In this session, we will apply what we had done in the first session to spot anomalies in Chest X-ray images. Which anomaly are we trying to spot? How do these anomalies look like? Let’s find the answers to these questions in an introduction of Chest X-ray images with Rohit.
You can download the notebook used in this session below. You can also download the CXR data directly on your Nimblebox using the instructions provided at the bottom of this page.
NOTE: The code in all of the notebooks provided for this session is updated occasionally. Therefore it may vary slightly from the one shown in the videos.
You saw that deep learning can be effectively used for spotting anomalies in X-rays to detect various diseases. You also saw that the application of deep learning to medical images has become easier due to the availability of open-source images. In this session, we will use X-Ray images from this CXR dataset (though we will be using only two of these classes – the download instructions for the subsetted dataset are provided at the bottom of this page).
Let’s go ahead and look at the data.
Getting CXR data on Nimblebox
For the purpose of this demonstration, we’ll be working with two classes – ‘effusion’ and ‘nofinding’ We have collected these images in this zip file link.
Notebook with dataset is present in the Nimblebox in the machine ”CNN Assignment -2”. Dataset is present in two folders: ‘effusion’ and ‘nofinding’ on your storage.
Building a Project
As you studying these images, you can build your own classifier for X-ray images and add it to your portfolio. Adding this mini-project to your portfolio will increase your chances of being shortlisted for AI companies. Here’s how you could go about it:
- Create a quick website – Medium or WordPress should do
- On the webpage. write a brief description of X-ray images from what you have learnt here and your own research
- Using the code you have learnt from this session, build a classifier on the images present on this X-ray dataset repository.
- Add your project to your Github profile
Report an error