Suppplying less series that required

[BegoneGIT]

Is it possible to use only one or few MRI sequences (for example only T1 and FLAIR) in preprocessor?

I currently have access to only T1 and T2/FLAIR from a study. I could probably remove references to series I haven't in the moment in code, but the question is if it would break completely or just become less accurate?

Cheers and thank you for that amazing tool.

[neuronflow]

Thanks for your interest in BTK.

We are currently actively developing an alternative preprocessing pipeline that allows the creation of preprocessing pipelines for arbitrary modalities. It is beta. The code still needs to be polished, the API is likely to change, and the documentation needs to be improved. It is functional for Linux systems, though: https://github.com/BrainLesion/preprocessing

Here is an example of how to use it: https://github.com/BrainLesion/preprocessing/blob/main/example_modality_centric_preprocessor.py

If you want to use the vanilla BTK preprocessing, you can also supply FLAIR two times as T2 and T2-FLAIR image and do the same with T1 / T1c.

The BraTS segmentation algorithms need all four sequences, though. We have an upcoming publication featuring segmentation models for other sequence combinations such as yours.

Further, we have published models to synthesize missing sequences.

[BegoneGIT]

Thank you for you answer.

I'd also like to point out that graphic at the README suggests that you can directly use DICOM files for preprocessor. I however had to convert them to nii.gz to get preprocessor running. I could've of course missed something in code.

May I also ask where I can find published models for synthetizing sequences? I read https://ui.adsabs.harvard.edu/abs/2023arXiv230509011B/abstract but from what I understood it announced challenge on predicting contents of missing sequences. I visited MICCAI 2023 webside but had trouble searching if models could be there.

Best regards

[neuronflow]

Nifti conversion is currently only available in our poorly supported GUI.

We recommend to use: https://github.com/rordenlab/dcm2niix

The whole preprocessor will probably be retired soon and we replaced by the BrainLes one above.

Here you can find the model I was referring to: https://github.com/CompImg/Glioma_GAN/

The new BraTS models are on the roadmap but are not released yet :)

[BegoneGIT]

Thank you, I used specifically this converter to get niftis :)

Thank you for a link to Gloma_GAN, I'll be testing how it behaves on my series.

By the way, how do you preview niftis during your workflows? Just using NiBabel, or some GUI tool?

[neuronflow]

For quickly inspecting niftis I can recommend: http://www.itksnap.org/ https://www.slicer.org/