(32, (3,3), activation='relu', input_shape=(224,224,3)),
       MaxPooling2D(2,2),
       Conv2D(64, (3,3), activation='relu'),
       MaxPooling2D(2,2),
       Flatten(),
       Dense(128, activation='relu'),
       Dense(num_classes, activation='softmax')
   ])

3. Compilation & Training

   model.compile(optimizer='adam',
                 loss='categorical_crossentropy',
                 metrics=['accuracy'])
   history = model.fit(train_data, validation_data=val_data, epochs=20)

4. Model Export

   model.save('TfractureCNNRev_model.keras')

---

🖥️ Running the Application

1️⃣ Clone the Repository

git clone https://github.com/meganfbr/detection_fractureTkinter.git
cd detection_fractureTkinter

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Run the Application

python main_dual_cnn.py

A PyQt5 window will appear, allowing you to load images and view predictions interactively.

---

⚠️ Notes

• Files like TfractureCNNRev_model.keras and model1_xray_filterCNN.keras are large (>50 MB).
  Consider using Git LFS for better handling.

• If you get errors related to TensorFlow GPU, you can disable GPU by adding:

  os.environ["CUDA_VISIBLE_DEVICES"] = "-1"

---

👩‍💻 Author

Megan Febriana
💼 GitHub Profile
📧 Add your contact info here if desired

---

🏷️ License

This project is licensed under the MIT License — free to use and modify for educational or research purposes.

---

💡 AI-powered bone fracture detection — smarter diagnostics through deep learning!