Workshop 2: Medical image segmentation with CNNs

We will be using a suite of pre-built pytorch segmentation models in the excellent segmentation-models package. This package simplifies the building of a pretrained segmentation network in 2D.

Pytorch can be quite intimidating, but is very powerful when you get to grips with it. In the interests of simplicity, we will use a wrapper around pytorch called pytorch-lightning. Lightning separates out the different bits of ML, allowing you to write a bit less boilerplate code, and letting us very quickly and easily use best-practise methods to train our models.

Pre-requisites:

• Google account
• Basic knowledge of Python

Part 1: Abdominal auto-segmentation

In Part 1, we will show you how to build an auto-segmentation model for abdominal organs-at-risk (OARs) using pytorch and pytorch-lightning. We will be using some open data from the MICCAI 2021 FLARE Challenge. This dataset contains ~500 patients, each of which has four OARs (liver, kidneys, pancreas and spleen) segmented. This notebook serves as an introduction and will walk you through the following steps:

1. Install prerequisites and set up
2. Load data containing CT and segmentations
3. Define some preprocessing and apply it to the CT slices
4. Create a segmentation model, using a library to make a pre-trained model for our segmentation task
5. Optimise a model on the training examples
6. Test the model against the testing data
7. Visualising the results! (You'll need to do this yourself)

Important: As soon as you open a notebook, you should save your own copy to your Google Drive. This will prevent people from editing the same notebook, which will lead to chaos...

To open this notebook in colab, click this link: colab

Part 2: Head and neck auto-segmentation

In Part 2, we will move sites from the abdomen to the Head and Neck (HnN). We now want you to train a second CNN model to segment OARs at this site. The data for this section is originally from The Cancer Imaging Archive (TCIA)

The notebook provided for this part has a skeleton code provided, you'll need to fill in the rest using what you learned from Part 1. Important: Remember to save your own copy as soon as you click the link below!

To open the second notebook in colab, click this link: colab

(Optional) Part 3: Image classification

In Part 3, you will train a CNN to classify head and neck CTs or abdominal CTs. You will then use the trained classifier to select the appropriate segmentation model (from Part 1 and Part 2) to segment random, unseen examples.

Important: Remember to save your own copy as soon as you click the link below! To open in colab, click this link: colab