Bone Fracture Detection Web Application

This project involves a web application that uses computer vision to predict bone fracture detection. The application utilizes three different models for classification and detection: YOLOv8, Faster R-CNN with ResNet, and VGG16 with SSD.

Data Source

The dataset used for bone fracture detection is available on Kaggle. You can find the dataset and additional information here.

About Data

The dataset includes X-ray images of different bone fractures. The classes available in the dataset are - Elbow Positive, Fingers Positive, Forearm Fracture, Humerus Fracture, Shoulder Fracture, and Wrist Positive. Each image includes a corresponding file with the annotated bounding box of the fracture.

Running the Repository

To run the web application locally, follow these steps:

• Clone the repository to your local machine.
• Create a virtual environment for the project.
• Install all the required dependencies.
• Download the pre-trained weights for the models here. Make sure to add them in a 'weights' folder in the project directory.
• Run the following command in your terminal to run the app

  streamlit run app.py
  

Web Application Features

The web application is built using Streamlit, providing user-friendly options for bone fracture detection:

Left Panel

• Model Selection: Users can choose from the three available models: YOLOv8, Faster R-CNN with ResNet, and VGG16 with SSD.
• Confidence Threshold: A slider in the left panel allows users to set the prediction confidence threshold.

Overview Tab

The Overview tab provides information about the different models used for bone fracture detection, their significance, and performance metrics.

Testing Tab

In the Testing tab, users can test the models by uploading images and observing the detection results that includes the output class and the respective bounding box and scores.

Working Demo

Streamlit.App.mp4

Get Started

To get started with the bone fracture detection web application, you'll need to clone this repository and follow the instructions above to run the application locally. Experiment with different models and confidence thresholds to explore their performance.

For any issues or feedback, please feel free to reach out