tueimage/8qa01

Image analysis for cancer risk assessment (8QA01)

This is a repository for the course 8QA01 at Eindhoven University of Technology. This course takes place on two platforms: Github and Canvas:

• In this Github repository you will find all content and technical part of the course
• Canvas is used for announcements, handing in your assignments, and discussion forum
• Meetings will be at the TU/e, COVID-rules permitting... Otherwise we will switch to MS Teams.

Your lecturer is Mark Janse, he can be reached at his TU/e email address (see Canvas).

Project goals

In this project you will learn to measure features in images of skin lesions, and predict the diagnosis (for example, melanoma) from these features in an automatic way. You will likely:

• Study literature to find out what kind of features may be relevant for skin lesion diagnosis
• Implement methods to measure such features
• Predict the lesion diagnosis, based on the features
• Come up with experiments to test if your predictions are good
• Think about other issues concerning automatic diagnosis

Slides & videos

There will be a two COCs, see Canvas for details.

Project code

In this project you will work with Python. This means you need to install software to run and edit Python code. Read more tips about getting started with Python here: https://github.com/tueimage/8qa01/blob/master/tips_code.md

Some basic steps needed to complete the project are found in the Jupyter notebook class2022_group00_script.ipynb, which already contains a general script to:

• go through all the images
• measure simple features in each image
• create a plot of the measurements
• predict the label of the image, based on the measurements

These steps use functions, which can be found in the module class2022_group00_functions.py. This is where you should add more functions for completing your project, and then call these functions from the notebook.

Data

For the project we use images from the following paper:

Codella N, Gutman D, Celebi ME, Helba B, Marchetti MA, Dusza S, Kalloo A, Liopyris K, Mishra N, Kittler H, Halpern A. "Skin Lesion Analysis Toward Melanoma Detection: A Challenge at the 2017 International Symposium on Biomedical Imaging (ISBI), Hosted by the International Skin Imaging Collaboration (ISIC)". arXiv: 1710.05006 [cs.CV] Available: https://arxiv.org/abs/1710.05006

There are more than 2000 images in this dataset. For each image, we have the following data:

• ISIC_[ID].png the image of the lesion
• ISIC_[ID]_segmentation.png the mask of the lesion, showing which pixels belong to the lesion or not
• The label of the image, i.e. whether it belongs to the Melanoma class (0 = no, 1 = yes), and/or the Keratosis class (0 = no, 1 = yes).

The full dataset is available via https://challenge.isic-archive.com/landing/2017, a copy is available here at https://surfdrive.surf.nl/files/index.php/s/yfAaln5OSh1vFCl (large file, around 11GB).

You can put all images into one folder (so you do not have the train/validation/test split), and all masks into one folder.

Assignments

This project consists of two assignments and a presentation:

• Intermediate assignment, which is NOT graded, but for which you will get feedback. For this assignment you need to think of some image features you want to measure, and label a number of images. You hand in a ZIP file on Canvas, with a short report and an Excel sheet with labels. More details here: https://github.com/tueimage/8qa01/blob/master/tips_assignment_intermediate.md

• Final assignment, which is graded. You hand in a ZIP file on Canvas, with your report and code. More details here: https://github.com/tueimage/8qa01/blob/master/tips_assignment_final.md

• The final assignment is also concluded with a "presentation", more details about this will follow in the coming weeks.

Tips from previous years

Here are some tips that students of previous years gave for you!

• “Don’t spend too long on literature in the beginning”
• “Make a plan, especially for time away from campus”
• “Start coding early”
• “Don’t be afraid to try things out”
• “You can do it even if nobody has programmed before!”

8qa01's Issues

Extend functions & notebook

• Simple classification, for example k-NN
• Randomly sample train/test set, or cross-validation setup