MRI to CT model

[plbenveniste]

The goal of this model is to accurately transform mri scans into ct scans. The greater objective is to be able to use Total Segmentator, which was trained on CT scans, on MRI images in order to perform segmentation.

To do:

• [x] Literature review (maybe look at this GH)
• [x] Data pre-processing (MRI and CT data are contained in same volume so not much to do in terms of pre-processing)
• [ ] Modify script for training and testing of the model
• [ ] Train and test a model
• [ ] Result analysis

[plbenveniste]

Added a new branch plb/mri_to_ct. Created a script to pre-process datasets for training and testing: preprocess_datasets.py. The goal is to use GoldAtlas and SynthRad2023 for training (because the first one has two modalities T1w and T2w and because the second one has a lot of data) and to validate/test on ICSFUS. The model chosen comes from this github : it is the denoising diffusion probabilistic model (DDPM) which proves to create great results when tranforming MRI to CT.

[plbenveniste]

Instead of the previous, I want to train a CycleGAN (Cycle-Consistent Generative Adversarial Network) or a UNIT (Unsupervised Image-to-Image Translation) model to see how these perform, before trying more complex models.