NA-MIC Project Weeks

Back to Projects List

MONAI based image augmentator

Key Investigators

• Ciro Benito, Raggio (Karlsruhe Institute of Technology, Germany)
• Paolo, Zaffino (Magna Graecia University of Catanzaro, Italy)
• Maria Francesca, Spadea (Karlsruhe Institute of Technology, Germany)

Presenter location: In-person

Project Description

MONAI and PyTorch based medical image augmentation tool that can be integrated in Slicer. The project aims to be a low-code version of the tool: https://github.com/ciroraggio/AugmentedDataLoader.

It’s designed to operate on a dataset of medical images and apply a series of specific transformations to each image. This process augments the original dataset, providing a greater variety of samples for training deep learning models.

Objective

1. Creating an intuitive graphic interface for the module
2. Parallelising the augmentation process to optimally utilise resources

Approach and Plan

1. Create the extension
2. Implement the augmentation process
3. Try to parallelise the process so that it takes as little time as possible on large data sets

Progress and Next Steps

1. Implemented graphic interface for loading images and masks, choosing transformations and saving images.
2. Implemented and tested MONAI spatial transformations such as Rotation, RandRotation, Flip, Resize.
3. Partially implemented input validation and MONAI intensity transformations, it will be completed in the future.
4. Partially implemented “Preview” feature, which allows the output of transformations to be viewed directly in the scene before saving them in the OS, will be completed in the future.

Illustrations

Files are saved as follows:

Example of image after transformation in the scene:

Background and References

GitHub Project: https://github.com/ciroraggio/SlicerAugmentator