Adding EPI or not ?

[abelsalm]

I finally managed to look into mni-bmpd dataset which contains EPI data and groundtruth for canal segmentation.
Btw images are here, not on data.neuro (I have to see with Mathieu to upload those on git annex properly)
https://drive.google.com/drive/folders/1LEw1f7mCWdNx4aFJZnLm3MmkTWTF9K5H?usp=sharing

The fact is that they are EPI would be intersting for the canal_t2w model, but segmentations are not always very good, and above all the resolution is not very good compared with TSE or STIR. Here it is 1.6x1.6x4.4
So I was wondering about the relevancy of adding EPI data even if it's T2 like.

The fact is also that since the resolution is not very good, I'm afraid it would be hard to distinguish properly where the canal boundary is, and that it could lead to a loss of performance on other type of acquisitions.

I would gladly take advice from anyone!

[valosekj]

Is the image shown above GT, or is that your prediction? If GT, can you please try to apply your model to see how it performs off the box?

[abelsalm]

Is the image shown above GT, or is that your prediction? If GT, can you please try to apply your model to see how it performs off the box?

It's a GT, sure I'll try!