Bone Break Classifier Team-18

An Automated Fracture Classifier

Comment 1

Inspiration

To build a unique project which has not been worked on before, and using existing technologies and data to increase the productivity and agility in various fields of science.

What it does

The Bone Break Classifier can accurately classify the given input image as the correct class of Fracture. This project can predict an image among 12 different classes of fracture. The project classifies the images using Convolutional Neural Network and Deep Learning.

How we built it

These are steps we followed to successfully build the Bone Break Classifier aka the Fracture Classifier.

  1. Collecting Data: We decided to use google images to build our dataset. The dataset consists of 1750 images which has around 150+ images for the 12 different types of fractures.
  2. Data cleaning and preparation: We first removed the unwanted images and then performed resizing and scaling of the images.
  3. Choosing and Training the Model: We decided to use Convolutional Neural Network (CNN) to build our model and trained it with some sample test images. A convolution is the simple application of a filter to an input that results in an activation.

  4. Evaluating the Model: The metrics used for evaluating our model was accuracy and loss.

  5. Prediction by the Model: We used a sample test data to test our model and verify its classification.

The model was successfully able to classify most of the fracture images correctly.

Challenges we ran into

The main challenge we ran into was about selecting and collecting the dataset. We were really focused on making an unique project which was not done anywhere else. Since collecting data is the most important part of a Data Science/ Machine Learning project/ AI project , we decided to create our own dataset. We had to give a lot of thought on selecting which type of dataset would be the most useful and help us correctly classify the fracture type. We finally decided to gather data from google images, due to the bulk availability of images, and good proportion of both accurate and not so accurate images for each class.

Accomplishments that we're proud of

  1. We were able to select a project title which is unique and useful.
  2. Created our own dataset.
  3. Build a model which successfully predicted the type of fracture.
  4. Worked as a team and enjoyed the IBM zDatathon event.

What we learned

How important it is to work in a team. How important is data gathering and the value of a dataset. Why data cleaning and data preparation is very important. How to use Machine Learning, Deep Learning and Data science libraries to the best to create a model. How to use the IBM Linux0NE features like - Docker Containers, tools like Python and Juptyer and libraries like Tensorflow and Keras.

What's next for Bone Break Classifier Team-18

We look forward to make many more improvements on our projects, which includes:

  1. Improving the accuracy of our model.
  2. Build a mobile or a web application, which could also recommend the nearby hospitals for treatment.
  3. Provide detailed description about the different types of injuries, which could be useful for educational purposes.
  4. Along with predicting type of fractures from X-rays, prediction of various different types of diseases and injuries using X-rays, MRI scans, CT scans etc.

Built With

Share this project:

Updates