By: 554947584 on June 20, 2021, 3:36 a.m.
Hi all,
I found that no matter how i normalize the target validation images which has 40 slices, my model cannot generalize to those images. In contrast, those images with 80 slices works very well. I tried min-max scale, histogram matching and etc...
Do you have any ideas to process them?
By: reubendo on June 20, 2021, 4:15 p.m.
Hi,
Did you try a zero-mean unit-variance normalisation? It has been succesfully used in ScribbleDA.
Moreover, note that not all images have the same voxel spacing. It may help to resample the images in order to have a similar spacing.
Are there any other ideas from the other participants? I really encourage you to collaborate on this!
Regards, Reuben
