UNETR with variable number of images as input

[oscarramirezs]

Hello, thanks for this great work.
I have somes CT of 512px*512px for image, but the number of images varies from one patient to another.
I already trained a UNET 3-D model, and using the transforms

DivisiblePadD(keys=["image", "label"], k=16, mode=('constant'), method= ("symmetric")),

I always get a CT input divisible by 16 to train the model.

I am trying now to train a UNETR, but, it looks like this model always need the same number of images for all patients, is this correct? Is there a way to use database with variable number of images to train the model?

thank for your help

Oscar