Use custom models and custom data loading functions

[sty21211250]

Hello, I have now written a custom model that needs to enter two data at once, such as
def forward(self, x_current, x_previous):
I need to modify the data_loader function again, but I see that the file of nnUNetDataset says that this is a read-only file, so I would like to ask which part I should modify to meet my requirements

And I certainly thought about just making a simple change to the nnUNetDataLoader3D function or not, transforming the tensor passed in to the model to get a new data. The model I defined takes into account the time factor, so I need to input data of two time points at the same time. If I want to transform the incoming tensor to obtain a new data, I should use the registration field to transform it. However, due to the influence of data enhancement operation, the deformation field may play a counterproductive role, so I am a little confused now.

Mainly because the pre-processing and post-processing of nnunet are too excellent, and the related auxiliary functions are also very perfect, which can greatly facilitate my task. If you can point out the direction, I would appreciate it very much, and then I will attach my model code

[.docx](https://github.com/user-attachments/files/18163876/default.docx)
`
this is my network